diff --git a/usr.sbin/bhyve/acpi.c b/usr.sbin/bhyve/acpi.c index fcf6858f40b9..c8fc65cb3b6e 100644 --- a/usr.sbin/bhyve/acpi.c +++ b/usr.sbin/bhyve/acpi.c @@ -1,825 +1,825 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2012 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ /* * bhyve ACPI table generator. * * Create the minimal set of ACPI tables required to boot FreeBSD (and * hopefully other o/s's). * * The tables are placed in the guest's ROM area just below 1MB physical, * above the MPTable. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "bhyverun.h" #include "acpi.h" #include "basl.h" #include "pci_emul.h" #include "vmgenc.h" #define BHYVE_ASL_TEMPLATE "bhyve.XXXXXXX" #define BHYVE_ASL_SUFFIX ".aml" #define BHYVE_ASL_COMPILER "/usr/sbin/iasl" #define BHYVE_ADDRESS_IOAPIC 0xFEC00000 #define BHYVE_ADDRESS_HPET 0xFED00000 #define BHYVE_ADDRESS_LAPIC 0xFEE00000 static int basl_keep_temps; static int basl_verbose_iasl; static int basl_ncpu; static uint32_t hpet_capabilities; /* * Contains the full pathname of the template to be passed * to mkstemp/mktemps(3) */ static char basl_template[MAXPATHLEN]; static char basl_stemplate[MAXPATHLEN]; /* * State for dsdt_line(), dsdt_indent(), and dsdt_unindent(). */ static FILE *dsdt_fp; static int dsdt_indent_level; static int dsdt_error; static struct basl_table *rsdt; static struct basl_table *xsdt; struct basl_fio { int fd; FILE *fp; char f_name[MAXPATHLEN]; }; #define EFPRINTF(...) \ if (fprintf(__VA_ARGS__) < 0) goto err_exit #define EFFLUSH(x) \ if (fflush(x) != 0) goto err_exit /* * A list for additional ACPI devices like a TPM. */ struct acpi_device_list_entry { SLIST_ENTRY(acpi_device_list_entry) chain; const struct acpi_device *dev; }; static SLIST_HEAD(acpi_device_list, acpi_device_list_entry) acpi_devices = SLIST_HEAD_INITIALIZER(acpi_devices); int acpi_tables_add_device(const struct acpi_device *const dev) { struct acpi_device_list_entry *const entry = calloc(1, sizeof(*entry)); if (entry == NULL) { return (ENOMEM); } entry->dev = dev; SLIST_INSERT_HEAD(&acpi_devices, entry, chain); return (0); } /* * Helper routines for writing to the DSDT from other modules. */ void dsdt_line(const char *fmt, ...) { va_list ap; if (dsdt_error != 0) return; if (strcmp(fmt, "") != 0) { if (dsdt_indent_level != 0) EFPRINTF(dsdt_fp, "%*c", dsdt_indent_level * 2, ' '); va_start(ap, fmt); if (vfprintf(dsdt_fp, fmt, ap) < 0) { va_end(ap); goto err_exit; } va_end(ap); } EFPRINTF(dsdt_fp, "\n"); return; err_exit: dsdt_error = errno; } void dsdt_indent(int levels) { dsdt_indent_level += levels; assert(dsdt_indent_level >= 0); } void dsdt_unindent(int levels) { assert(dsdt_indent_level >= levels); dsdt_indent_level -= levels; } void dsdt_fixed_ioport(uint16_t iobase, uint16_t length) { dsdt_line("IO (Decode16,"); dsdt_line(" 0x%04X, // Range Minimum", iobase); dsdt_line(" 0x%04X, // Range Maximum", iobase); dsdt_line(" 0x01, // Alignment"); dsdt_line(" 0x%02X, // Length", length); dsdt_line(" )"); } void dsdt_fixed_irq(uint8_t irq) { dsdt_line("IRQNoFlags ()"); dsdt_line(" {%d}", irq); } void dsdt_fixed_mem32(uint32_t base, uint32_t length) { dsdt_line("Memory32Fixed (ReadWrite,"); dsdt_line(" 0x%08X, // Address Base", base); dsdt_line(" 0x%08X, // Address Length", length); dsdt_line(" )"); } static int basl_fwrite_dsdt(FILE *fp) { dsdt_fp = fp; dsdt_error = 0; dsdt_indent_level = 0; dsdt_line("/*"); dsdt_line(" * bhyve DSDT template"); dsdt_line(" */"); dsdt_line("DefinitionBlock (\"bhyve_dsdt.aml\", \"DSDT\", 2," "\"BHYVE \", \"BVDSDT \", 0x00000001)"); dsdt_line("{"); dsdt_line(" Name (_S5, Package ()"); dsdt_line(" {"); dsdt_line(" 0x05,"); dsdt_line(" Zero,"); dsdt_line(" })"); pci_write_dsdt(); dsdt_line(""); dsdt_line(" Scope (_SB.PC00)"); dsdt_line(" {"); dsdt_line(" Device (HPET)"); dsdt_line(" {"); dsdt_line(" Name (_HID, EISAID(\"PNP0103\"))"); dsdt_line(" Name (_UID, 0)"); dsdt_line(" Name (_CRS, ResourceTemplate ()"); dsdt_line(" {"); dsdt_indent(4); dsdt_fixed_mem32(0xFED00000, 0x400); dsdt_unindent(4); dsdt_line(" })"); dsdt_line(" }"); dsdt_line(" }"); vmgenc_write_dsdt(); const struct acpi_device_list_entry *entry; SLIST_FOREACH(entry, &acpi_devices, chain) { BASL_EXEC(acpi_device_write_dsdt(entry->dev)); } dsdt_line("}"); if (dsdt_error != 0) return (dsdt_error); EFFLUSH(fp); return (0); err_exit: return (errno); } static int basl_open(struct basl_fio *bf, int suffix) { int err; err = 0; if (suffix) { strlcpy(bf->f_name, basl_stemplate, MAXPATHLEN); bf->fd = mkstemps(bf->f_name, strlen(BHYVE_ASL_SUFFIX)); } else { strlcpy(bf->f_name, basl_template, MAXPATHLEN); bf->fd = mkstemp(bf->f_name); } if (bf->fd > 0) { bf->fp = fdopen(bf->fd, "w+"); if (bf->fp == NULL) { unlink(bf->f_name); close(bf->fd); } } else { err = 1; } return (err); } static void basl_close(struct basl_fio *bf) { if (!basl_keep_temps) unlink(bf->f_name); fclose(bf->fp); } static int basl_start(struct basl_fio *in, struct basl_fio *out) { int err; err = basl_open(in, 0); if (!err) { err = basl_open(out, 1); if (err) { basl_close(in); } } return (err); } static void basl_end(struct basl_fio *in, struct basl_fio *out) { basl_close(in); basl_close(out); } static int basl_load(struct vmctx *ctx, int fd) { struct stat sb; void *addr; if (fstat(fd, &sb) < 0) return (errno); addr = calloc(1, sb.st_size); if (addr == NULL) return (EFAULT); if (read(fd, addr, sb.st_size) < 0) return (errno); struct basl_table *table; uint8_t name[ACPI_NAMESEG_SIZE + 1] = { 0 }; memcpy(name, addr, sizeof(name) - 1 /* last char is '\0' */); BASL_EXEC(basl_table_create(&table, ctx, name, BASL_TABLE_ALIGNMENT)); BASL_EXEC(basl_table_append_bytes(table, addr, sb.st_size)); return (0); } static int basl_compile(struct vmctx *ctx, int (*fwrite_section)(FILE *)) { struct basl_fio io[2]; static char iaslbuf[3*MAXPATHLEN + 10]; const char *fmt; int err; err = basl_start(&io[0], &io[1]); if (!err) { err = (*fwrite_section)(io[0].fp); if (!err) { /* * iasl sends the results of the compilation to * stdout. Shut this down by using the shell to * redirect stdout to /dev/null, unless the user * has requested verbose output for debugging * purposes */ fmt = basl_verbose_iasl ? "%s -p %s %s" : "/bin/sh -c \"%s -p %s %s\" 1> /dev/null"; snprintf(iaslbuf, sizeof(iaslbuf), fmt, BHYVE_ASL_COMPILER, io[1].f_name, io[0].f_name); err = system(iaslbuf); if (!err) { /* * Copy the aml output file into guest * memory at the specified location */ err = basl_load(ctx, io[1].fd); } } basl_end(&io[0], &io[1]); } return (err); } static int basl_make_templates(void) { const char *tmpdir; int err; int len; err = 0; /* * */ if ((tmpdir = getenv("BHYVE_TMPDIR")) == NULL || *tmpdir == '\0' || (tmpdir = getenv("TMPDIR")) == NULL || *tmpdir == '\0') { tmpdir = _PATH_TMP; } len = strlen(tmpdir); if ((len + sizeof(BHYVE_ASL_TEMPLATE) + 1) < MAXPATHLEN) { strcpy(basl_template, tmpdir); while (len > 0 && basl_template[len - 1] == '/') len--; basl_template[len] = '/'; strcpy(&basl_template[len + 1], BHYVE_ASL_TEMPLATE); } else err = E2BIG; if (!err) { /* - * len has been intialized (and maybe adjusted) above + * len has been initialized (and maybe adjusted) above */ if ((len + sizeof(BHYVE_ASL_TEMPLATE) + 1 + sizeof(BHYVE_ASL_SUFFIX)) < MAXPATHLEN) { strcpy(basl_stemplate, tmpdir); basl_stemplate[len] = '/'; strcpy(&basl_stemplate[len + 1], BHYVE_ASL_TEMPLATE); len = strlen(basl_stemplate); strcpy(&basl_stemplate[len], BHYVE_ASL_SUFFIX); } else err = E2BIG; } return (err); } static int build_dsdt(struct vmctx *const ctx) { BASL_EXEC(basl_compile(ctx, basl_fwrite_dsdt)); return (0); } static int build_facs(struct vmctx *const ctx) { ACPI_TABLE_FACS facs; struct basl_table *table; BASL_EXEC(basl_table_create(&table, ctx, ACPI_SIG_FACS, BASL_TABLE_ALIGNMENT_FACS)); memset(&facs, 0, sizeof(facs)); memcpy(facs.Signature, ACPI_SIG_FACS, ACPI_NAMESEG_SIZE); facs.Length = sizeof(facs); facs.Version = htole32(2); BASL_EXEC(basl_table_append_bytes(table, &facs, sizeof(facs))); return (0); } static int build_fadt(struct vmctx *const ctx) { ACPI_TABLE_FADT fadt; struct basl_table *table; BASL_EXEC(basl_table_create(&table, ctx, ACPI_SIG_FADT, BASL_TABLE_ALIGNMENT)); memset(&fadt, 0, sizeof(fadt)); BASL_EXEC(basl_table_append_header(table, ACPI_SIG_FADT, 5, 1)); fadt.Facs = htole32(0); /* patched by basl */ fadt.Dsdt = htole32(0); /* patched by basl */ fadt.SciInterrupt = htole16(SCI_INT); fadt.SmiCommand = htole32(SMI_CMD); fadt.AcpiEnable = BHYVE_ACPI_ENABLE; fadt.AcpiDisable = BHYVE_ACPI_DISABLE; fadt.Pm1aEventBlock = htole32(PM1A_EVT_ADDR); fadt.Pm1aControlBlock = htole32(PM1A_CNT_ADDR); fadt.PmTimerBlock = htole32(IO_PMTMR); fadt.Gpe0Block = htole32(IO_GPE0_BLK); fadt.Pm1EventLength = 4; fadt.Pm1ControlLength = 2; fadt.PmTimerLength = 4; fadt.Gpe0BlockLength = IO_GPE0_LEN; fadt.Century = 0x32; fadt.BootFlags = htole16(ACPI_FADT_NO_VGA | ACPI_FADT_NO_ASPM); fadt.Flags = htole32(ACPI_FADT_WBINVD | ACPI_FADT_C1_SUPPORTED | ACPI_FADT_SLEEP_BUTTON | ACPI_FADT_32BIT_TIMER | ACPI_FADT_RESET_REGISTER | ACPI_FADT_HEADLESS | ACPI_FADT_APIC_PHYSICAL); basl_fill_gas(&fadt.ResetRegister, ACPI_ADR_SPACE_SYSTEM_IO, 8, 0, ACPI_GAS_ACCESS_WIDTH_BYTE, 0xCF9); fadt.ResetValue = 6; fadt.MinorRevision = 1; fadt.XFacs = htole64(0); /* patched by basl */ fadt.XDsdt = htole64(0); /* patched by basl */ basl_fill_gas(&fadt.XPm1aEventBlock, ACPI_ADR_SPACE_SYSTEM_IO, 0x20, 0, ACPI_GAS_ACCESS_WIDTH_WORD, PM1A_EVT_ADDR); basl_fill_gas(&fadt.XPm1bEventBlock, ACPI_ADR_SPACE_SYSTEM_IO, 0, 0, ACPI_GAS_ACCESS_WIDTH_UNDEFINED, 0); basl_fill_gas(&fadt.XPm1aControlBlock, ACPI_ADR_SPACE_SYSTEM_IO, 0x10, 0, ACPI_GAS_ACCESS_WIDTH_WORD, PM1A_CNT_ADDR); basl_fill_gas(&fadt.XPm1bControlBlock, ACPI_ADR_SPACE_SYSTEM_IO, 0, 0, ACPI_GAS_ACCESS_WIDTH_UNDEFINED, 0); basl_fill_gas(&fadt.XPm2ControlBlock, ACPI_ADR_SPACE_SYSTEM_IO, 8, 0, ACPI_GAS_ACCESS_WIDTH_UNDEFINED, 0); basl_fill_gas(&fadt.XPmTimerBlock, ACPI_ADR_SPACE_SYSTEM_IO, 0x20, 0, ACPI_GAS_ACCESS_WIDTH_DWORD, IO_PMTMR); basl_fill_gas(&fadt.XGpe0Block, ACPI_ADR_SPACE_SYSTEM_IO, IO_GPE0_LEN * 8, 0, ACPI_GAS_ACCESS_WIDTH_BYTE, IO_GPE0_BLK); basl_fill_gas(&fadt.XGpe1Block, ACPI_ADR_SPACE_SYSTEM_IO, 0, 0, ACPI_GAS_ACCESS_WIDTH_UNDEFINED, 0); basl_fill_gas(&fadt.SleepControl, ACPI_ADR_SPACE_SYSTEM_IO, 8, 0, ACPI_GAS_ACCESS_WIDTH_BYTE, 0); basl_fill_gas(&fadt.SleepStatus, ACPI_ADR_SPACE_SYSTEM_IO, 8, 0, ACPI_GAS_ACCESS_WIDTH_BYTE, 0); BASL_EXEC(basl_table_append_content(table, &fadt, sizeof(fadt))); BASL_EXEC(basl_table_add_pointer(table, ACPI_SIG_FACS, offsetof(ACPI_TABLE_FADT, Facs), sizeof(fadt.Facs))); BASL_EXEC(basl_table_add_pointer(table, ACPI_SIG_DSDT, offsetof(ACPI_TABLE_FADT, Dsdt), sizeof(fadt.Dsdt))); BASL_EXEC(basl_table_add_pointer(table, ACPI_SIG_FACS, offsetof(ACPI_TABLE_FADT, XFacs), sizeof(fadt.XFacs))); BASL_EXEC(basl_table_add_pointer(table, ACPI_SIG_DSDT, offsetof(ACPI_TABLE_FADT, XDsdt), sizeof(fadt.XDsdt))); BASL_EXEC(basl_table_append_pointer(rsdt, ACPI_SIG_FADT, ACPI_RSDT_ENTRY_SIZE)); BASL_EXEC(basl_table_append_pointer(xsdt, ACPI_SIG_FADT, ACPI_XSDT_ENTRY_SIZE)); return (0); } static int build_hpet(struct vmctx *const ctx) { ACPI_TABLE_HPET hpet; struct basl_table *table; BASL_EXEC(basl_table_create(&table, ctx, ACPI_SIG_HPET, BASL_TABLE_ALIGNMENT)); memset(&hpet, 0, sizeof(hpet)); BASL_EXEC(basl_table_append_header(table, ACPI_SIG_HPET, 1, 1)); hpet.Id = htole32(hpet_capabilities); basl_fill_gas(&hpet.Address, ACPI_ADR_SPACE_SYSTEM_MEMORY, 0, 0, ACPI_GAS_ACCESS_WIDTH_LEGACY, BHYVE_ADDRESS_HPET); hpet.Flags = ACPI_HPET_PAGE_PROTECT4; BASL_EXEC(basl_table_append_content(table, &hpet, sizeof(hpet))); BASL_EXEC(basl_table_append_pointer(rsdt, ACPI_SIG_HPET, ACPI_RSDT_ENTRY_SIZE)); BASL_EXEC(basl_table_append_pointer(xsdt, ACPI_SIG_HPET, ACPI_XSDT_ENTRY_SIZE)); return (0); } static int build_madt(struct vmctx *const ctx) { ACPI_TABLE_MADT madt; ACPI_MADT_LOCAL_APIC madt_lapic; ACPI_MADT_IO_APIC madt_ioapic; ACPI_MADT_INTERRUPT_OVERRIDE madt_irq_override; ACPI_MADT_LOCAL_APIC_NMI madt_lapic_nmi; struct basl_table *table; BASL_EXEC(basl_table_create(&table, ctx, ACPI_SIG_MADT, BASL_TABLE_ALIGNMENT)); memset(&madt, 0, sizeof(madt)); BASL_EXEC(basl_table_append_header(table, ACPI_SIG_MADT, 1, 1)); madt.Address = htole32(BHYVE_ADDRESS_LAPIC); madt.Flags = htole32(ACPI_MADT_PCAT_COMPAT); BASL_EXEC(basl_table_append_content(table, &madt, sizeof(madt))); /* Local APIC for each CPU */ for (int i = 0; i < basl_ncpu; ++i) { memset(&madt_lapic, 0, sizeof(madt_lapic)); madt_lapic.Header.Type = ACPI_MADT_TYPE_LOCAL_APIC; madt_lapic.Header.Length = sizeof(madt_lapic); madt_lapic.ProcessorId = i; madt_lapic.Id = i; madt_lapic.LapicFlags = htole32(ACPI_MADT_ENABLED); BASL_EXEC(basl_table_append_bytes(table, &madt_lapic, sizeof(madt_lapic))); } /* I/O APIC */ memset(&madt_ioapic, 0, sizeof(madt_ioapic)); madt_ioapic.Header.Type = ACPI_MADT_TYPE_IO_APIC; madt_ioapic.Header.Length = sizeof(madt_ioapic); madt_ioapic.Address = htole32(BHYVE_ADDRESS_IOAPIC); BASL_EXEC( basl_table_append_bytes(table, &madt_ioapic, sizeof(madt_ioapic))); /* Legacy IRQ0 is connected to pin 2 of the I/O APIC */ memset(&madt_irq_override, 0, sizeof(madt_irq_override)); madt_irq_override.Header.Type = ACPI_MADT_TYPE_INTERRUPT_OVERRIDE; madt_irq_override.Header.Length = sizeof(madt_irq_override); madt_irq_override.GlobalIrq = htole32(2); madt_irq_override.IntiFlags = htole16( ACPI_MADT_POLARITY_ACTIVE_HIGH | ACPI_MADT_TRIGGER_EDGE); BASL_EXEC(basl_table_append_bytes(table, &madt_irq_override, sizeof(madt_irq_override))); memset(&madt_irq_override, 0, sizeof(madt_irq_override)); madt_irq_override.Header.Type = ACPI_MADT_TYPE_INTERRUPT_OVERRIDE; madt_irq_override.Header.Length = sizeof(madt_irq_override); madt_irq_override.SourceIrq = SCI_INT; madt_irq_override.GlobalIrq = htole32(SCI_INT); madt_irq_override.IntiFlags = htole16( ACPI_MADT_POLARITY_ACTIVE_LOW | ACPI_MADT_TRIGGER_LEVEL); BASL_EXEC(basl_table_append_bytes(table, &madt_irq_override, sizeof(madt_irq_override))); /* Local APIC NMI is conntected to LINT 1 on all CPUs */ memset(&madt_lapic_nmi, 0, sizeof(madt_lapic_nmi)); madt_lapic_nmi.Header.Type = ACPI_MADT_TYPE_LOCAL_APIC_NMI; madt_lapic_nmi.Header.Length = sizeof(madt_lapic_nmi); madt_lapic_nmi.ProcessorId = 0xFF; madt_lapic_nmi.IntiFlags = htole16( ACPI_MADT_POLARITY_ACTIVE_HIGH | ACPI_MADT_TRIGGER_EDGE); madt_lapic_nmi.Lint = 1; BASL_EXEC(basl_table_append_bytes(table, &madt_lapic_nmi, sizeof(madt_lapic_nmi))); BASL_EXEC(basl_table_append_pointer(rsdt, ACPI_SIG_MADT, ACPI_RSDT_ENTRY_SIZE)); BASL_EXEC(basl_table_append_pointer(xsdt, ACPI_SIG_MADT, ACPI_XSDT_ENTRY_SIZE)); return (0); } static int build_mcfg(struct vmctx *const ctx) { ACPI_TABLE_MCFG mcfg; ACPI_MCFG_ALLOCATION mcfg_allocation; struct basl_table *table; BASL_EXEC(basl_table_create(&table, ctx, ACPI_SIG_MCFG, BASL_TABLE_ALIGNMENT)); memset(&mcfg, 0, sizeof(mcfg)); BASL_EXEC(basl_table_append_header(table, ACPI_SIG_MCFG, 1, 1)); BASL_EXEC(basl_table_append_content(table, &mcfg, sizeof(mcfg))); memset(&mcfg_allocation, 0, sizeof(mcfg_allocation)); mcfg_allocation.Address = htole64(pci_ecfg_base()); mcfg_allocation.EndBusNumber = 0xFF; BASL_EXEC(basl_table_append_bytes(table, &mcfg_allocation, sizeof(mcfg_allocation))); BASL_EXEC(basl_table_append_pointer(rsdt, ACPI_SIG_MCFG, ACPI_RSDT_ENTRY_SIZE)); BASL_EXEC(basl_table_append_pointer(xsdt, ACPI_SIG_MCFG, ACPI_XSDT_ENTRY_SIZE)); return (0); } static int build_rsdp(struct vmctx *const ctx) { ACPI_TABLE_RSDP rsdp; struct basl_table *table; BASL_EXEC(basl_table_create(&table, ctx, ACPI_RSDP_NAME, BASL_TABLE_ALIGNMENT)); memset(&rsdp, 0, sizeof(rsdp)); memcpy(rsdp.Signature, ACPI_SIG_RSDP, 8); rsdp.Checksum = 0; /* patched by basl */ memcpy(rsdp.OemId, "BHYVE ", ACPI_OEM_ID_SIZE); rsdp.Revision = 2; rsdp.RsdtPhysicalAddress = htole32(0); /* patched by basl */ rsdp.Length = htole32(0); /* patched by basl */ rsdp.XsdtPhysicalAddress = htole64(0); /* patched by basl */ rsdp.ExtendedChecksum = 0; /* patched by basl */ BASL_EXEC(basl_table_append_bytes(table, &rsdp, sizeof(rsdp))); BASL_EXEC(basl_table_add_checksum(table, offsetof(ACPI_TABLE_RSDP, Checksum), 0, 20)); BASL_EXEC(basl_table_add_pointer(table, ACPI_SIG_RSDT, offsetof(ACPI_TABLE_RSDP, RsdtPhysicalAddress), sizeof(rsdp.RsdtPhysicalAddress))); BASL_EXEC(basl_table_add_length(table, offsetof(ACPI_TABLE_RSDP, Length), sizeof(rsdp.Length))); BASL_EXEC(basl_table_add_pointer(table, ACPI_SIG_XSDT, offsetof(ACPI_TABLE_RSDP, XsdtPhysicalAddress), sizeof(rsdp.XsdtPhysicalAddress))); BASL_EXEC(basl_table_add_checksum(table, offsetof(ACPI_TABLE_RSDP, ExtendedChecksum), 0, BASL_TABLE_CHECKSUM_LEN_FULL_TABLE)); return (0); } static int build_rsdt(struct vmctx *const ctx) { BASL_EXEC( basl_table_create(&rsdt, ctx, ACPI_SIG_RSDT, BASL_TABLE_ALIGNMENT)); /* Header */ BASL_EXEC(basl_table_append_header(rsdt, ACPI_SIG_RSDT, 1, 1)); /* Pointers (added by other build_XXX funcs) */ return (0); } static int build_spcr(struct vmctx *const ctx) { ACPI_TABLE_SPCR spcr; struct basl_table *table; BASL_EXEC(basl_table_create(&table, ctx, ACPI_SIG_SPCR, BASL_TABLE_ALIGNMENT)); memset(&spcr, 0, sizeof(spcr)); BASL_EXEC(basl_table_append_header(table, ACPI_SIG_SPCR, 1, 1)); spcr.InterfaceType = ACPI_DBG2_16550_COMPATIBLE; basl_fill_gas(&spcr.SerialPort, ACPI_ADR_SPACE_SYSTEM_IO, 8, 0, ACPI_GAS_ACCESS_WIDTH_LEGACY, 0x3F8); spcr.InterruptType = ACPI_SPCR_INTERRUPT_TYPE_8259; spcr.PcInterrupt = 4; spcr.BaudRate = ACPI_SPCR_BAUD_RATE_115200; spcr.Parity = ACPI_SPCR_PARITY_NO_PARITY; spcr.StopBits = ACPI_SPCR_STOP_BITS_1; spcr.FlowControl = 3; /* RTS/CTS | DCD */ spcr.TerminalType = ACPI_SPCR_TERMINAL_TYPE_VT_UTF8; BASL_EXEC(basl_table_append_content(table, &spcr, sizeof(spcr))); BASL_EXEC(basl_table_append_pointer(rsdt, ACPI_SIG_SPCR, ACPI_RSDT_ENTRY_SIZE)); BASL_EXEC(basl_table_append_pointer(xsdt, ACPI_SIG_SPCR, ACPI_XSDT_ENTRY_SIZE)); return (0); } static int build_xsdt(struct vmctx *const ctx) { BASL_EXEC( basl_table_create(&xsdt, ctx, ACPI_SIG_XSDT, BASL_TABLE_ALIGNMENT)); /* Header */ BASL_EXEC(basl_table_append_header(xsdt, ACPI_SIG_XSDT, 1, 1)); /* Pointers (added by other build_XXX funcs) */ return (0); } int acpi_build(struct vmctx *ctx, int ncpu) { int err; basl_ncpu = ncpu; err = vm_get_hpet_capabilities(ctx, &hpet_capabilities); if (err != 0) return (err); /* * For debug, allow the user to have iasl compiler output sent * to stdout rather than /dev/null */ if (getenv("BHYVE_ACPI_VERBOSE_IASL")) basl_verbose_iasl = 1; /* * Allow the user to keep the generated ASL files for debugging * instead of deleting them following use */ if (getenv("BHYVE_ACPI_KEEPTMPS")) basl_keep_temps = 1; BASL_EXEC(basl_init()); BASL_EXEC(basl_make_templates()); /* * Generate ACPI tables and copy them into guest memory. * * According to UEFI Specification v6.3 chapter 5.1 the FADT should be * the first table pointed to by XSDT. For that reason, build it as the * first table after XSDT. */ BASL_EXEC(build_rsdp(ctx)); BASL_EXEC(build_rsdt(ctx)); BASL_EXEC(build_xsdt(ctx)); BASL_EXEC(build_fadt(ctx)); BASL_EXEC(build_madt(ctx)); BASL_EXEC(build_hpet(ctx)); BASL_EXEC(build_mcfg(ctx)); BASL_EXEC(build_facs(ctx)); BASL_EXEC(build_spcr(ctx)); /* Build ACPI device-specific tables such as a TPM2 table. */ const struct acpi_device_list_entry *entry; SLIST_FOREACH(entry, &acpi_devices, chain) { BASL_EXEC(acpi_device_build_table(entry->dev)); } BASL_EXEC(build_dsdt(ctx)); BASL_EXEC(basl_finish()); return (0); } diff --git a/usr.sbin/bhyve/bhyverun.c b/usr.sbin/bhyve/bhyverun.c index ef5968ba6385..df9e1924a525 100644 --- a/usr.sbin/bhyve/bhyverun.c +++ b/usr.sbin/bhyve/bhyverun.c @@ -1,1655 +1,1655 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #ifndef WITHOUT_CAPSICUM #include #endif #include #ifdef BHYVE_SNAPSHOT #include #include #endif #include #ifdef BHYVE_SNAPSHOT #include #endif #include #include #include #include #ifndef WITHOUT_CAPSICUM #include #endif #include #include #include #include #include #ifdef BHYVE_SNAPSHOT #include #endif #include #include #include #include #include #include #include #include #ifdef BHYVE_SNAPSHOT #include #include #include #endif #include #ifndef WITHOUT_CAPSICUM #include #endif #include #include #include "bhyverun.h" #include "acpi.h" #include "atkbdc.h" #include "bootrom.h" #include "config.h" #include "inout.h" #include "debug.h" #include "e820.h" #include "fwctl.h" #include "gdb.h" #include "ioapic.h" #include "kernemu_dev.h" #include "mem.h" #include "mevent.h" #include "mptbl.h" #include "pci_emul.h" #include "pci_irq.h" #include "pci_lpc.h" #include "qemu_fwcfg.h" #include "smbiostbl.h" #ifdef BHYVE_SNAPSHOT #include "snapshot.h" #endif #include "xmsr.h" #include "spinup_ap.h" #include "rtc.h" #include "vmgenc.h" #define GUEST_NIO_PORT 0x488 /* guest upcalls via i/o port */ #define MB (1024UL * 1024) #define GB (1024UL * MB) static const char * const vmx_exit_reason_desc[] = { [EXIT_REASON_EXCEPTION] = "Exception or non-maskable interrupt (NMI)", [EXIT_REASON_EXT_INTR] = "External interrupt", [EXIT_REASON_TRIPLE_FAULT] = "Triple fault", [EXIT_REASON_INIT] = "INIT signal", [EXIT_REASON_SIPI] = "Start-up IPI (SIPI)", [EXIT_REASON_IO_SMI] = "I/O system-management interrupt (SMI)", [EXIT_REASON_SMI] = "Other SMI", [EXIT_REASON_INTR_WINDOW] = "Interrupt window", [EXIT_REASON_NMI_WINDOW] = "NMI window", [EXIT_REASON_TASK_SWITCH] = "Task switch", [EXIT_REASON_CPUID] = "CPUID", [EXIT_REASON_GETSEC] = "GETSEC", [EXIT_REASON_HLT] = "HLT", [EXIT_REASON_INVD] = "INVD", [EXIT_REASON_INVLPG] = "INVLPG", [EXIT_REASON_RDPMC] = "RDPMC", [EXIT_REASON_RDTSC] = "RDTSC", [EXIT_REASON_RSM] = "RSM", [EXIT_REASON_VMCALL] = "VMCALL", [EXIT_REASON_VMCLEAR] = "VMCLEAR", [EXIT_REASON_VMLAUNCH] = "VMLAUNCH", [EXIT_REASON_VMPTRLD] = "VMPTRLD", [EXIT_REASON_VMPTRST] = "VMPTRST", [EXIT_REASON_VMREAD] = "VMREAD", [EXIT_REASON_VMRESUME] = "VMRESUME", [EXIT_REASON_VMWRITE] = "VMWRITE", [EXIT_REASON_VMXOFF] = "VMXOFF", [EXIT_REASON_VMXON] = "VMXON", [EXIT_REASON_CR_ACCESS] = "Control-register accesses", [EXIT_REASON_DR_ACCESS] = "MOV DR", [EXIT_REASON_INOUT] = "I/O instruction", [EXIT_REASON_RDMSR] = "RDMSR", [EXIT_REASON_WRMSR] = "WRMSR", [EXIT_REASON_INVAL_VMCS] = "VM-entry failure due to invalid guest state", [EXIT_REASON_INVAL_MSR] = "VM-entry failure due to MSR loading", [EXIT_REASON_MWAIT] = "MWAIT", [EXIT_REASON_MTF] = "Monitor trap flag", [EXIT_REASON_MONITOR] = "MONITOR", [EXIT_REASON_PAUSE] = "PAUSE", [EXIT_REASON_MCE_DURING_ENTRY] = "VM-entry failure due to machine-check event", [EXIT_REASON_TPR] = "TPR below threshold", [EXIT_REASON_APIC_ACCESS] = "APIC access", [EXIT_REASON_VIRTUALIZED_EOI] = "Virtualized EOI", [EXIT_REASON_GDTR_IDTR] = "Access to GDTR or IDTR", [EXIT_REASON_LDTR_TR] = "Access to LDTR or TR", [EXIT_REASON_EPT_FAULT] = "EPT violation", [EXIT_REASON_EPT_MISCONFIG] = "EPT misconfiguration", [EXIT_REASON_INVEPT] = "INVEPT", [EXIT_REASON_RDTSCP] = "RDTSCP", [EXIT_REASON_VMX_PREEMPT] = "VMX-preemption timer expired", [EXIT_REASON_INVVPID] = "INVVPID", [EXIT_REASON_WBINVD] = "WBINVD", [EXIT_REASON_XSETBV] = "XSETBV", [EXIT_REASON_APIC_WRITE] = "APIC write", [EXIT_REASON_RDRAND] = "RDRAND", [EXIT_REASON_INVPCID] = "INVPCID", [EXIT_REASON_VMFUNC] = "VMFUNC", [EXIT_REASON_ENCLS] = "ENCLS", [EXIT_REASON_RDSEED] = "RDSEED", [EXIT_REASON_PM_LOG_FULL] = "Page-modification log full", [EXIT_REASON_XSAVES] = "XSAVES", [EXIT_REASON_XRSTORS] = "XRSTORS" }; typedef int (*vmexit_handler_t)(struct vmctx *, struct vcpu *, struct vm_run *); int guest_ncpus; uint16_t cpu_cores, cpu_sockets, cpu_threads; int raw_stdio = 0; static char *progname; static const int BSP = 0; static cpuset_t cpumask; static void vm_loop(struct vmctx *ctx, struct vcpu *vcpu); static struct vcpu_info { struct vmctx *ctx; struct vcpu *vcpu; int vcpuid; } *vcpu_info; static cpuset_t **vcpumap; static void usage(int code) { fprintf(stderr, "Usage: %s [-AaCDeHhPSuWwxY]\n" " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n" " %*s [-G port] [-k config_file] [-l lpc] [-m mem] [-o var=value]\n" " %*s [-p vcpu:hostcpu] [-r file] [-s pci] [-U uuid] vmname\n" " -A: create ACPI tables\n" " -a: local apic is in xAPIC mode (deprecated)\n" " -C: include guest memory in core file\n" " -c: number of CPUs and/or topology specification\n" " -D: destroy on power-off\n" " -e: exit on unhandled I/O access\n" " -G: start a debug server\n" " -H: vmexit from the guest on HLT\n" " -h: help\n" " -k: key=value flat config file\n" " -K: PS2 keyboard layout\n" " -l: LPC device configuration\n" " -m: memory size\n" " -o: set config 'var' to 'value'\n" " -P: vmexit from the guest on pause\n" " -p: pin 'vcpu' to 'hostcpu'\n" #ifdef BHYVE_SNAPSHOT " -r: path to checkpoint file\n" #endif " -S: guest memory cannot be swapped\n" " -s: PCI slot config\n" " -U: UUID\n" " -u: RTC keeps UTC time\n" " -W: force virtio to use single-vector MSI\n" " -w: ignore unimplemented MSRs\n" " -x: local APIC is in x2APIC mode\n" " -Y: disable MPtable generation\n", progname, (int)strlen(progname), "", (int)strlen(progname), "", (int)strlen(progname), ""); exit(code); } /* * XXX This parser is known to have the following issues: * 1. It accepts null key=value tokens ",," as setting "cpus" to an * empty string. * * The acceptance of a null specification ('-c ""') is by design to match the * manual page syntax specification, this results in a topology of 1 vCPU. */ static int topology_parse(const char *opt) { char *cp, *str, *tofree; if (*opt == '\0') { set_config_value("sockets", "1"); set_config_value("cores", "1"); set_config_value("threads", "1"); set_config_value("cpus", "1"); return (0); } tofree = str = strdup(opt); if (str == NULL) errx(4, "Failed to allocate memory"); while ((cp = strsep(&str, ",")) != NULL) { if (strncmp(cp, "cpus=", strlen("cpus=")) == 0) set_config_value("cpus", cp + strlen("cpus=")); else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0) set_config_value("sockets", cp + strlen("sockets=")); else if (strncmp(cp, "cores=", strlen("cores=")) == 0) set_config_value("cores", cp + strlen("cores=")); else if (strncmp(cp, "threads=", strlen("threads=")) == 0) set_config_value("threads", cp + strlen("threads=")); #ifdef notyet /* Do not expose this until vmm.ko implements it */ else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0) set_config_value("maxcpus", cp + strlen("maxcpus=")); #endif else if (strchr(cp, '=') != NULL) goto out; else set_config_value("cpus", cp); } free(tofree); return (0); out: free(tofree); return (-1); } static int parse_int_value(const char *key, const char *value, int minval, int maxval) { char *cp; long lval; errno = 0; lval = strtol(value, &cp, 0); if (errno != 0 || *cp != '\0' || cp == value || lval < minval || lval > maxval) errx(4, "Invalid value for %s: '%s'", key, value); return (lval); } /* * Set the sockets, cores, threads, and guest_cpus variables based on * the configured topology. * * The limits of UINT16_MAX are due to the types passed to * vm_set_topology(). vmm.ko may enforce tighter limits. */ static void calc_topology(void) { const char *value; bool explicit_cpus; uint64_t ncpus; value = get_config_value("cpus"); if (value != NULL) { guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX); explicit_cpus = true; } else { guest_ncpus = 1; explicit_cpus = false; } value = get_config_value("cores"); if (value != NULL) cpu_cores = parse_int_value("cores", value, 1, UINT16_MAX); else cpu_cores = 1; value = get_config_value("threads"); if (value != NULL) cpu_threads = parse_int_value("threads", value, 1, UINT16_MAX); else cpu_threads = 1; value = get_config_value("sockets"); if (value != NULL) cpu_sockets = parse_int_value("sockets", value, 1, UINT16_MAX); else cpu_sockets = guest_ncpus; /* * Compute sockets * cores * threads avoiding overflow. The * range check above insures these are 16 bit values. */ ncpus = (uint64_t)cpu_sockets * cpu_cores * cpu_threads; if (ncpus > UINT16_MAX) errx(4, "Computed number of vCPUs too high: %ju", (uintmax_t)ncpus); if (explicit_cpus) { if (guest_ncpus != (int)ncpus) errx(4, "Topology (%d sockets, %d cores, %d threads) " "does not match %d vCPUs", cpu_sockets, cpu_cores, cpu_threads, guest_ncpus); } else guest_ncpus = ncpus; } static int pincpu_parse(const char *opt) { const char *value; char *newval; char key[16]; int vcpu, pcpu; if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) { fprintf(stderr, "invalid format: %s\n", opt); return (-1); } if (vcpu < 0) { fprintf(stderr, "invalid vcpu '%d'\n", vcpu); return (-1); } if (pcpu < 0 || pcpu >= CPU_SETSIZE) { fprintf(stderr, "hostcpu '%d' outside valid range from " "0 to %d\n", pcpu, CPU_SETSIZE - 1); return (-1); } snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); value = get_config_value(key); if (asprintf(&newval, "%s%s%d", value != NULL ? value : "", value != NULL ? "," : "", pcpu) == -1) { perror("failed to build new cpuset string"); return (-1); } set_config_value(key, newval); free(newval); return (0); } static void parse_cpuset(int vcpu, const char *list, cpuset_t *set) { char *cp, *token; int pcpu, start; CPU_ZERO(set); start = -1; token = __DECONST(char *, list); for (;;) { pcpu = strtoul(token, &cp, 0); if (cp == token) errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); if (pcpu < 0 || pcpu >= CPU_SETSIZE) errx(4, "hostcpu '%d' outside valid range from 0 to %d", pcpu, CPU_SETSIZE - 1); switch (*cp) { case ',': case '\0': if (start >= 0) { if (start > pcpu) errx(4, "Invalid hostcpu range %d-%d", start, pcpu); while (start < pcpu) { CPU_SET(start, set); start++; } start = -1; } CPU_SET(pcpu, set); break; case '-': if (start >= 0) errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); start = pcpu; break; default: errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); } if (*cp == '\0') break; token = cp + 1; } } static void build_vcpumaps(void) { char key[16]; const char *value; int vcpu; vcpumap = calloc(guest_ncpus, sizeof(*vcpumap)); for (vcpu = 0; vcpu < guest_ncpus; vcpu++) { snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); value = get_config_value(key); if (value == NULL) continue; vcpumap[vcpu] = malloc(sizeof(cpuset_t)); if (vcpumap[vcpu] == NULL) err(4, "Failed to allocate cpuset for vcpu %d", vcpu); parse_cpuset(vcpu, value, vcpumap[vcpu]); } } void vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode) { int error, restart_instruction; restart_instruction = 1; error = vm_inject_exception(vcpu, vector, errcode_valid, errcode, restart_instruction); assert(error == 0); } void * paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len) { return (vm_map_gpa(ctx, gaddr, len)); } #ifdef BHYVE_SNAPSHOT uintptr_t paddr_host2guest(struct vmctx *ctx, void *addr) { return (vm_rev_map_gpa(ctx, addr)); } #endif int fbsdrun_virtio_msix(void) { return (get_config_bool_default("virtio_msix", true)); } static void * fbsdrun_start_thread(void *param) { char tname[MAXCOMLEN + 1]; struct vcpu_info *vi = param; int error; snprintf(tname, sizeof(tname), "vcpu %d", vi->vcpuid); pthread_set_name_np(pthread_self(), tname); if (vcpumap[vi->vcpuid] != NULL) { error = pthread_setaffinity_np(pthread_self(), sizeof(cpuset_t), vcpumap[vi->vcpuid]); assert(error == 0); } #ifdef BHYVE_SNAPSHOT checkpoint_cpu_add(vi->vcpuid); #endif gdb_cpu_add(vi->vcpu); vm_loop(vi->ctx, vi->vcpu); /* not reached */ exit(1); return (NULL); } static void fbsdrun_addcpu(struct vcpu_info *vi) { pthread_t thr; int error; error = vm_activate_cpu(vi->vcpu); if (error != 0) err(EX_OSERR, "could not activate CPU %d", vi->vcpuid); CPU_SET_ATOMIC(vi->vcpuid, &cpumask); vm_suspend_cpu(vi->vcpu); error = pthread_create(&thr, NULL, fbsdrun_start_thread, vi); assert(error == 0); } static int fbsdrun_deletecpu(int vcpu) { if (!CPU_ISSET(vcpu, &cpumask)) { fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu); exit(4); } CPU_CLR_ATOMIC(vcpu, &cpumask); return (CPU_EMPTY(&cpumask)); } static int vmexit_handle_notify(struct vmctx *ctx __unused, struct vcpu *vcpu __unused, struct vm_exit *vme __unused, uint32_t eax __unused) { #if BHYVE_DEBUG /* * put guest-driven debug here */ #endif return (VMEXIT_CONTINUE); } static int vmexit_inout(struct vmctx *ctx, struct vcpu *vcpu, struct vm_run *vmrun) { struct vm_exit *vme; int error; int bytes, port, in, out; vme = vmrun->vm_exit; port = vme->u.inout.port; bytes = vme->u.inout.bytes; in = vme->u.inout.in; out = !in; /* Extra-special case of host notifications */ if (out && port == GUEST_NIO_PORT) { error = vmexit_handle_notify(ctx, vcpu, vme, vme->u.inout.eax); return (error); } error = emulate_inout(ctx, vcpu, vme); if (error) { fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n", in ? "in" : "out", bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'), port, vme->rip); return (VMEXIT_ABORT); } else { return (VMEXIT_CONTINUE); } } static int vmexit_rdmsr(struct vmctx *ctx __unused, struct vcpu *vcpu, struct vm_run *vmrun) { struct vm_exit *vme; uint64_t val; uint32_t eax, edx; int error; vme = vmrun->vm_exit; val = 0; error = emulate_rdmsr(vcpu, vme->u.msr.code, &val); if (error != 0) { fprintf(stderr, "rdmsr to register %#x on vcpu %d\n", vme->u.msr.code, vcpu_id(vcpu)); if (get_config_bool("x86.strictmsr")) { vm_inject_gp(vcpu); return (VMEXIT_CONTINUE); } } eax = val; error = vm_set_register(vcpu, VM_REG_GUEST_RAX, eax); assert(error == 0); edx = val >> 32; error = vm_set_register(vcpu, VM_REG_GUEST_RDX, edx); assert(error == 0); return (VMEXIT_CONTINUE); } static int vmexit_wrmsr(struct vmctx *ctx __unused, struct vcpu *vcpu, struct vm_run *vmrun) { struct vm_exit *vme; int error; vme = vmrun->vm_exit; error = emulate_wrmsr(vcpu, vme->u.msr.code, vme->u.msr.wval); if (error != 0) { fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n", vme->u.msr.code, vme->u.msr.wval, vcpu_id(vcpu)); if (get_config_bool("x86.strictmsr")) { vm_inject_gp(vcpu); return (VMEXIT_CONTINUE); } } return (VMEXIT_CONTINUE); } #define DEBUG_EPT_MISCONFIG #ifdef DEBUG_EPT_MISCONFIG #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4]; static int ept_misconfig_ptenum; #endif static const char * vmexit_vmx_desc(uint32_t exit_reason) { if (exit_reason >= nitems(vmx_exit_reason_desc) || vmx_exit_reason_desc[exit_reason] == NULL) return ("Unknown"); return (vmx_exit_reason_desc[exit_reason]); } static int vmexit_vmx(struct vmctx *ctx, struct vcpu *vcpu, struct vm_run *vmrun) { struct vm_exit *vme; vme = vmrun->vm_exit; fprintf(stderr, "vm exit[%d]\n", vcpu_id(vcpu)); fprintf(stderr, "\treason\t\tVMX\n"); fprintf(stderr, "\trip\t\t0x%016lx\n", vme->rip); fprintf(stderr, "\tinst_length\t%d\n", vme->inst_length); fprintf(stderr, "\tstatus\t\t%d\n", vme->u.vmx.status); fprintf(stderr, "\texit_reason\t%u (%s)\n", vme->u.vmx.exit_reason, vmexit_vmx_desc(vme->u.vmx.exit_reason)); fprintf(stderr, "\tqualification\t0x%016lx\n", vme->u.vmx.exit_qualification); fprintf(stderr, "\tinst_type\t\t%d\n", vme->u.vmx.inst_type); fprintf(stderr, "\tinst_error\t\t%d\n", vme->u.vmx.inst_error); #ifdef DEBUG_EPT_MISCONFIG if (vme->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) { vm_get_register(vcpu, VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS), &ept_misconfig_gpa); vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte, &ept_misconfig_ptenum); fprintf(stderr, "\tEPT misconfiguration:\n"); fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa); fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n", ept_misconfig_ptenum, ept_misconfig_pte[0], ept_misconfig_pte[1], ept_misconfig_pte[2], ept_misconfig_pte[3]); } #endif /* DEBUG_EPT_MISCONFIG */ return (VMEXIT_ABORT); } static int vmexit_svm(struct vmctx *ctx __unused, struct vcpu *vcpu, struct vm_run *vmrun) { struct vm_exit *vme; vme = vmrun->vm_exit; fprintf(stderr, "vm exit[%d]\n", vcpu_id(vcpu)); fprintf(stderr, "\treason\t\tSVM\n"); fprintf(stderr, "\trip\t\t0x%016lx\n", vme->rip); fprintf(stderr, "\tinst_length\t%d\n", vme->inst_length); fprintf(stderr, "\texitcode\t%#lx\n", vme->u.svm.exitcode); fprintf(stderr, "\texitinfo1\t%#lx\n", vme->u.svm.exitinfo1); fprintf(stderr, "\texitinfo2\t%#lx\n", vme->u.svm.exitinfo2); return (VMEXIT_ABORT); } static int vmexit_bogus(struct vmctx *ctx __unused, struct vcpu *vcpu __unused, struct vm_run *vmrun) { assert(vmrun->vm_exit->inst_length == 0); return (VMEXIT_CONTINUE); } static int vmexit_reqidle(struct vmctx *ctx __unused, struct vcpu *vcpu __unused, struct vm_run *vmrun) { assert(vmrun->vm_exit->inst_length == 0); return (VMEXIT_CONTINUE); } static int vmexit_hlt(struct vmctx *ctx __unused, struct vcpu *vcpu __unused, struct vm_run *vmrun __unused) { /* * Just continue execution with the next instruction. We use * the HLT VM exit as a way to be friendly with the host * scheduler. */ return (VMEXIT_CONTINUE); } static int vmexit_pause(struct vmctx *ctx __unused, struct vcpu *vcpu __unused, struct vm_run *vmrun __unused) { return (VMEXIT_CONTINUE); } static int vmexit_mtrap(struct vmctx *ctx __unused, struct vcpu *vcpu, struct vm_run *vmrun) { assert(vmrun->vm_exit->inst_length == 0); #ifdef BHYVE_SNAPSHOT checkpoint_cpu_suspend(vcpu_id(vcpu)); #endif gdb_cpu_mtrap(vcpu); #ifdef BHYVE_SNAPSHOT checkpoint_cpu_resume(vcpu_id(vcpu)); #endif return (VMEXIT_CONTINUE); } static int vmexit_inst_emul(struct vmctx *ctx __unused, struct vcpu *vcpu, struct vm_run *vmrun) { struct vm_exit *vme; struct vie *vie; int err, i, cs_d; enum vm_cpu_mode mode; vme = vmrun->vm_exit; vie = &vme->u.inst_emul.vie; if (!vie->decoded) { /* * Attempt to decode in userspace as a fallback. This allows * updating instruction decode in bhyve without rebooting the * kernel (rapid prototyping), albeit with much slower * emulation. */ vie_restart(vie); mode = vme->u.inst_emul.paging.cpu_mode; cs_d = vme->u.inst_emul.cs_d; if (vmm_decode_instruction(mode, cs_d, vie) != 0) goto fail; if (vm_set_register(vcpu, VM_REG_GUEST_RIP, vme->rip + vie->num_processed) != 0) goto fail; } err = emulate_mem(vcpu, vme->u.inst_emul.gpa, vie, &vme->u.inst_emul.paging); if (err) { if (err == ESRCH) { EPRINTLN("Unhandled memory access to 0x%lx\n", vme->u.inst_emul.gpa); } goto fail; } return (VMEXIT_CONTINUE); fail: fprintf(stderr, "Failed to emulate instruction sequence [ "); for (i = 0; i < vie->num_valid; i++) fprintf(stderr, "%02x", vie->inst[i]); FPRINTLN(stderr, " ] at 0x%lx", vme->rip); return (VMEXIT_ABORT); } static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER; static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER; static int vmexit_suspend(struct vmctx *ctx, struct vcpu *vcpu, struct vm_run *vmrun) { struct vm_exit *vme; enum vm_suspend_how how; int vcpuid = vcpu_id(vcpu); vme = vmrun->vm_exit; how = vme->u.suspended.how; fbsdrun_deletecpu(vcpuid); if (vcpuid != BSP) { pthread_mutex_lock(&resetcpu_mtx); pthread_cond_signal(&resetcpu_cond); pthread_mutex_unlock(&resetcpu_mtx); pthread_exit(NULL); } pthread_mutex_lock(&resetcpu_mtx); while (!CPU_EMPTY(&cpumask)) { pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx); } pthread_mutex_unlock(&resetcpu_mtx); switch (how) { case VM_SUSPEND_RESET: exit(0); case VM_SUSPEND_POWEROFF: if (get_config_bool_default("destroy_on_poweroff", false)) vm_destroy(ctx); exit(1); case VM_SUSPEND_HALT: exit(2); case VM_SUSPEND_TRIPLEFAULT: exit(3); default: fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how); exit(100); } return (0); /* NOTREACHED */ } static int vmexit_debug(struct vmctx *ctx __unused, struct vcpu *vcpu, struct vm_run *vmrun __unused) { #ifdef BHYVE_SNAPSHOT checkpoint_cpu_suspend(vcpu_id(vcpu)); #endif gdb_cpu_suspend(vcpu); #ifdef BHYVE_SNAPSHOT checkpoint_cpu_resume(vcpu_id(vcpu)); #endif /* * XXX-MJ sleep for a short period to avoid chewing up the CPU in the * window between activation of the vCPU thread and the STARTUP IPI. */ usleep(1000); return (VMEXIT_CONTINUE); } static int vmexit_breakpoint(struct vmctx *ctx __unused, struct vcpu *vcpu, struct vm_run *vmrun) { gdb_cpu_breakpoint(vcpu, vmrun->vm_exit); return (VMEXIT_CONTINUE); } static int vmexit_ipi(struct vmctx *ctx __unused, struct vcpu *vcpu __unused, struct vm_run *vmrun) { struct vm_exit *vme; cpuset_t *dmask; int error = -1; int i; dmask = vmrun->cpuset; vme = vmrun->vm_exit; switch (vme->u.ipi.mode) { case APIC_DELMODE_INIT: CPU_FOREACH_ISSET(i, dmask) { error = vm_suspend_cpu(vcpu_info[i].vcpu); if (error) { warnx("%s: failed to suspend cpu %d\n", __func__, i); break; } } break; case APIC_DELMODE_STARTUP: CPU_FOREACH_ISSET(i, dmask) { spinup_ap(vcpu_info[i].vcpu, vme->u.ipi.vector << PAGE_SHIFT); } error = 0; break; default: break; } return (error); } static vmexit_handler_t handler[VM_EXITCODE_MAX] = { [VM_EXITCODE_INOUT] = vmexit_inout, [VM_EXITCODE_INOUT_STR] = vmexit_inout, [VM_EXITCODE_VMX] = vmexit_vmx, [VM_EXITCODE_SVM] = vmexit_svm, [VM_EXITCODE_BOGUS] = vmexit_bogus, [VM_EXITCODE_REQIDLE] = vmexit_reqidle, [VM_EXITCODE_RDMSR] = vmexit_rdmsr, [VM_EXITCODE_WRMSR] = vmexit_wrmsr, [VM_EXITCODE_MTRAP] = vmexit_mtrap, [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul, [VM_EXITCODE_SUSPENDED] = vmexit_suspend, [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch, [VM_EXITCODE_DEBUG] = vmexit_debug, [VM_EXITCODE_BPT] = vmexit_breakpoint, [VM_EXITCODE_IPI] = vmexit_ipi, }; static void vm_loop(struct vmctx *ctx, struct vcpu *vcpu) { struct vm_exit vme; struct vm_run vmrun; int error, rc; enum vm_exitcode exitcode; cpuset_t active_cpus, dmask; error = vm_active_cpus(ctx, &active_cpus); assert(CPU_ISSET(vcpu_id(vcpu), &active_cpus)); vmrun.vm_exit = &vme; vmrun.cpuset = &dmask; vmrun.cpusetsize = sizeof(dmask); while (1) { error = vm_run(vcpu, &vmrun); if (error != 0) break; exitcode = vme.exitcode; if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) { fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n", exitcode); exit(4); } rc = (*handler[exitcode])(ctx, vcpu, &vmrun); switch (rc) { case VMEXIT_CONTINUE: break; case VMEXIT_ABORT: abort(); default: exit(4); } } fprintf(stderr, "vm_run error %d, errno %d\n", error, errno); } static int num_vcpus_allowed(struct vmctx *ctx, struct vcpu *vcpu) { uint16_t sockets, cores, threads, maxcpus; int tmp, error; /* * The guest is allowed to spinup more than one processor only if the * UNRESTRICTED_GUEST capability is available. */ error = vm_get_capability(vcpu, VM_CAP_UNRESTRICTED_GUEST, &tmp); if (error != 0) return (1); error = vm_get_topology(ctx, &sockets, &cores, &threads, &maxcpus); if (error == 0) return (maxcpus); else return (1); } static void fbsdrun_set_capabilities(struct vcpu *vcpu, bool bsp) { int err, tmp; if (get_config_bool_default("x86.vmexit_on_hlt", false)) { err = vm_get_capability(vcpu, VM_CAP_HALT_EXIT, &tmp); if (err < 0) { fprintf(stderr, "VM exit on HLT not supported\n"); exit(4); } vm_set_capability(vcpu, VM_CAP_HALT_EXIT, 1); if (bsp) handler[VM_EXITCODE_HLT] = vmexit_hlt; } if (get_config_bool_default("x86.vmexit_on_pause", false)) { /* * pause exit support required for this mode */ err = vm_get_capability(vcpu, VM_CAP_PAUSE_EXIT, &tmp); if (err < 0) { fprintf(stderr, "SMP mux requested, no pause support\n"); exit(4); } vm_set_capability(vcpu, VM_CAP_PAUSE_EXIT, 1); if (bsp) handler[VM_EXITCODE_PAUSE] = vmexit_pause; } if (get_config_bool_default("x86.x2apic", false)) err = vm_set_x2apic_state(vcpu, X2APIC_ENABLED); else err = vm_set_x2apic_state(vcpu, X2APIC_DISABLED); if (err) { fprintf(stderr, "Unable to set x2apic state (%d)\n", err); exit(4); } vm_set_capability(vcpu, VM_CAP_ENABLE_INVPCID, 1); err = vm_set_capability(vcpu, VM_CAP_IPI_EXIT, 1); assert(err == 0); } static struct vmctx * do_open(const char *vmname) { struct vmctx *ctx; int error; bool reinit, romboot; reinit = romboot = false; if (lpc_bootrom()) romboot = true; error = vm_create(vmname); if (error) { if (errno == EEXIST) { if (romboot) { reinit = true; } else { /* * The virtual machine has been setup by the * userspace bootloader. */ } } else { perror("vm_create"); exit(4); } } else { if (!romboot) { /* * If the virtual machine was just created then a * bootrom must be configured to boot it. */ fprintf(stderr, "virtual machine cannot be booted\n"); exit(4); } } ctx = vm_open(vmname); if (ctx == NULL) { perror("vm_open"); exit(4); } #ifndef WITHOUT_CAPSICUM if (vm_limit_rights(ctx) != 0) err(EX_OSERR, "vm_limit_rights"); #endif if (reinit) { error = vm_reinit(ctx); if (error) { perror("vm_reinit"); exit(4); } } error = vm_set_topology(ctx, cpu_sockets, cpu_cores, cpu_threads, 0 /* maxcpus, unimplemented */); if (error) errx(EX_OSERR, "vm_set_topology"); return (ctx); } static void spinup_vcpu(struct vcpu_info *vi, bool bsp) { int error; if (!bsp) { fbsdrun_set_capabilities(vi->vcpu, false); /* * Enable the 'unrestricted guest' mode for APs. * * APs startup in power-on 16-bit mode. */ error = vm_set_capability(vi->vcpu, VM_CAP_UNRESTRICTED_GUEST, 1); assert(error == 0); } fbsdrun_addcpu(vi); } static bool parse_config_option(const char *option) { const char *value; char *path; value = strchr(option, '='); if (value == NULL || value[1] == '\0') return (false); path = strndup(option, value - option); if (path == NULL) err(4, "Failed to allocate memory"); set_config_value(path, value + 1); return (true); } static void parse_simple_config_file(const char *path) { FILE *fp; char *line, *cp; size_t linecap; unsigned int lineno; fp = fopen(path, "r"); if (fp == NULL) err(4, "Failed to open configuration file %s", path); line = NULL; linecap = 0; lineno = 1; for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) { if (*line == '#' || *line == '\n') continue; cp = strchr(line, '\n'); if (cp != NULL) *cp = '\0'; if (!parse_config_option(line)) errx(4, "%s line %u: invalid config option '%s'", path, lineno, line); } free(line); fclose(fp); } static void parse_gdb_options(const char *opt) { const char *sport; char *colon; if (opt[0] == 'w') { set_config_bool("gdb.wait", true); opt++; } colon = strrchr(opt, ':'); if (colon == NULL) { sport = opt; } else { *colon = '\0'; colon++; sport = colon; set_config_value("gdb.address", opt); } set_config_value("gdb.port", sport); } static void set_defaults(void) { set_config_bool("acpi_tables", false); set_config_value("memory.size", "256M"); set_config_bool("x86.strictmsr", true); set_config_value("lpc.fwcfg", "bhyve"); } int main(int argc, char *argv[]) { int c, error; int max_vcpus, memflags; struct vcpu *bsp; struct vmctx *ctx; struct qemu_fwcfg_item *e820_fwcfg_item; uint64_t rip; size_t memsize; const char *optstr, *value, *vmname; #ifdef BHYVE_SNAPSHOT char *restore_file; struct restore_state rstate; restore_file = NULL; #endif init_config(); set_defaults(); progname = basename(argv[0]); #ifdef BHYVE_SNAPSHOT optstr = "aehuwxACDHIPSWYk:f:o:p:G:c:s:m:l:K:U:r:"; #else optstr = "aehuwxACDHIPSWYk:f:o:p:G:c:s:m:l:K:U:"; #endif while ((c = getopt(argc, argv, optstr)) != -1) { switch (c) { case 'a': set_config_bool("x86.x2apic", false); break; case 'A': set_config_bool("acpi_tables", true); break; case 'D': set_config_bool("destroy_on_poweroff", true); break; case 'p': if (pincpu_parse(optarg) != 0) { errx(EX_USAGE, "invalid vcpu pinning " "configuration '%s'", optarg); } break; case 'c': if (topology_parse(optarg) != 0) { errx(EX_USAGE, "invalid cpu topology " "'%s'", optarg); } break; case 'C': set_config_bool("memory.guest_in_core", true); break; case 'f': if (qemu_fwcfg_parse_cmdline_arg(optarg) != 0) { errx(EX_USAGE, "invalid fwcfg item '%s'", optarg); } break; case 'G': parse_gdb_options(optarg); break; case 'k': parse_simple_config_file(optarg); break; case 'K': set_config_value("keyboard.layout", optarg); break; case 'l': if (strncmp(optarg, "help", strlen(optarg)) == 0) { lpc_print_supported_devices(); exit(0); } else if (lpc_device_parse(optarg) != 0) { errx(EX_USAGE, "invalid lpc device " "configuration '%s'", optarg); } break; #ifdef BHYVE_SNAPSHOT case 'r': restore_file = optarg; break; #endif case 's': if (strncmp(optarg, "help", strlen(optarg)) == 0) { pci_print_supported_devices(); exit(0); } else if (pci_parse_slot(optarg) != 0) exit(4); else break; case 'S': set_config_bool("memory.wired", true); break; case 'm': set_config_value("memory.size", optarg); break; case 'o': if (!parse_config_option(optarg)) errx(EX_USAGE, "invalid configuration option '%s'", optarg); break; case 'H': set_config_bool("x86.vmexit_on_hlt", true); break; case 'I': /* * The "-I" option was used to add an ioapic to the * virtual machine. * * An ioapic is now provided unconditionally for each * virtual machine and this option is now deprecated. */ break; case 'P': set_config_bool("x86.vmexit_on_pause", true); break; case 'e': set_config_bool("x86.strictio", true); break; case 'u': set_config_bool("rtc.use_localtime", false); break; case 'U': set_config_value("uuid", optarg); break; case 'w': set_config_bool("x86.strictmsr", false); break; case 'W': set_config_bool("virtio_msix", false); break; case 'x': set_config_bool("x86.x2apic", true); break; case 'Y': set_config_bool("x86.mptable", false); break; case 'h': usage(0); default: usage(1); } } argc -= optind; argv += optind; if (argc > 1) usage(1); #ifdef BHYVE_SNAPSHOT if (restore_file != NULL) { error = load_restore_file(restore_file, &rstate); if (error) { fprintf(stderr, "Failed to read checkpoint info from " "file: '%s'.\n", restore_file); exit(1); } vmname = lookup_vmname(&rstate); if (vmname != NULL) set_config_value("name", vmname); } #endif if (argc == 1) set_config_value("name", argv[0]); vmname = get_config_value("name"); if (vmname == NULL) usage(1); if (get_config_bool_default("config.dump", false)) { dump_config(); exit(1); } calc_topology(); build_vcpumaps(); value = get_config_value("memory.size"); error = vm_parse_memsize(value, &memsize); if (error) errx(EX_USAGE, "invalid memsize '%s'", value); ctx = do_open(vmname); #ifdef BHYVE_SNAPSHOT if (restore_file != NULL) { guest_ncpus = lookup_guest_ncpus(&rstate); memflags = lookup_memflags(&rstate); memsize = lookup_memsize(&rstate); } if (guest_ncpus < 1) { fprintf(stderr, "Invalid guest vCPUs (%d)\n", guest_ncpus); exit(1); } #endif bsp = vm_vcpu_open(ctx, BSP); max_vcpus = num_vcpus_allowed(ctx, bsp); if (guest_ncpus > max_vcpus) { fprintf(stderr, "%d vCPUs requested but only %d available\n", guest_ncpus, max_vcpus); exit(4); } fbsdrun_set_capabilities(bsp, true); /* Allocate per-VCPU resources. */ vcpu_info = calloc(guest_ncpus, sizeof(*vcpu_info)); for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++) { vcpu_info[vcpuid].ctx = ctx; vcpu_info[vcpuid].vcpuid = vcpuid; if (vcpuid == BSP) vcpu_info[vcpuid].vcpu = bsp; else vcpu_info[vcpuid].vcpu = vm_vcpu_open(ctx, vcpuid); } memflags = 0; if (get_config_bool_default("memory.wired", false)) memflags |= VM_MEM_F_WIRED; if (get_config_bool_default("memory.guest_in_core", false)) memflags |= VM_MEM_F_INCORE; vm_set_memflags(ctx, memflags); error = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); if (error) { fprintf(stderr, "Unable to setup memory (%d)\n", errno); exit(4); } error = init_msr(); if (error) { fprintf(stderr, "init_msr error %d", error); exit(4); } init_mem(guest_ncpus); init_inout(); kernemu_dev_init(); init_bootrom(ctx); atkbdc_init(ctx); pci_irq_init(ctx); ioapic_init(ctx); rtc_init(ctx); sci_init(ctx); if (qemu_fwcfg_init(ctx) != 0) { fprintf(stderr, "qemu fwcfg initialization error"); exit(4); } if (qemu_fwcfg_add_file("opt/bhyve/hw.ncpu", sizeof(guest_ncpus), &guest_ncpus) != 0) { fprintf(stderr, "Could not add qemu fwcfg opt/bhyve/hw.ncpu"); exit(4); } if (e820_init(ctx) != 0) { fprintf(stderr, "Unable to setup E820"); exit(4); } /* - * Exit if a device emulation finds an error in its initilization + * Exit if a device emulation finds an error in its initialization */ if (init_pci(ctx) != 0) { perror("device emulation initialization error"); exit(4); } /* * Initialize after PCI, to allow a bootrom file to reserve the high * region. */ if (get_config_bool("acpi_tables")) vmgenc_init(ctx); init_gdb(ctx); if (lpc_bootrom()) { if (vm_set_capability(bsp, VM_CAP_UNRESTRICTED_GUEST, 1)) { fprintf(stderr, "ROM boot failed: unrestricted guest " "capability not available\n"); exit(4); } error = vcpu_reset(bsp); assert(error == 0); } /* * Add all vCPUs. */ for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++) spinup_vcpu(&vcpu_info[vcpuid], vcpuid == BSP); #ifdef BHYVE_SNAPSHOT if (restore_file != NULL) { fprintf(stdout, "Pausing pci devs...\r\n"); if (vm_pause_user_devs() != 0) { fprintf(stderr, "Failed to pause PCI device state.\n"); exit(1); } fprintf(stdout, "Restoring vm mem...\r\n"); if (restore_vm_mem(ctx, &rstate) != 0) { fprintf(stderr, "Failed to restore VM memory.\n"); exit(1); } fprintf(stdout, "Restoring pci devs...\r\n"); if (vm_restore_user_devs(&rstate) != 0) { fprintf(stderr, "Failed to restore PCI device state.\n"); exit(1); } fprintf(stdout, "Restoring kernel structs...\r\n"); if (vm_restore_kern_structs(ctx, &rstate) != 0) { fprintf(stderr, "Failed to restore kernel structs.\n"); exit(1); } fprintf(stdout, "Resuming pci devs...\r\n"); if (vm_resume_user_devs() != 0) { fprintf(stderr, "Failed to resume PCI device state.\n"); exit(1); } } #endif error = vm_get_register(bsp, VM_REG_GUEST_RIP, &rip); assert(error == 0); /* * build the guest tables, MP etc. */ if (get_config_bool_default("x86.mptable", true)) { error = mptable_build(ctx, guest_ncpus); if (error) { perror("error to build the guest tables"); exit(4); } } error = smbios_build(ctx); if (error != 0) exit(4); if (get_config_bool("acpi_tables")) { error = acpi_build(ctx, guest_ncpus); assert(error == 0); } e820_fwcfg_item = e820_get_fwcfg_item(); if (e820_fwcfg_item == NULL) { fprintf(stderr, "invalid e820 table"); exit(4); } if (qemu_fwcfg_add_file("etc/e820", e820_fwcfg_item->size, e820_fwcfg_item->data) != 0) { fprintf(stderr, "could not add qemu fwcfg etc/e820"); exit(4); } free(e820_fwcfg_item); if (lpc_bootrom() && strcmp(lpc_fwcfg(), "bhyve") == 0) { fwctl_init(); } /* * Change the proc title to include the VM name. */ setproctitle("%s", vmname); #ifdef BHYVE_SNAPSHOT /* initialize mutex/cond variables */ init_snapshot(); /* * checkpointing thread for communication with bhyvectl */ if (init_checkpoint_thread(ctx) != 0) errx(EX_OSERR, "Failed to start checkpoint thread"); #endif #ifndef WITHOUT_CAPSICUM caph_cache_catpages(); if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1) errx(EX_OSERR, "Unable to apply rights for sandbox"); if (caph_enter() == -1) errx(EX_OSERR, "cap_enter() failed"); #endif #ifdef BHYVE_SNAPSHOT if (restore_file != NULL) { destroy_restore_state(&rstate); if (vm_restore_time(ctx) < 0) err(EX_OSERR, "Unable to restore time"); for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++) vm_resume_cpu(vcpu_info[vcpuid].vcpu); } else #endif vm_resume_cpu(bsp); /* * Head off to the main event dispatch loop */ mevent_dispatch(); exit(4); } diff --git a/usr.sbin/bhyve/config.c b/usr.sbin/bhyve/config.c index ba51f81dd011..c844d30f2f74 100644 --- a/usr.sbin/bhyve/config.c +++ b/usr.sbin/bhyve/config.c @@ -1,466 +1,466 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2021 John H. Baldwin * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include "config.h" static nvlist_t *config_root; void init_config(void) { config_root = nvlist_create(0); if (config_root == NULL) err(4, "Failed to create configuration root nvlist"); } static nvlist_t * _lookup_config_node(nvlist_t *parent, const char *path, bool create) { char *copy, *name, *tofree; nvlist_t *nvl, *new_nvl; copy = strdup(path); if (copy == NULL) errx(4, "Failed to allocate memory"); tofree = copy; nvl = parent; while ((name = strsep(©, ".")) != NULL) { if (*name == '\0') { warnx("Invalid configuration node: %s", path); nvl = NULL; break; } if (nvlist_exists_nvlist(nvl, name)) /* * XXX-MJ it is incorrect to cast away the const * qualifier like this since the contract with nvlist * says that values are immutable, and some consumers * will indeed add nodes to the returned nvlist. In * practice, however, it appears to be harmless with the * current nvlist implementation, so we just live with * it until the implementation is reworked. */ nvl = __DECONST(nvlist_t *, nvlist_get_nvlist(nvl, name)); else if (nvlist_exists(nvl, name)) { for (copy = tofree; copy < name; copy++) if (*copy == '\0') *copy = '.'; warnx( "Configuration node %s is a child of existing variable %s", path, tofree); nvl = NULL; break; } else if (create) { /* * XXX-MJ as with the case above, "new_nvl" shouldn't be * mutated after its ownership is given to "nvl". */ new_nvl = nvlist_create(0); if (new_nvl == NULL) errx(4, "Failed to allocate memory"); nvlist_move_nvlist(nvl, name, new_nvl); nvl = new_nvl; } else { nvl = NULL; break; } } free(tofree); return (nvl); } nvlist_t * create_config_node(const char *path) { return (_lookup_config_node(config_root, path, true)); } nvlist_t * find_config_node(const char *path) { return (_lookup_config_node(config_root, path, false)); } nvlist_t * create_relative_config_node(nvlist_t *parent, const char *path) { return (_lookup_config_node(parent, path, true)); } nvlist_t * find_relative_config_node(nvlist_t *parent, const char *path) { return (_lookup_config_node(parent, path, false)); } void set_config_value_node(nvlist_t *parent, const char *name, const char *value) { if (strchr(name, '.') != NULL) errx(4, "Invalid config node name %s", name); if (parent == NULL) parent = config_root; if (nvlist_exists_string(parent, name)) nvlist_free_string(parent, name); else if (nvlist_exists(parent, name)) errx(4, - "Attemping to add value %s to existing node %s of list %p", + "Attempting to add value %s to existing node %s of list %p", value, name, parent); nvlist_add_string(parent, name, value); } void set_config_value_node_if_unset(nvlist_t *const parent, const char *const name, const char *const value) { if (get_config_value_node(parent, name) != NULL) { return; } set_config_value_node(parent, name, value); } void set_config_value(const char *path, const char *value) { const char *name; char *node_name; nvlist_t *nvl; /* Look for last separator. */ name = strrchr(path, '.'); if (name == NULL) { nvl = config_root; name = path; } else { node_name = strndup(path, name - path); if (node_name == NULL) errx(4, "Failed to allocate memory"); nvl = create_config_node(node_name); if (nvl == NULL) errx(4, "Failed to create configuration node %s", node_name); free(node_name); /* Skip over '.'. */ name++; } if (nvlist_exists_nvlist(nvl, name)) errx(4, "Attempting to add value %s to existing node %s", value, path); set_config_value_node(nvl, name, value); } void set_config_value_if_unset(const char *const path, const char *const value) { if (get_config_value(path) != NULL) { return; } set_config_value(path, value); } static const char * get_raw_config_value(const char *path) { const char *name; char *node_name; nvlist_t *nvl; /* Look for last separator. */ name = strrchr(path, '.'); if (name == NULL) { nvl = config_root; name = path; } else { node_name = strndup(path, name - path); if (node_name == NULL) errx(4, "Failed to allocate memory"); nvl = find_config_node(node_name); free(node_name); if (nvl == NULL) return (NULL); /* Skip over '.'. */ name++; } if (nvlist_exists_string(nvl, name)) return (nvlist_get_string(nvl, name)); if (nvlist_exists_nvlist(nvl, name)) warnx("Attempting to fetch value of node %s", path); return (NULL); } static char * _expand_config_value(const char *value, int depth) { FILE *valfp; const char *cp, *vp; char *nestedval, *path, *valbuf; size_t valsize; valfp = open_memstream(&valbuf, &valsize); if (valfp == NULL) errx(4, "Failed to allocate memory"); vp = value; while (*vp != '\0') { switch (*vp) { case '%': if (depth > 15) { warnx( "Too many recursive references in configuration value"); fputc('%', valfp); vp++; break; } if (vp[1] != '(' || vp[2] == '\0') cp = NULL; else cp = strchr(vp + 2, ')'); if (cp == NULL) { warnx( "Invalid reference in configuration value \"%s\"", value); fputc('%', valfp); vp++; break; } vp += 2; if (cp == vp) { warnx( "Empty reference in configuration value \"%s\"", value); vp++; break; } /* Allocate a C string holding the path. */ path = strndup(vp, cp - vp); if (path == NULL) errx(4, "Failed to allocate memory"); /* Advance 'vp' past the reference. */ vp = cp + 1; /* Fetch the referenced value. */ cp = get_raw_config_value(path); if (cp == NULL) warnx( "Failed to fetch referenced configuration variable %s", path); else { nestedval = _expand_config_value(cp, depth + 1); fputs(nestedval, valfp); free(nestedval); } free(path); break; case '\\': vp++; if (*vp == '\0') { warnx( "Trailing \\ in configuration value \"%s\"", value); break; } /* FALLTHROUGH */ default: fputc(*vp, valfp); vp++; break; } } fclose(valfp); return (valbuf); } static const char * expand_config_value(const char *value) { static char *valbuf; if (strchr(value, '%') == NULL) return (value); free(valbuf); valbuf = _expand_config_value(value, 0); return (valbuf); } const char * get_config_value(const char *path) { const char *value; value = get_raw_config_value(path); if (value == NULL) return (NULL); return (expand_config_value(value)); } const char * get_config_value_node(const nvlist_t *parent, const char *name) { if (strchr(name, '.') != NULL) errx(4, "Invalid config node name %s", name); if (parent == NULL) parent = config_root; if (nvlist_exists_nvlist(parent, name)) warnx("Attempt to fetch value of node %s of list %p", name, parent); if (!nvlist_exists_string(parent, name)) return (NULL); return (expand_config_value(nvlist_get_string(parent, name))); } static bool _bool_value(const char *name, const char *value) { if (strcasecmp(value, "true") == 0 || strcasecmp(value, "on") == 0 || strcasecmp(value, "yes") == 0 || strcmp(value, "1") == 0) return (true); if (strcasecmp(value, "false") == 0 || strcasecmp(value, "off") == 0 || strcasecmp(value, "no") == 0 || strcmp(value, "0") == 0) return (false); err(4, "Invalid value %s for boolean variable %s", value, name); } bool get_config_bool(const char *path) { const char *value; value = get_config_value(path); if (value == NULL) err(4, "Failed to fetch boolean variable %s", path); return (_bool_value(path, value)); } bool get_config_bool_default(const char *path, bool def) { const char *value; value = get_config_value(path); if (value == NULL) return (def); return (_bool_value(path, value)); } bool get_config_bool_node(const nvlist_t *parent, const char *name) { const char *value; value = get_config_value_node(parent, name); if (value == NULL) err(4, "Failed to fetch boolean variable %s", name); return (_bool_value(name, value)); } bool get_config_bool_node_default(const nvlist_t *parent, const char *name, bool def) { const char *value; value = get_config_value_node(parent, name); if (value == NULL) return (def); return (_bool_value(name, value)); } void set_config_bool(const char *path, bool value) { set_config_value(path, value ? "true" : "false"); } void set_config_bool_node(nvlist_t *parent, const char *name, bool value) { set_config_value_node(parent, name, value ? "true" : "false"); } static void dump_tree(const char *prefix, const nvlist_t *nvl) { const char *name; void *cookie; int type; cookie = NULL; while ((name = nvlist_next(nvl, &type, &cookie)) != NULL) { if (type == NV_TYPE_NVLIST) { char *new_prefix; asprintf(&new_prefix, "%s%s.", prefix, name); dump_tree(new_prefix, nvlist_get_nvlist(nvl, name)); free(new_prefix); } else { assert(type == NV_TYPE_STRING); printf("%s%s=%s\n", prefix, name, nvlist_get_string(nvl, name)); } } } void dump_config(void) { dump_tree("", config_root); } diff --git a/usr.sbin/bhyve/net_backends.h b/usr.sbin/bhyve/net_backends.h index 844024e7bc9d..cc8c7b5cff32 100644 --- a/usr.sbin/bhyve/net_backends.h +++ b/usr.sbin/bhyve/net_backends.h @@ -1,96 +1,96 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2019 Vincenzo Maffione * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ */ #ifndef __NET_BACKENDS_H__ #define __NET_BACKENDS_H__ #include /* Opaque type representing a network backend. */ typedef struct net_backend net_backend_t; /* Interface between network frontends and the network backends. */ typedef void (*net_be_rxeof_t)(int, enum ev_type, void *param); int netbe_init(net_backend_t **be, nvlist_t *nvl, net_be_rxeof_t cb, void *param); int netbe_legacy_config(nvlist_t *nvl, const char *opts); void netbe_cleanup(net_backend_t *be); uint64_t netbe_get_cap(net_backend_t *be); int netbe_set_cap(net_backend_t *be, uint64_t cap, unsigned vnet_hdr_len); size_t netbe_get_vnet_hdr_len(net_backend_t *be); ssize_t netbe_send(net_backend_t *be, const struct iovec *iov, int iovcnt); ssize_t netbe_peek_recvlen(net_backend_t *be); ssize_t netbe_recv(net_backend_t *be, const struct iovec *iov, int iovcnt); ssize_t netbe_rx_discard(net_backend_t *be); void netbe_rx_disable(net_backend_t *be); void netbe_rx_enable(net_backend_t *be); /* * Network device capabilities taken from the VirtIO standard. - * Despite the name, these capabilities can be used by different frontents + * Despite the name, these capabilities can be used by different frontends * (virtio-net, ptnet) and supported by different backends (netmap, tap, ...). */ #define VIRTIO_NET_F_CSUM (1 << 0) /* host handles partial cksum */ #define VIRTIO_NET_F_GUEST_CSUM (1 << 1) /* guest handles partial cksum */ #define VIRTIO_NET_F_MTU (1 << 3) /* initial MTU advice */ #define VIRTIO_NET_F_MAC (1 << 5) /* host supplies MAC */ #define VIRTIO_NET_F_GSO_DEPREC (1 << 6) /* deprecated: host handles GSO */ #define VIRTIO_NET_F_GUEST_TSO4 (1 << 7) /* guest can rcv TSOv4 */ #define VIRTIO_NET_F_GUEST_TSO6 (1 << 8) /* guest can rcv TSOv6 */ #define VIRTIO_NET_F_GUEST_ECN (1 << 9) /* guest can rcv TSO with ECN */ #define VIRTIO_NET_F_GUEST_UFO (1 << 10) /* guest can rcv UFO */ #define VIRTIO_NET_F_HOST_TSO4 (1 << 11) /* host can rcv TSOv4 */ #define VIRTIO_NET_F_HOST_TSO6 (1 << 12) /* host can rcv TSOv6 */ #define VIRTIO_NET_F_HOST_ECN (1 << 13) /* host can rcv TSO with ECN */ #define VIRTIO_NET_F_HOST_UFO (1 << 14) /* host can rcv UFO */ #define VIRTIO_NET_F_MRG_RXBUF (1 << 15) /* host can merge RX buffers */ #define VIRTIO_NET_F_STATUS (1 << 16) /* config status field available */ #define VIRTIO_NET_F_CTRL_VQ (1 << 17) /* control channel available */ #define VIRTIO_NET_F_CTRL_RX (1 << 18) /* control channel RX mode support */ #define VIRTIO_NET_F_CTRL_VLAN (1 << 19) /* control channel VLAN filtering */ #define VIRTIO_NET_F_GUEST_ANNOUNCE \ (1 << 21) /* guest can send gratuitous pkts */ #define VIRTIO_NET_F_MQ (1 << 22) /* host supports multiple VQ pairs */ /* * Fixed network header size */ struct virtio_net_rxhdr { uint8_t vrh_flags; uint8_t vrh_gso_type; uint16_t vrh_hdr_len; uint16_t vrh_gso_size; uint16_t vrh_csum_start; uint16_t vrh_csum_offset; uint16_t vrh_bufs; } __packed; #endif /* __NET_BACKENDS_H__ */ diff --git a/usr.sbin/bhyve/pci_nvme.c b/usr.sbin/bhyve/pci_nvme.c index f3f10cba7502..de5865220155 100644 --- a/usr.sbin/bhyve/pci_nvme.c +++ b/usr.sbin/bhyve/pci_nvme.c @@ -1,3342 +1,3342 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2017 Shunsuke Mie * Copyright (c) 2018 Leon Dang * Copyright (c) 2020 Chuck Tuffli * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * bhyve PCIe-NVMe device emulation. * * options: * -s ,nvme,devpath,maxq=#,qsz=#,ioslots=#,sectsz=#,ser=A-Z,eui64=#,dsm= * * accepted devpath: * /dev/blockdev * /path/to/image * ram=size_in_MiB * * maxq = max number of queues * qsz = max elements in each queue * ioslots = max number of concurrent io requests * sectsz = sector size (defaults to blockif sector size) * ser = serial number (20-chars max) * eui64 = IEEE Extended Unique Identifier (8 byte value) * dsm = DataSet Management support. Option is one of auto, enable,disable * */ /* TODO: - create async event for smart and log - intr coalesce */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "bhyverun.h" #include "block_if.h" #include "config.h" #include "debug.h" #include "pci_emul.h" static int nvme_debug = 0; #define DPRINTF(fmt, args...) if (nvme_debug) PRINTLN(fmt, ##args) #define WPRINTF(fmt, args...) PRINTLN(fmt, ##args) /* defaults; can be overridden */ #define NVME_MSIX_BAR 4 #define NVME_IOSLOTS 8 /* The NVMe spec defines bits 13:4 in BAR0 as reserved */ #define NVME_MMIO_SPACE_MIN (1 << 14) #define NVME_QUEUES 16 #define NVME_MAX_QENTRIES 2048 /* Memory Page size Minimum reported in CAP register */ #define NVME_MPSMIN 0 /* MPSMIN converted to bytes */ #define NVME_MPSMIN_BYTES (1 << (12 + NVME_MPSMIN)) #define NVME_PRP2_ITEMS (PAGE_SIZE/sizeof(uint64_t)) #define NVME_MDTS 9 /* Note the + 1 allows for the initial descriptor to not be page aligned */ #define NVME_MAX_IOVEC ((1 << NVME_MDTS) + 1) #define NVME_MAX_DATA_SIZE ((1 << NVME_MDTS) * NVME_MPSMIN_BYTES) /* This is a synthetic status code to indicate there is no status */ #define NVME_NO_STATUS 0xffff #define NVME_COMPLETION_VALID(c) ((c).status != NVME_NO_STATUS) /* Reported temperature in Kelvin (i.e. room temperature) */ #define NVME_TEMPERATURE 296 /* helpers */ /* Convert a zero-based value into a one-based value */ #define ONE_BASED(zero) ((zero) + 1) /* Convert a one-based value into a zero-based value */ #define ZERO_BASED(one) ((one) - 1) /* Encode number of SQ's and CQ's for Set/Get Features */ #define NVME_FEATURE_NUM_QUEUES(sc) \ (ZERO_BASED((sc)->num_squeues) & 0xffff) | \ (ZERO_BASED((sc)->num_cqueues) & 0xffff) << 16 #define NVME_DOORBELL_OFFSET offsetof(struct nvme_registers, doorbell) enum nvme_controller_register_offsets { NVME_CR_CAP_LOW = 0x00, NVME_CR_CAP_HI = 0x04, NVME_CR_VS = 0x08, NVME_CR_INTMS = 0x0c, NVME_CR_INTMC = 0x10, NVME_CR_CC = 0x14, NVME_CR_CSTS = 0x1c, NVME_CR_NSSR = 0x20, NVME_CR_AQA = 0x24, NVME_CR_ASQ_LOW = 0x28, NVME_CR_ASQ_HI = 0x2c, NVME_CR_ACQ_LOW = 0x30, NVME_CR_ACQ_HI = 0x34, }; enum nvme_cmd_cdw11 { NVME_CMD_CDW11_PC = 0x0001, NVME_CMD_CDW11_IEN = 0x0002, NVME_CMD_CDW11_IV = 0xFFFF0000, }; enum nvme_copy_dir { NVME_COPY_TO_PRP, NVME_COPY_FROM_PRP, }; #define NVME_CQ_INTEN 0x01 #define NVME_CQ_INTCOAL 0x02 struct nvme_completion_queue { struct nvme_completion *qbase; pthread_mutex_t mtx; uint32_t size; uint16_t tail; /* nvme progress */ uint16_t head; /* guest progress */ uint16_t intr_vec; uint32_t intr_en; }; struct nvme_submission_queue { struct nvme_command *qbase; pthread_mutex_t mtx; uint32_t size; uint16_t head; /* nvme progress */ uint16_t tail; /* guest progress */ uint16_t cqid; /* completion queue id */ int qpriority; }; enum nvme_storage_type { NVME_STOR_BLOCKIF = 0, NVME_STOR_RAM = 1, }; struct pci_nvme_blockstore { enum nvme_storage_type type; void *ctx; uint64_t size; uint32_t sectsz; uint32_t sectsz_bits; uint64_t eui64; uint32_t deallocate:1; }; /* * Calculate the number of additional page descriptors for guest IO requests * based on the advertised Max Data Transfer (MDTS) and given the number of * default iovec's in a struct blockif_req. */ #define MDTS_PAD_SIZE \ ( NVME_MAX_IOVEC > BLOCKIF_IOV_MAX ? \ NVME_MAX_IOVEC - BLOCKIF_IOV_MAX : \ 0 ) struct pci_nvme_ioreq { struct pci_nvme_softc *sc; STAILQ_ENTRY(pci_nvme_ioreq) link; struct nvme_submission_queue *nvme_sq; uint16_t sqid; /* command information */ uint16_t opc; uint16_t cid; uint32_t nsid; uint64_t prev_gpaddr; size_t prev_size; size_t bytes; struct blockif_req io_req; struct iovec iovpadding[MDTS_PAD_SIZE]; }; enum nvme_dsm_type { /* Dataset Management bit in ONCS reflects backing storage capability */ NVME_DATASET_MANAGEMENT_AUTO, /* Unconditionally set Dataset Management bit in ONCS */ NVME_DATASET_MANAGEMENT_ENABLE, /* Unconditionally clear Dataset Management bit in ONCS */ NVME_DATASET_MANAGEMENT_DISABLE, }; struct pci_nvme_softc; struct nvme_feature_obj; typedef void (*nvme_feature_cb)(struct pci_nvme_softc *, struct nvme_feature_obj *, struct nvme_command *, struct nvme_completion *); struct nvme_feature_obj { uint32_t cdw11; nvme_feature_cb set; nvme_feature_cb get; bool namespace_specific; }; #define NVME_FID_MAX (NVME_FEAT_ENDURANCE_GROUP_EVENT_CONFIGURATION + 1) typedef enum { PCI_NVME_AE_TYPE_ERROR = 0, PCI_NVME_AE_TYPE_SMART, PCI_NVME_AE_TYPE_NOTICE, PCI_NVME_AE_TYPE_IO_CMD = 6, PCI_NVME_AE_TYPE_VENDOR = 7, PCI_NVME_AE_TYPE_MAX /* Must be last */ } pci_nvme_async_type; /* Asynchronous Event Requests */ struct pci_nvme_aer { STAILQ_ENTRY(pci_nvme_aer) link; uint16_t cid; /* Command ID of the submitted AER */ }; /** Asynchronous Event Information - Notice */ typedef enum { PCI_NVME_AEI_NOTICE_NS_ATTR_CHANGED = 0, PCI_NVME_AEI_NOTICE_FW_ACTIVATION, PCI_NVME_AEI_NOTICE_TELEMETRY_CHANGE, PCI_NVME_AEI_NOTICE_ANA_CHANGE, PCI_NVME_AEI_NOTICE_PREDICT_LATENCY_CHANGE, PCI_NVME_AEI_NOTICE_LBA_STATUS_ALERT, PCI_NVME_AEI_NOTICE_ENDURANCE_GROUP_CHANGE, PCI_NVME_AEI_NOTICE_MAX, } pci_nvme_async_event_info_notice; #define PCI_NVME_AEI_NOTICE_SHIFT 8 #define PCI_NVME_AEI_NOTICE_MASK(event) (1 << (event + PCI_NVME_AEI_NOTICE_SHIFT)) /* Asynchronous Event Notifications */ struct pci_nvme_aen { pci_nvme_async_type atype; uint32_t event_data; bool posted; }; /* * By default, enable all Asynchrnous Event Notifications: * SMART / Health Critical Warnings * Namespace Attribute Notices */ #define PCI_NVME_AEN_DEFAULT_MASK 0x11f typedef enum { NVME_CNTRLTYPE_IO = 1, NVME_CNTRLTYPE_DISCOVERY = 2, NVME_CNTRLTYPE_ADMIN = 3, } pci_nvme_cntrl_type; struct pci_nvme_softc { struct pci_devinst *nsc_pi; pthread_mutex_t mtx; struct nvme_registers regs; struct nvme_namespace_data nsdata; struct nvme_controller_data ctrldata; struct nvme_error_information_entry err_log; struct nvme_health_information_page health_log; struct nvme_firmware_page fw_log; struct nvme_ns_list ns_log; struct pci_nvme_blockstore nvstore; uint16_t max_qentries; /* max entries per queue */ uint32_t max_queues; /* max number of IO SQ's or CQ's */ uint32_t num_cqueues; uint32_t num_squeues; bool num_q_is_set; /* Has host set Number of Queues */ struct pci_nvme_ioreq *ioreqs; STAILQ_HEAD(, pci_nvme_ioreq) ioreqs_free; /* free list of ioreqs */ uint32_t pending_ios; uint32_t ioslots; sem_t iosemlock; /* * Memory mapped Submission and Completion queues * Each array includes both Admin and IO queues */ struct nvme_completion_queue *compl_queues; struct nvme_submission_queue *submit_queues; struct nvme_feature_obj feat[NVME_FID_MAX]; enum nvme_dsm_type dataset_management; /* Accounting for SMART data */ __uint128_t read_data_units; __uint128_t write_data_units; __uint128_t read_commands; __uint128_t write_commands; uint32_t read_dunits_remainder; uint32_t write_dunits_remainder; STAILQ_HEAD(, pci_nvme_aer) aer_list; pthread_mutex_t aer_mtx; uint32_t aer_count; struct pci_nvme_aen aen[PCI_NVME_AE_TYPE_MAX]; pthread_t aen_tid; pthread_mutex_t aen_mtx; pthread_cond_t aen_cond; }; static void pci_nvme_cq_update(struct pci_nvme_softc *sc, struct nvme_completion_queue *cq, uint32_t cdw0, uint16_t cid, uint16_t sqid, uint16_t status); static struct pci_nvme_ioreq *pci_nvme_get_ioreq(struct pci_nvme_softc *); static void pci_nvme_release_ioreq(struct pci_nvme_softc *, struct pci_nvme_ioreq *); static void pci_nvme_io_done(struct blockif_req *, int); /* Controller Configuration utils */ #define NVME_CC_GET_EN(cc) \ ((cc) >> NVME_CC_REG_EN_SHIFT & NVME_CC_REG_EN_MASK) #define NVME_CC_GET_CSS(cc) \ ((cc) >> NVME_CC_REG_CSS_SHIFT & NVME_CC_REG_CSS_MASK) #define NVME_CC_GET_SHN(cc) \ ((cc) >> NVME_CC_REG_SHN_SHIFT & NVME_CC_REG_SHN_MASK) #define NVME_CC_GET_IOSQES(cc) \ ((cc) >> NVME_CC_REG_IOSQES_SHIFT & NVME_CC_REG_IOSQES_MASK) #define NVME_CC_GET_IOCQES(cc) \ ((cc) >> NVME_CC_REG_IOCQES_SHIFT & NVME_CC_REG_IOCQES_MASK) #define NVME_CC_WRITE_MASK \ ((NVME_CC_REG_EN_MASK << NVME_CC_REG_EN_SHIFT) | \ (NVME_CC_REG_IOSQES_MASK << NVME_CC_REG_IOSQES_SHIFT) | \ (NVME_CC_REG_IOCQES_MASK << NVME_CC_REG_IOCQES_SHIFT)) #define NVME_CC_NEN_WRITE_MASK \ ((NVME_CC_REG_CSS_MASK << NVME_CC_REG_CSS_SHIFT) | \ (NVME_CC_REG_MPS_MASK << NVME_CC_REG_MPS_SHIFT) | \ (NVME_CC_REG_AMS_MASK << NVME_CC_REG_AMS_SHIFT)) /* Controller Status utils */ #define NVME_CSTS_GET_RDY(sts) \ ((sts) >> NVME_CSTS_REG_RDY_SHIFT & NVME_CSTS_REG_RDY_MASK) #define NVME_CSTS_RDY (1 << NVME_CSTS_REG_RDY_SHIFT) #define NVME_CSTS_CFS (1 << NVME_CSTS_REG_CFS_SHIFT) /* Completion Queue status word utils */ #define NVME_STATUS_P (1 << NVME_STATUS_P_SHIFT) #define NVME_STATUS_MASK \ ((NVME_STATUS_SCT_MASK << NVME_STATUS_SCT_SHIFT) |\ (NVME_STATUS_SC_MASK << NVME_STATUS_SC_SHIFT)) #define NVME_ONCS_DSM (NVME_CTRLR_DATA_ONCS_DSM_MASK << \ NVME_CTRLR_DATA_ONCS_DSM_SHIFT) static void nvme_feature_invalid_cb(struct pci_nvme_softc *, struct nvme_feature_obj *, struct nvme_command *, struct nvme_completion *); static void nvme_feature_temperature(struct pci_nvme_softc *, struct nvme_feature_obj *, struct nvme_command *, struct nvme_completion *); static void nvme_feature_num_queues(struct pci_nvme_softc *, struct nvme_feature_obj *, struct nvme_command *, struct nvme_completion *); static void nvme_feature_iv_config(struct pci_nvme_softc *, struct nvme_feature_obj *, struct nvme_command *, struct nvme_completion *); static void nvme_feature_async_event(struct pci_nvme_softc *, struct nvme_feature_obj *, struct nvme_command *, struct nvme_completion *); static void *aen_thr(void *arg); static __inline void cpywithpad(char *dst, size_t dst_size, const char *src, char pad) { size_t len; len = strnlen(src, dst_size); memset(dst, pad, dst_size); memcpy(dst, src, len); } static __inline void pci_nvme_status_tc(uint16_t *status, uint16_t type, uint16_t code) { *status &= ~NVME_STATUS_MASK; *status |= (type & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT | (code & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT; } static __inline void pci_nvme_status_genc(uint16_t *status, uint16_t code) { pci_nvme_status_tc(status, NVME_SCT_GENERIC, code); } /* * Initialize the requested number or IO Submission and Completion Queues. * Admin queues are allocated implicitly. */ static void pci_nvme_init_queues(struct pci_nvme_softc *sc, uint32_t nsq, uint32_t ncq) { uint32_t i; /* * Allocate and initialize the Submission Queues */ if (nsq > NVME_QUEUES) { WPRINTF("%s: clamping number of SQ from %u to %u", __func__, nsq, NVME_QUEUES); nsq = NVME_QUEUES; } sc->num_squeues = nsq; sc->submit_queues = calloc(sc->num_squeues + 1, sizeof(struct nvme_submission_queue)); if (sc->submit_queues == NULL) { WPRINTF("%s: SQ allocation failed", __func__); sc->num_squeues = 0; } else { struct nvme_submission_queue *sq = sc->submit_queues; for (i = 0; i < sc->num_squeues + 1; i++) pthread_mutex_init(&sq[i].mtx, NULL); } /* * Allocate and initialize the Completion Queues */ if (ncq > NVME_QUEUES) { WPRINTF("%s: clamping number of CQ from %u to %u", __func__, ncq, NVME_QUEUES); ncq = NVME_QUEUES; } sc->num_cqueues = ncq; sc->compl_queues = calloc(sc->num_cqueues + 1, sizeof(struct nvme_completion_queue)); if (sc->compl_queues == NULL) { WPRINTF("%s: CQ allocation failed", __func__); sc->num_cqueues = 0; } else { struct nvme_completion_queue *cq = sc->compl_queues; for (i = 0; i < sc->num_cqueues + 1; i++) pthread_mutex_init(&cq[i].mtx, NULL); } } static void pci_nvme_init_ctrldata(struct pci_nvme_softc *sc) { struct nvme_controller_data *cd = &sc->ctrldata; cd->vid = 0xFB5D; cd->ssvid = 0x0000; cpywithpad((char *)cd->mn, sizeof(cd->mn), "bhyve-NVMe", ' '); cpywithpad((char *)cd->fr, sizeof(cd->fr), "1.0", ' '); /* Num of submission commands that we can handle at a time (2^rab) */ cd->rab = 4; /* FreeBSD OUI */ cd->ieee[0] = 0x58; cd->ieee[1] = 0x9c; cd->ieee[2] = 0xfc; cd->mic = 0; cd->mdts = NVME_MDTS; /* max data transfer size (2^mdts * CAP.MPSMIN) */ cd->ver = NVME_REV(1,4); cd->cntrltype = NVME_CNTRLTYPE_IO; cd->oacs = 1 << NVME_CTRLR_DATA_OACS_FORMAT_SHIFT; cd->oaes = NVMEB(NVME_CTRLR_DATA_OAES_NS_ATTR); cd->acl = 2; cd->aerl = 4; /* Advertise 1, Read-only firmware slot */ cd->frmw = NVMEB(NVME_CTRLR_DATA_FRMW_SLOT1_RO) | (1 << NVME_CTRLR_DATA_FRMW_NUM_SLOTS_SHIFT); cd->lpa = 0; /* TODO: support some simple things like SMART */ cd->elpe = 0; /* max error log page entries */ /* * Report a single power state (zero-based value) * power_state[] values are left as zero to indicate "Not reported" */ cd->npss = 0; /* Warning Composite Temperature Threshold */ cd->wctemp = 0x0157; cd->cctemp = 0x0157; /* SANICAP must not be 0 for Revision 1.4 and later NVMe Controllers */ cd->sanicap = (NVME_CTRLR_DATA_SANICAP_NODMMAS_NO << NVME_CTRLR_DATA_SANICAP_NODMMAS_SHIFT); cd->sqes = (6 << NVME_CTRLR_DATA_SQES_MAX_SHIFT) | (6 << NVME_CTRLR_DATA_SQES_MIN_SHIFT); cd->cqes = (4 << NVME_CTRLR_DATA_CQES_MAX_SHIFT) | (4 << NVME_CTRLR_DATA_CQES_MIN_SHIFT); cd->nn = 1; /* number of namespaces */ cd->oncs = 0; switch (sc->dataset_management) { case NVME_DATASET_MANAGEMENT_AUTO: if (sc->nvstore.deallocate) cd->oncs |= NVME_ONCS_DSM; break; case NVME_DATASET_MANAGEMENT_ENABLE: cd->oncs |= NVME_ONCS_DSM; break; default: break; } cd->fna = NVME_CTRLR_DATA_FNA_FORMAT_ALL_MASK << NVME_CTRLR_DATA_FNA_FORMAT_ALL_SHIFT; cd->vwc = NVME_CTRLR_DATA_VWC_ALL_NO << NVME_CTRLR_DATA_VWC_ALL_SHIFT; } static void pci_nvme_init_nsdata_size(struct pci_nvme_blockstore *nvstore, struct nvme_namespace_data *nd) { /* Get capacity and block size information from backing store */ nd->nsze = nvstore->size / nvstore->sectsz; nd->ncap = nd->nsze; nd->nuse = nd->nsze; } static void pci_nvme_init_nsdata(struct pci_nvme_softc *sc, struct nvme_namespace_data *nd, uint32_t nsid, struct pci_nvme_blockstore *nvstore) { pci_nvme_init_nsdata_size(nvstore, nd); if (nvstore->type == NVME_STOR_BLOCKIF) nvstore->deallocate = blockif_candelete(nvstore->ctx); nd->nlbaf = 0; /* NLBAF is a 0's based value (i.e. 1 LBA Format) */ nd->flbas = 0; /* Create an EUI-64 if user did not provide one */ if (nvstore->eui64 == 0) { char *data = NULL; uint64_t eui64 = nvstore->eui64; asprintf(&data, "%s%u%u%u", get_config_value("name"), sc->nsc_pi->pi_bus, sc->nsc_pi->pi_slot, sc->nsc_pi->pi_func); if (data != NULL) { eui64 = OUI_FREEBSD_NVME_LOW | crc16(0, data, strlen(data)); free(data); } nvstore->eui64 = (eui64 << 16) | (nsid & 0xffff); } be64enc(nd->eui64, nvstore->eui64); /* LBA data-sz = 2^lbads */ nd->lbaf[0] = nvstore->sectsz_bits << NVME_NS_DATA_LBAF_LBADS_SHIFT; } static void pci_nvme_init_logpages(struct pci_nvme_softc *sc) { __uint128_t power_cycles = 1; memset(&sc->err_log, 0, sizeof(sc->err_log)); memset(&sc->health_log, 0, sizeof(sc->health_log)); memset(&sc->fw_log, 0, sizeof(sc->fw_log)); memset(&sc->ns_log, 0, sizeof(sc->ns_log)); /* Set read/write remainder to round up according to spec */ sc->read_dunits_remainder = 999; sc->write_dunits_remainder = 999; /* Set nominal Health values checked by implementations */ sc->health_log.temperature = NVME_TEMPERATURE; sc->health_log.available_spare = 100; sc->health_log.available_spare_threshold = 10; /* Set Active Firmware Info to slot 1 */ sc->fw_log.afi = (1 << NVME_FIRMWARE_PAGE_AFI_SLOT_SHIFT); memcpy(&sc->fw_log.revision[0], sc->ctrldata.fr, sizeof(sc->fw_log.revision[0])); memcpy(&sc->health_log.power_cycles, &power_cycles, sizeof(sc->health_log.power_cycles)); } static void pci_nvme_init_features(struct pci_nvme_softc *sc) { enum nvme_feature fid; for (fid = 0; fid < NVME_FID_MAX; fid++) { switch (fid) { case NVME_FEAT_ARBITRATION: case NVME_FEAT_POWER_MANAGEMENT: case NVME_FEAT_INTERRUPT_COALESCING: //XXX case NVME_FEAT_WRITE_ATOMICITY: /* Mandatory but no special handling required */ //XXX hang - case NVME_FEAT_PREDICTABLE_LATENCY_MODE_CONFIG: //XXX hang - case NVME_FEAT_HOST_BEHAVIOR_SUPPORT: // this returns a data buffer break; case NVME_FEAT_TEMPERATURE_THRESHOLD: sc->feat[fid].set = nvme_feature_temperature; break; case NVME_FEAT_ERROR_RECOVERY: sc->feat[fid].namespace_specific = true; break; case NVME_FEAT_NUMBER_OF_QUEUES: sc->feat[fid].set = nvme_feature_num_queues; break; case NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION: sc->feat[fid].set = nvme_feature_iv_config; break; case NVME_FEAT_ASYNC_EVENT_CONFIGURATION: sc->feat[fid].set = nvme_feature_async_event; /* Enable all AENs by default */ sc->feat[fid].cdw11 = PCI_NVME_AEN_DEFAULT_MASK; break; default: sc->feat[fid].set = nvme_feature_invalid_cb; sc->feat[fid].get = nvme_feature_invalid_cb; } } } static void pci_nvme_aer_reset(struct pci_nvme_softc *sc) { STAILQ_INIT(&sc->aer_list); sc->aer_count = 0; } static void pci_nvme_aer_init(struct pci_nvme_softc *sc) { pthread_mutex_init(&sc->aer_mtx, NULL); pci_nvme_aer_reset(sc); } static void pci_nvme_aer_destroy(struct pci_nvme_softc *sc) { struct pci_nvme_aer *aer = NULL; pthread_mutex_lock(&sc->aer_mtx); while (!STAILQ_EMPTY(&sc->aer_list)) { aer = STAILQ_FIRST(&sc->aer_list); STAILQ_REMOVE_HEAD(&sc->aer_list, link); free(aer); } pthread_mutex_unlock(&sc->aer_mtx); pci_nvme_aer_reset(sc); } static bool pci_nvme_aer_available(struct pci_nvme_softc *sc) { return (sc->aer_count != 0); } static bool pci_nvme_aer_limit_reached(struct pci_nvme_softc *sc) { struct nvme_controller_data *cd = &sc->ctrldata; /* AERL is a zero based value while aer_count is one's based */ return (sc->aer_count == (cd->aerl + 1U)); } /* * Add an Async Event Request * * Stores an AER to be returned later if the Controller needs to notify the * host of an event. * Note that while the NVMe spec doesn't require Controllers to return AER's * in order, this implementation does preserve the order. */ static int pci_nvme_aer_add(struct pci_nvme_softc *sc, uint16_t cid) { struct pci_nvme_aer *aer = NULL; aer = calloc(1, sizeof(struct pci_nvme_aer)); if (aer == NULL) return (-1); /* Save the Command ID for use in the completion message */ aer->cid = cid; pthread_mutex_lock(&sc->aer_mtx); sc->aer_count++; STAILQ_INSERT_TAIL(&sc->aer_list, aer, link); pthread_mutex_unlock(&sc->aer_mtx); return (0); } /* * Get an Async Event Request structure * * Returns a pointer to an AER previously submitted by the host or NULL if * no AER's exist. Caller is responsible for freeing the returned struct. */ static struct pci_nvme_aer * pci_nvme_aer_get(struct pci_nvme_softc *sc) { struct pci_nvme_aer *aer = NULL; pthread_mutex_lock(&sc->aer_mtx); aer = STAILQ_FIRST(&sc->aer_list); if (aer != NULL) { STAILQ_REMOVE_HEAD(&sc->aer_list, link); sc->aer_count--; } pthread_mutex_unlock(&sc->aer_mtx); return (aer); } static void pci_nvme_aen_reset(struct pci_nvme_softc *sc) { uint32_t atype; memset(sc->aen, 0, PCI_NVME_AE_TYPE_MAX * sizeof(struct pci_nvme_aen)); for (atype = 0; atype < PCI_NVME_AE_TYPE_MAX; atype++) { sc->aen[atype].atype = atype; } } static void pci_nvme_aen_init(struct pci_nvme_softc *sc) { char nstr[80]; pci_nvme_aen_reset(sc); pthread_mutex_init(&sc->aen_mtx, NULL); pthread_create(&sc->aen_tid, NULL, aen_thr, sc); snprintf(nstr, sizeof(nstr), "nvme-aen-%d:%d", sc->nsc_pi->pi_slot, sc->nsc_pi->pi_func); pthread_set_name_np(sc->aen_tid, nstr); } static void pci_nvme_aen_destroy(struct pci_nvme_softc *sc) { pci_nvme_aen_reset(sc); } /* Notify the AEN thread of pending work */ static void pci_nvme_aen_notify(struct pci_nvme_softc *sc) { pthread_cond_signal(&sc->aen_cond); } /* * Post an Asynchronous Event Notification */ static int32_t pci_nvme_aen_post(struct pci_nvme_softc *sc, pci_nvme_async_type atype, uint32_t event_data) { struct pci_nvme_aen *aen; if (atype >= PCI_NVME_AE_TYPE_MAX) { return(EINVAL); } pthread_mutex_lock(&sc->aen_mtx); aen = &sc->aen[atype]; /* Has the controller already posted an event of this type? */ if (aen->posted) { pthread_mutex_unlock(&sc->aen_mtx); return(EALREADY); } aen->event_data = event_data; aen->posted = true; pthread_mutex_unlock(&sc->aen_mtx); pci_nvme_aen_notify(sc); return(0); } static void pci_nvme_aen_process(struct pci_nvme_softc *sc) { struct pci_nvme_aer *aer; struct pci_nvme_aen *aen; pci_nvme_async_type atype; uint32_t mask; uint16_t status; uint8_t lid; assert(pthread_mutex_isowned_np(&sc->aen_mtx)); for (atype = 0; atype < PCI_NVME_AE_TYPE_MAX; atype++) { aen = &sc->aen[atype]; /* Previous iterations may have depleted the available AER's */ if (!pci_nvme_aer_available(sc)) { DPRINTF("%s: no AER", __func__); break; } if (!aen->posted) { DPRINTF("%s: no AEN posted for atype=%#x", __func__, atype); continue; } status = NVME_SC_SUCCESS; /* Is the event masked? */ mask = sc->feat[NVME_FEAT_ASYNC_EVENT_CONFIGURATION].cdw11; DPRINTF("%s: atype=%#x mask=%#x event_data=%#x", __func__, atype, mask, aen->event_data); switch (atype) { case PCI_NVME_AE_TYPE_ERROR: lid = NVME_LOG_ERROR; break; case PCI_NVME_AE_TYPE_SMART: mask &= 0xff; if ((mask & aen->event_data) == 0) continue; lid = NVME_LOG_HEALTH_INFORMATION; break; case PCI_NVME_AE_TYPE_NOTICE: if (aen->event_data >= PCI_NVME_AEI_NOTICE_MAX) { EPRINTLN("%s unknown AEN notice type %u", __func__, aen->event_data); status = NVME_SC_INTERNAL_DEVICE_ERROR; lid = 0; break; } if ((PCI_NVME_AEI_NOTICE_MASK(aen->event_data) & mask) == 0) continue; switch (aen->event_data) { case PCI_NVME_AEI_NOTICE_NS_ATTR_CHANGED: lid = NVME_LOG_CHANGED_NAMESPACE; break; case PCI_NVME_AEI_NOTICE_FW_ACTIVATION: lid = NVME_LOG_FIRMWARE_SLOT; break; case PCI_NVME_AEI_NOTICE_TELEMETRY_CHANGE: lid = NVME_LOG_TELEMETRY_CONTROLLER_INITIATED; break; case PCI_NVME_AEI_NOTICE_ANA_CHANGE: lid = NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS; break; case PCI_NVME_AEI_NOTICE_PREDICT_LATENCY_CHANGE: lid = NVME_LOG_PREDICTABLE_LATENCY_EVENT_AGGREGATE; break; case PCI_NVME_AEI_NOTICE_LBA_STATUS_ALERT: lid = NVME_LOG_LBA_STATUS_INFORMATION; break; case PCI_NVME_AEI_NOTICE_ENDURANCE_GROUP_CHANGE: lid = NVME_LOG_ENDURANCE_GROUP_EVENT_AGGREGATE; break; default: lid = 0; } break; default: /* bad type?!? */ EPRINTLN("%s unknown AEN type %u", __func__, atype); status = NVME_SC_INTERNAL_DEVICE_ERROR; lid = 0; break; } aer = pci_nvme_aer_get(sc); assert(aer != NULL); DPRINTF("%s: CID=%#x CDW0=%#x", __func__, aer->cid, (lid << 16) | (aen->event_data << 8) | atype); pci_nvme_cq_update(sc, &sc->compl_queues[0], (lid << 16) | (aen->event_data << 8) | atype, /* cdw0 */ aer->cid, 0, /* SQID */ status); aen->event_data = 0; aen->posted = false; pci_generate_msix(sc->nsc_pi, 0); } } static void * aen_thr(void *arg) { struct pci_nvme_softc *sc; sc = arg; pthread_mutex_lock(&sc->aen_mtx); for (;;) { pci_nvme_aen_process(sc); pthread_cond_wait(&sc->aen_cond, &sc->aen_mtx); } pthread_mutex_unlock(&sc->aen_mtx); pthread_exit(NULL); return (NULL); } static void pci_nvme_reset_locked(struct pci_nvme_softc *sc) { uint32_t i; DPRINTF("%s", __func__); sc->regs.cap_lo = (ZERO_BASED(sc->max_qentries) & NVME_CAP_LO_REG_MQES_MASK) | (1 << NVME_CAP_LO_REG_CQR_SHIFT) | (60 << NVME_CAP_LO_REG_TO_SHIFT); sc->regs.cap_hi = 1 << NVME_CAP_HI_REG_CSS_NVM_SHIFT; sc->regs.vs = NVME_REV(1,4); /* NVMe v1.4 */ sc->regs.cc = 0; assert(sc->submit_queues != NULL); for (i = 0; i < sc->num_squeues + 1; i++) { sc->submit_queues[i].qbase = NULL; sc->submit_queues[i].size = 0; sc->submit_queues[i].cqid = 0; sc->submit_queues[i].tail = 0; sc->submit_queues[i].head = 0; } assert(sc->compl_queues != NULL); for (i = 0; i < sc->num_cqueues + 1; i++) { sc->compl_queues[i].qbase = NULL; sc->compl_queues[i].size = 0; sc->compl_queues[i].tail = 0; sc->compl_queues[i].head = 0; } sc->num_q_is_set = false; pci_nvme_aer_destroy(sc); pci_nvme_aen_destroy(sc); /* * Clear CSTS.RDY last to prevent the host from enabling Controller * before cleanup completes */ sc->regs.csts = 0; } static void pci_nvme_reset(struct pci_nvme_softc *sc) { pthread_mutex_lock(&sc->mtx); pci_nvme_reset_locked(sc); pthread_mutex_unlock(&sc->mtx); } static int pci_nvme_init_controller(struct pci_nvme_softc *sc) { uint16_t acqs, asqs; DPRINTF("%s", __func__); /* * NVMe 2.0 states that "enabling a controller while this field is * cleared to 0h produces undefined results" for both ACQS and * ASQS. If zero, set CFS and do not become ready. */ asqs = ONE_BASED(sc->regs.aqa & NVME_AQA_REG_ASQS_MASK); if (asqs < 2) { EPRINTLN("%s: illegal ASQS value %#x (aqa=%#x)", __func__, asqs - 1, sc->regs.aqa); sc->regs.csts |= NVME_CSTS_CFS; return (-1); } sc->submit_queues[0].size = asqs; sc->submit_queues[0].qbase = vm_map_gpa(sc->nsc_pi->pi_vmctx, sc->regs.asq, sizeof(struct nvme_command) * asqs); if (sc->submit_queues[0].qbase == NULL) { EPRINTLN("%s: ASQ vm_map_gpa(%lx) failed", __func__, sc->regs.asq); sc->regs.csts |= NVME_CSTS_CFS; return (-1); } DPRINTF("%s mapping Admin-SQ guest 0x%lx, host: %p", __func__, sc->regs.asq, sc->submit_queues[0].qbase); acqs = ONE_BASED((sc->regs.aqa >> NVME_AQA_REG_ACQS_SHIFT) & NVME_AQA_REG_ACQS_MASK); if (acqs < 2) { EPRINTLN("%s: illegal ACQS value %#x (aqa=%#x)", __func__, acqs - 1, sc->regs.aqa); sc->regs.csts |= NVME_CSTS_CFS; return (-1); } sc->compl_queues[0].size = acqs; sc->compl_queues[0].qbase = vm_map_gpa(sc->nsc_pi->pi_vmctx, sc->regs.acq, sizeof(struct nvme_completion) * acqs); if (sc->compl_queues[0].qbase == NULL) { EPRINTLN("%s: ACQ vm_map_gpa(%lx) failed", __func__, sc->regs.acq); sc->regs.csts |= NVME_CSTS_CFS; return (-1); } sc->compl_queues[0].intr_en = NVME_CQ_INTEN; DPRINTF("%s mapping Admin-CQ guest 0x%lx, host: %p", __func__, sc->regs.acq, sc->compl_queues[0].qbase); return (0); } static int nvme_prp_memcpy(struct vmctx *ctx, uint64_t prp1, uint64_t prp2, uint8_t *b, size_t len, enum nvme_copy_dir dir) { uint8_t *p; size_t bytes; if (len > (8 * 1024)) { return (-1); } /* Copy from the start of prp1 to the end of the physical page */ bytes = PAGE_SIZE - (prp1 & PAGE_MASK); bytes = MIN(bytes, len); p = vm_map_gpa(ctx, prp1, bytes); if (p == NULL) { return (-1); } if (dir == NVME_COPY_TO_PRP) memcpy(p, b, bytes); else memcpy(b, p, bytes); b += bytes; len -= bytes; if (len == 0) { return (0); } len = MIN(len, PAGE_SIZE); p = vm_map_gpa(ctx, prp2, len); if (p == NULL) { return (-1); } if (dir == NVME_COPY_TO_PRP) memcpy(p, b, len); else memcpy(b, p, len); return (0); } /* * Write a Completion Queue Entry update * * Write the completion and update the doorbell value */ static void pci_nvme_cq_update(struct pci_nvme_softc *sc, struct nvme_completion_queue *cq, uint32_t cdw0, uint16_t cid, uint16_t sqid, uint16_t status) { struct nvme_submission_queue *sq = &sc->submit_queues[sqid]; struct nvme_completion *cqe; assert(cq->qbase != NULL); pthread_mutex_lock(&cq->mtx); cqe = &cq->qbase[cq->tail]; /* Flip the phase bit */ status |= (cqe->status ^ NVME_STATUS_P) & NVME_STATUS_P_MASK; cqe->cdw0 = cdw0; cqe->sqhd = sq->head; cqe->sqid = sqid; cqe->cid = cid; cqe->status = status; cq->tail++; if (cq->tail >= cq->size) { cq->tail = 0; } pthread_mutex_unlock(&cq->mtx); } static int nvme_opc_delete_io_sq(struct pci_nvme_softc* sc, struct nvme_command* command, struct nvme_completion* compl) { uint16_t qid = command->cdw10 & 0xffff; DPRINTF("%s DELETE_IO_SQ %u", __func__, qid); if (qid == 0 || qid > sc->num_squeues || (sc->submit_queues[qid].qbase == NULL)) { WPRINTF("%s NOT PERMITTED queue id %u / num_squeues %u", __func__, qid, sc->num_squeues); pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_INVALID_QUEUE_IDENTIFIER); return (1); } sc->submit_queues[qid].qbase = NULL; sc->submit_queues[qid].cqid = 0; pci_nvme_status_genc(&compl->status, NVME_SC_SUCCESS); return (1); } static int nvme_opc_create_io_sq(struct pci_nvme_softc* sc, struct nvme_command* command, struct nvme_completion* compl) { if (command->cdw11 & NVME_CMD_CDW11_PC) { uint16_t qid = command->cdw10 & 0xffff; struct nvme_submission_queue *nsq; if ((qid == 0) || (qid > sc->num_squeues) || (sc->submit_queues[qid].qbase != NULL)) { WPRINTF("%s queue index %u > num_squeues %u", __func__, qid, sc->num_squeues); pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_INVALID_QUEUE_IDENTIFIER); return (1); } nsq = &sc->submit_queues[qid]; nsq->size = ONE_BASED((command->cdw10 >> 16) & 0xffff); DPRINTF("%s size=%u (max=%u)", __func__, nsq->size, sc->max_qentries); if ((nsq->size < 2) || (nsq->size > sc->max_qentries)) { /* * Queues must specify at least two entries * NOTE: "MAXIMUM QUEUE SIZE EXCEEDED" was renamed to * "INVALID QUEUE SIZE" in the NVM Express 1.3 Spec */ pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_MAXIMUM_QUEUE_SIZE_EXCEEDED); return (1); } nsq->head = nsq->tail = 0; nsq->cqid = (command->cdw11 >> 16) & 0xffff; if ((nsq->cqid == 0) || (nsq->cqid > sc->num_cqueues)) { pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_INVALID_QUEUE_IDENTIFIER); return (1); } if (sc->compl_queues[nsq->cqid].qbase == NULL) { pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_COMPLETION_QUEUE_INVALID); return (1); } nsq->qpriority = (command->cdw11 >> 1) & 0x03; nsq->qbase = vm_map_gpa(sc->nsc_pi->pi_vmctx, command->prp1, sizeof(struct nvme_command) * (size_t)nsq->size); DPRINTF("%s sq %u size %u gaddr %p cqid %u", __func__, qid, nsq->size, nsq->qbase, nsq->cqid); pci_nvme_status_genc(&compl->status, NVME_SC_SUCCESS); DPRINTF("%s completed creating IOSQ qid %u", __func__, qid); } else { /* * Guest sent non-cont submission queue request. * This setting is unsupported by this emulation. */ WPRINTF("%s unsupported non-contig (list-based) " "create i/o submission queue", __func__); pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); } return (1); } static int nvme_opc_delete_io_cq(struct pci_nvme_softc* sc, struct nvme_command* command, struct nvme_completion* compl) { uint16_t qid = command->cdw10 & 0xffff; uint16_t sqid; DPRINTF("%s DELETE_IO_CQ %u", __func__, qid); if (qid == 0 || qid > sc->num_cqueues || (sc->compl_queues[qid].qbase == NULL)) { WPRINTF("%s queue index %u / num_cqueues %u", __func__, qid, sc->num_cqueues); pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_INVALID_QUEUE_IDENTIFIER); return (1); } /* Deleting an Active CQ is an error */ for (sqid = 1; sqid < sc->num_squeues + 1; sqid++) if (sc->submit_queues[sqid].cqid == qid) { pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_INVALID_QUEUE_DELETION); return (1); } sc->compl_queues[qid].qbase = NULL; pci_nvme_status_genc(&compl->status, NVME_SC_SUCCESS); return (1); } static int nvme_opc_create_io_cq(struct pci_nvme_softc* sc, struct nvme_command* command, struct nvme_completion* compl) { struct nvme_completion_queue *ncq; uint16_t qid = command->cdw10 & 0xffff; /* Only support Physically Contiguous queues */ if ((command->cdw11 & NVME_CMD_CDW11_PC) == 0) { WPRINTF("%s unsupported non-contig (list-based) " "create i/o completion queue", __func__); pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); return (1); } if ((qid == 0) || (qid > sc->num_cqueues) || (sc->compl_queues[qid].qbase != NULL)) { WPRINTF("%s queue index %u > num_cqueues %u", __func__, qid, sc->num_cqueues); pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_INVALID_QUEUE_IDENTIFIER); return (1); } ncq = &sc->compl_queues[qid]; ncq->intr_en = (command->cdw11 & NVME_CMD_CDW11_IEN) >> 1; ncq->intr_vec = (command->cdw11 >> 16) & 0xffff; if (ncq->intr_vec > (sc->max_queues + 1)) { pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_INVALID_INTERRUPT_VECTOR); return (1); } ncq->size = ONE_BASED((command->cdw10 >> 16) & 0xffff); if ((ncq->size < 2) || (ncq->size > sc->max_qentries)) { /* * Queues must specify at least two entries * NOTE: "MAXIMUM QUEUE SIZE EXCEEDED" was renamed to * "INVALID QUEUE SIZE" in the NVM Express 1.3 Spec */ pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_MAXIMUM_QUEUE_SIZE_EXCEEDED); return (1); } ncq->head = ncq->tail = 0; ncq->qbase = vm_map_gpa(sc->nsc_pi->pi_vmctx, command->prp1, sizeof(struct nvme_command) * (size_t)ncq->size); pci_nvme_status_genc(&compl->status, NVME_SC_SUCCESS); return (1); } static int nvme_opc_get_log_page(struct pci_nvme_softc* sc, struct nvme_command* command, struct nvme_completion* compl) { uint64_t logoff; uint32_t logsize; uint8_t logpage; pci_nvme_status_genc(&compl->status, NVME_SC_SUCCESS); /* * Command specifies the number of dwords to return in fields NUMDU * and NUMDL. This is a zero-based value. */ logpage = command->cdw10 & 0xFF; logsize = ((command->cdw11 << 16) | (command->cdw10 >> 16)) + 1; logsize *= sizeof(uint32_t); logoff = ((uint64_t)(command->cdw13) << 32) | command->cdw12; DPRINTF("%s log page %u len %u", __func__, logpage, logsize); switch (logpage) { case NVME_LOG_ERROR: if (logoff >= sizeof(sc->err_log)) { pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); break; } nvme_prp_memcpy(sc->nsc_pi->pi_vmctx, command->prp1, command->prp2, (uint8_t *)&sc->err_log + logoff, MIN(logsize - logoff, sizeof(sc->err_log)), NVME_COPY_TO_PRP); break; case NVME_LOG_HEALTH_INFORMATION: if (logoff >= sizeof(sc->health_log)) { pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); break; } pthread_mutex_lock(&sc->mtx); memcpy(&sc->health_log.data_units_read, &sc->read_data_units, sizeof(sc->health_log.data_units_read)); memcpy(&sc->health_log.data_units_written, &sc->write_data_units, sizeof(sc->health_log.data_units_written)); memcpy(&sc->health_log.host_read_commands, &sc->read_commands, sizeof(sc->health_log.host_read_commands)); memcpy(&sc->health_log.host_write_commands, &sc->write_commands, sizeof(sc->health_log.host_write_commands)); pthread_mutex_unlock(&sc->mtx); nvme_prp_memcpy(sc->nsc_pi->pi_vmctx, command->prp1, command->prp2, (uint8_t *)&sc->health_log + logoff, MIN(logsize - logoff, sizeof(sc->health_log)), NVME_COPY_TO_PRP); break; case NVME_LOG_FIRMWARE_SLOT: if (logoff >= sizeof(sc->fw_log)) { pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); break; } nvme_prp_memcpy(sc->nsc_pi->pi_vmctx, command->prp1, command->prp2, (uint8_t *)&sc->fw_log + logoff, MIN(logsize - logoff, sizeof(sc->fw_log)), NVME_COPY_TO_PRP); break; case NVME_LOG_CHANGED_NAMESPACE: if (logoff >= sizeof(sc->ns_log)) { pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); break; } nvme_prp_memcpy(sc->nsc_pi->pi_vmctx, command->prp1, command->prp2, (uint8_t *)&sc->ns_log + logoff, MIN(logsize - logoff, sizeof(sc->ns_log)), NVME_COPY_TO_PRP); memset(&sc->ns_log, 0, sizeof(sc->ns_log)); break; default: DPRINTF("%s get log page %x command not supported", __func__, logpage); pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_INVALID_LOG_PAGE); } return (1); } static int nvme_opc_identify(struct pci_nvme_softc* sc, struct nvme_command* command, struct nvme_completion* compl) { void *dest; uint16_t status; DPRINTF("%s identify 0x%x nsid 0x%x", __func__, command->cdw10 & 0xFF, command->nsid); status = 0; pci_nvme_status_genc(&status, NVME_SC_SUCCESS); switch (command->cdw10 & 0xFF) { case 0x00: /* return Identify Namespace data structure */ /* Global NS only valid with NS Management */ if (command->nsid == NVME_GLOBAL_NAMESPACE_TAG) { pci_nvme_status_genc(&status, NVME_SC_INVALID_NAMESPACE_OR_FORMAT); break; } nvme_prp_memcpy(sc->nsc_pi->pi_vmctx, command->prp1, command->prp2, (uint8_t *)&sc->nsdata, sizeof(sc->nsdata), NVME_COPY_TO_PRP); break; case 0x01: /* return Identify Controller data structure */ nvme_prp_memcpy(sc->nsc_pi->pi_vmctx, command->prp1, command->prp2, (uint8_t *)&sc->ctrldata, sizeof(sc->ctrldata), NVME_COPY_TO_PRP); break; case 0x02: /* list of 1024 active NSIDs > CDW1.NSID */ dest = vm_map_gpa(sc->nsc_pi->pi_vmctx, command->prp1, sizeof(uint32_t) * 1024); /* All unused entries shall be zero */ memset(dest, 0, sizeof(uint32_t) * 1024); ((uint32_t *)dest)[0] = 1; break; case 0x03: /* list of NSID structures in CDW1.NSID, 4096 bytes */ if (command->nsid != 1) { pci_nvme_status_genc(&status, NVME_SC_INVALID_NAMESPACE_OR_FORMAT); break; } dest = vm_map_gpa(sc->nsc_pi->pi_vmctx, command->prp1, sizeof(uint32_t) * 1024); /* All bytes after the descriptor shall be zero */ memset(dest, 0, sizeof(uint32_t) * 1024); /* Return NIDT=1 (i.e. EUI64) descriptor */ ((uint8_t *)dest)[0] = 1; ((uint8_t *)dest)[1] = sizeof(uint64_t); memcpy(((uint8_t *)dest) + 4, sc->nsdata.eui64, sizeof(uint64_t)); break; case 0x13: /* * Controller list is optional but used by UNH tests. Return * a valid but empty list. */ dest = vm_map_gpa(sc->nsc_pi->pi_vmctx, command->prp1, sizeof(uint16_t) * 2048); memset(dest, 0, sizeof(uint16_t) * 2048); break; default: DPRINTF("%s unsupported identify command requested 0x%x", __func__, command->cdw10 & 0xFF); pci_nvme_status_genc(&status, NVME_SC_INVALID_FIELD); break; } compl->status = status; return (1); } static const char * nvme_fid_to_name(uint8_t fid) { const char *name; switch (fid) { case NVME_FEAT_ARBITRATION: name = "Arbitration"; break; case NVME_FEAT_POWER_MANAGEMENT: name = "Power Management"; break; case NVME_FEAT_LBA_RANGE_TYPE: name = "LBA Range Type"; break; case NVME_FEAT_TEMPERATURE_THRESHOLD: name = "Temperature Threshold"; break; case NVME_FEAT_ERROR_RECOVERY: name = "Error Recovery"; break; case NVME_FEAT_VOLATILE_WRITE_CACHE: name = "Volatile Write Cache"; break; case NVME_FEAT_NUMBER_OF_QUEUES: name = "Number of Queues"; break; case NVME_FEAT_INTERRUPT_COALESCING: name = "Interrupt Coalescing"; break; case NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION: name = "Interrupt Vector Configuration"; break; case NVME_FEAT_WRITE_ATOMICITY: name = "Write Atomicity Normal"; break; case NVME_FEAT_ASYNC_EVENT_CONFIGURATION: name = "Asynchronous Event Configuration"; break; case NVME_FEAT_AUTONOMOUS_POWER_STATE_TRANSITION: name = "Autonomous Power State Transition"; break; case NVME_FEAT_HOST_MEMORY_BUFFER: name = "Host Memory Buffer"; break; case NVME_FEAT_TIMESTAMP: name = "Timestamp"; break; case NVME_FEAT_KEEP_ALIVE_TIMER: name = "Keep Alive Timer"; break; case NVME_FEAT_HOST_CONTROLLED_THERMAL_MGMT: name = "Host Controlled Thermal Management"; break; case NVME_FEAT_NON_OP_POWER_STATE_CONFIG: name = "Non-Operation Power State Config"; break; case NVME_FEAT_READ_RECOVERY_LEVEL_CONFIG: name = "Read Recovery Level Config"; break; case NVME_FEAT_PREDICTABLE_LATENCY_MODE_CONFIG: name = "Predictable Latency Mode Config"; break; case NVME_FEAT_PREDICTABLE_LATENCY_MODE_WINDOW: name = "Predictable Latency Mode Window"; break; case NVME_FEAT_LBA_STATUS_INFORMATION_ATTRIBUTES: name = "LBA Status Information Report Interval"; break; case NVME_FEAT_HOST_BEHAVIOR_SUPPORT: name = "Host Behavior Support"; break; case NVME_FEAT_SANITIZE_CONFIG: name = "Sanitize Config"; break; case NVME_FEAT_ENDURANCE_GROUP_EVENT_CONFIGURATION: name = "Endurance Group Event Configuration"; break; case NVME_FEAT_SOFTWARE_PROGRESS_MARKER: name = "Software Progress Marker"; break; case NVME_FEAT_HOST_IDENTIFIER: name = "Host Identifier"; break; case NVME_FEAT_RESERVATION_NOTIFICATION_MASK: name = "Reservation Notification Mask"; break; case NVME_FEAT_RESERVATION_PERSISTENCE: name = "Reservation Persistence"; break; case NVME_FEAT_NAMESPACE_WRITE_PROTECTION_CONFIG: name = "Namespace Write Protection Config"; break; default: name = "Unknown"; break; } return (name); } static void nvme_feature_invalid_cb(struct pci_nvme_softc *sc __unused, struct nvme_feature_obj *feat __unused, struct nvme_command *command __unused, struct nvme_completion *compl) { pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); } static void nvme_feature_iv_config(struct pci_nvme_softc *sc, struct nvme_feature_obj *feat __unused, struct nvme_command *command, struct nvme_completion *compl) { uint32_t i; uint32_t cdw11 = command->cdw11; uint16_t iv; bool cd; pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); iv = cdw11 & 0xffff; cd = cdw11 & (1 << 16); if (iv > (sc->max_queues + 1)) { return; } /* No Interrupt Coalescing (i.e. not Coalescing Disable) for Admin Q */ if ((iv == 0) && !cd) return; /* Requested Interrupt Vector must be used by a CQ */ for (i = 0; i < sc->num_cqueues + 1; i++) { if (sc->compl_queues[i].intr_vec == iv) { pci_nvme_status_genc(&compl->status, NVME_SC_SUCCESS); } } } #define NVME_ASYNC_EVENT_ENDURANCE_GROUP (0x4000) static void nvme_feature_async_event(struct pci_nvme_softc *sc __unused, struct nvme_feature_obj *feat __unused, struct nvme_command *command, struct nvme_completion *compl) { if (command->cdw11 & NVME_ASYNC_EVENT_ENDURANCE_GROUP) pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); } #define NVME_TEMP_THRESH_OVER 0 #define NVME_TEMP_THRESH_UNDER 1 static void nvme_feature_temperature(struct pci_nvme_softc *sc, struct nvme_feature_obj *feat __unused, struct nvme_command *command, struct nvme_completion *compl) { uint16_t tmpth; /* Temperature Threshold */ uint8_t tmpsel; /* Threshold Temperature Select */ uint8_t thsel; /* Threshold Type Select */ bool set_crit = false; bool report_crit; tmpth = command->cdw11 & 0xffff; tmpsel = (command->cdw11 >> 16) & 0xf; thsel = (command->cdw11 >> 20) & 0x3; DPRINTF("%s: tmpth=%#x tmpsel=%#x thsel=%#x", __func__, tmpth, tmpsel, thsel); /* Check for unsupported values */ if (((tmpsel != 0) && (tmpsel != 0xf)) || (thsel > NVME_TEMP_THRESH_UNDER)) { pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); return; } if (((thsel == NVME_TEMP_THRESH_OVER) && (NVME_TEMPERATURE >= tmpth)) || ((thsel == NVME_TEMP_THRESH_UNDER) && (NVME_TEMPERATURE <= tmpth))) set_crit = true; pthread_mutex_lock(&sc->mtx); if (set_crit) sc->health_log.critical_warning |= NVME_CRIT_WARN_ST_TEMPERATURE; else sc->health_log.critical_warning &= ~NVME_CRIT_WARN_ST_TEMPERATURE; pthread_mutex_unlock(&sc->mtx); report_crit = sc->feat[NVME_FEAT_ASYNC_EVENT_CONFIGURATION].cdw11 & NVME_CRIT_WARN_ST_TEMPERATURE; if (set_crit && report_crit) pci_nvme_aen_post(sc, PCI_NVME_AE_TYPE_SMART, sc->health_log.critical_warning); DPRINTF("%s: set_crit=%c critical_warning=%#x status=%#x", __func__, set_crit ? 'T':'F', sc->health_log.critical_warning, compl->status); } static void nvme_feature_num_queues(struct pci_nvme_softc *sc, struct nvme_feature_obj *feat __unused, struct nvme_command *command, struct nvme_completion *compl) { uint16_t nqr; /* Number of Queues Requested */ if (sc->num_q_is_set) { WPRINTF("%s: Number of Queues already set", __func__); pci_nvme_status_genc(&compl->status, NVME_SC_COMMAND_SEQUENCE_ERROR); return; } nqr = command->cdw11 & 0xFFFF; if (nqr == 0xffff) { WPRINTF("%s: Illegal NSQR value %#x", __func__, nqr); pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); return; } sc->num_squeues = ONE_BASED(nqr); if (sc->num_squeues > sc->max_queues) { DPRINTF("NSQR=%u is greater than max %u", sc->num_squeues, sc->max_queues); sc->num_squeues = sc->max_queues; } nqr = (command->cdw11 >> 16) & 0xFFFF; if (nqr == 0xffff) { WPRINTF("%s: Illegal NCQR value %#x", __func__, nqr); pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); return; } sc->num_cqueues = ONE_BASED(nqr); if (sc->num_cqueues > sc->max_queues) { DPRINTF("NCQR=%u is greater than max %u", sc->num_cqueues, sc->max_queues); sc->num_cqueues = sc->max_queues; } /* Patch the command value which will be saved on callback's return */ command->cdw11 = NVME_FEATURE_NUM_QUEUES(sc); compl->cdw0 = NVME_FEATURE_NUM_QUEUES(sc); sc->num_q_is_set = true; } static int nvme_opc_set_features(struct pci_nvme_softc *sc, struct nvme_command *command, struct nvme_completion *compl) { struct nvme_feature_obj *feat; uint32_t nsid = command->nsid; uint8_t fid = NVMEV(NVME_FEAT_SET_FID, command->cdw10); bool sv = NVMEV(NVME_FEAT_SET_SV, command->cdw10); DPRINTF("%s: Feature ID 0x%x (%s)", __func__, fid, nvme_fid_to_name(fid)); if (fid >= NVME_FID_MAX) { DPRINTF("%s invalid feature 0x%x", __func__, fid); pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); return (1); } if (sv) { pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_FEATURE_NOT_SAVEABLE); return (1); } feat = &sc->feat[fid]; if (feat->namespace_specific && (nsid == NVME_GLOBAL_NAMESPACE_TAG)) { pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); return (1); } if (!feat->namespace_specific && !((nsid == 0) || (nsid == NVME_GLOBAL_NAMESPACE_TAG))) { pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_FEATURE_NOT_NS_SPECIFIC); return (1); } compl->cdw0 = 0; pci_nvme_status_genc(&compl->status, NVME_SC_SUCCESS); if (feat->set) feat->set(sc, feat, command, compl); else { pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_FEATURE_NOT_CHANGEABLE); return (1); } DPRINTF("%s: status=%#x cdw11=%#x", __func__, compl->status, command->cdw11); if (compl->status == NVME_SC_SUCCESS) { feat->cdw11 = command->cdw11; if ((fid == NVME_FEAT_ASYNC_EVENT_CONFIGURATION) && (command->cdw11 != 0)) pci_nvme_aen_notify(sc); } return (0); } #define NVME_FEATURES_SEL_SUPPORTED 0x3 #define NVME_FEATURES_NS_SPECIFIC (1 << 1) static int nvme_opc_get_features(struct pci_nvme_softc* sc, struct nvme_command* command, struct nvme_completion* compl) { struct nvme_feature_obj *feat; uint8_t fid = command->cdw10 & 0xFF; uint8_t sel = (command->cdw10 >> 8) & 0x7; DPRINTF("%s: Feature ID 0x%x (%s)", __func__, fid, nvme_fid_to_name(fid)); if (fid >= NVME_FID_MAX) { DPRINTF("%s invalid feature 0x%x", __func__, fid); pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); return (1); } compl->cdw0 = 0; pci_nvme_status_genc(&compl->status, NVME_SC_SUCCESS); feat = &sc->feat[fid]; if (feat->get) { feat->get(sc, feat, command, compl); } if (compl->status == NVME_SC_SUCCESS) { if ((sel == NVME_FEATURES_SEL_SUPPORTED) && feat->namespace_specific) compl->cdw0 = NVME_FEATURES_NS_SPECIFIC; else compl->cdw0 = feat->cdw11; } return (0); } static int nvme_opc_format_nvm(struct pci_nvme_softc* sc, struct nvme_command* command, struct nvme_completion* compl) { uint8_t ses, lbaf, pi; /* Only supports Secure Erase Setting - User Data Erase */ ses = (command->cdw10 >> 9) & 0x7; if (ses > 0x1) { pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); return (1); } /* Only supports a single LBA Format */ lbaf = command->cdw10 & 0xf; if (lbaf != 0) { pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_INVALID_FORMAT); return (1); } - /* Doesn't support Protection Infomation */ + /* Doesn't support Protection Information */ pi = (command->cdw10 >> 5) & 0x7; if (pi != 0) { pci_nvme_status_genc(&compl->status, NVME_SC_INVALID_FIELD); return (1); } if (sc->nvstore.type == NVME_STOR_RAM) { if (sc->nvstore.ctx) free(sc->nvstore.ctx); sc->nvstore.ctx = calloc(1, sc->nvstore.size); pci_nvme_status_genc(&compl->status, NVME_SC_SUCCESS); } else { struct pci_nvme_ioreq *req; int err; req = pci_nvme_get_ioreq(sc); if (req == NULL) { pci_nvme_status_genc(&compl->status, NVME_SC_INTERNAL_DEVICE_ERROR); WPRINTF("%s: unable to allocate IO req", __func__); return (1); } req->nvme_sq = &sc->submit_queues[0]; req->sqid = 0; req->opc = command->opc; req->cid = command->cid; req->nsid = command->nsid; req->io_req.br_offset = 0; req->io_req.br_resid = sc->nvstore.size; req->io_req.br_callback = pci_nvme_io_done; err = blockif_delete(sc->nvstore.ctx, &req->io_req); if (err) { pci_nvme_status_genc(&compl->status, NVME_SC_INTERNAL_DEVICE_ERROR); pci_nvme_release_ioreq(sc, req); } else compl->status = NVME_NO_STATUS; } return (1); } static int nvme_opc_abort(struct pci_nvme_softc *sc __unused, struct nvme_command *command, struct nvme_completion *compl) { DPRINTF("%s submission queue %u, command ID 0x%x", __func__, command->cdw10 & 0xFFFF, (command->cdw10 >> 16) & 0xFFFF); /* TODO: search for the command ID and abort it */ compl->cdw0 = 1; pci_nvme_status_genc(&compl->status, NVME_SC_SUCCESS); return (1); } static int nvme_opc_async_event_req(struct pci_nvme_softc* sc, struct nvme_command* command, struct nvme_completion* compl) { DPRINTF("%s async event request count=%u aerl=%u cid=%#x", __func__, sc->aer_count, sc->ctrldata.aerl, command->cid); /* Don't exceed the Async Event Request Limit (AERL). */ if (pci_nvme_aer_limit_reached(sc)) { pci_nvme_status_tc(&compl->status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED); return (1); } if (pci_nvme_aer_add(sc, command->cid)) { pci_nvme_status_tc(&compl->status, NVME_SCT_GENERIC, NVME_SC_INTERNAL_DEVICE_ERROR); return (1); } /* * Raise events when they happen based on the Set Features cmd. * These events happen async, so only set completion successful if * there is an event reflective of the request to get event. */ compl->status = NVME_NO_STATUS; pci_nvme_aen_notify(sc); return (0); } static void pci_nvme_handle_admin_cmd(struct pci_nvme_softc* sc, uint64_t value) { struct nvme_completion compl; struct nvme_command *cmd; struct nvme_submission_queue *sq; struct nvme_completion_queue *cq; uint16_t sqhead; DPRINTF("%s index %u", __func__, (uint32_t)value); sq = &sc->submit_queues[0]; cq = &sc->compl_queues[0]; pthread_mutex_lock(&sq->mtx); sqhead = sq->head; DPRINTF("sqhead %u, tail %u", sqhead, sq->tail); while (sqhead != atomic_load_acq_short(&sq->tail)) { cmd = &(sq->qbase)[sqhead]; compl.cdw0 = 0; compl.status = 0; switch (cmd->opc) { case NVME_OPC_DELETE_IO_SQ: DPRINTF("%s command DELETE_IO_SQ", __func__); nvme_opc_delete_io_sq(sc, cmd, &compl); break; case NVME_OPC_CREATE_IO_SQ: DPRINTF("%s command CREATE_IO_SQ", __func__); nvme_opc_create_io_sq(sc, cmd, &compl); break; case NVME_OPC_DELETE_IO_CQ: DPRINTF("%s command DELETE_IO_CQ", __func__); nvme_opc_delete_io_cq(sc, cmd, &compl); break; case NVME_OPC_CREATE_IO_CQ: DPRINTF("%s command CREATE_IO_CQ", __func__); nvme_opc_create_io_cq(sc, cmd, &compl); break; case NVME_OPC_GET_LOG_PAGE: DPRINTF("%s command GET_LOG_PAGE", __func__); nvme_opc_get_log_page(sc, cmd, &compl); break; case NVME_OPC_IDENTIFY: DPRINTF("%s command IDENTIFY", __func__); nvme_opc_identify(sc, cmd, &compl); break; case NVME_OPC_ABORT: DPRINTF("%s command ABORT", __func__); nvme_opc_abort(sc, cmd, &compl); break; case NVME_OPC_SET_FEATURES: DPRINTF("%s command SET_FEATURES", __func__); nvme_opc_set_features(sc, cmd, &compl); break; case NVME_OPC_GET_FEATURES: DPRINTF("%s command GET_FEATURES", __func__); nvme_opc_get_features(sc, cmd, &compl); break; case NVME_OPC_FIRMWARE_ACTIVATE: DPRINTF("%s command FIRMWARE_ACTIVATE", __func__); pci_nvme_status_tc(&compl.status, NVME_SCT_COMMAND_SPECIFIC, NVME_SC_INVALID_FIRMWARE_SLOT); break; case NVME_OPC_ASYNC_EVENT_REQUEST: DPRINTF("%s command ASYNC_EVENT_REQ", __func__); nvme_opc_async_event_req(sc, cmd, &compl); break; case NVME_OPC_FORMAT_NVM: DPRINTF("%s command FORMAT_NVM", __func__); if ((sc->ctrldata.oacs & (1 << NVME_CTRLR_DATA_OACS_FORMAT_SHIFT)) == 0) { pci_nvme_status_genc(&compl.status, NVME_SC_INVALID_OPCODE); break; } nvme_opc_format_nvm(sc, cmd, &compl); break; case NVME_OPC_SECURITY_SEND: case NVME_OPC_SECURITY_RECEIVE: case NVME_OPC_SANITIZE: case NVME_OPC_GET_LBA_STATUS: DPRINTF("%s command OPC=%#x (unsupported)", __func__, cmd->opc); /* Valid but unsupported opcodes */ pci_nvme_status_genc(&compl.status, NVME_SC_INVALID_FIELD); break; default: DPRINTF("%s command OPC=%#X (not implemented)", __func__, cmd->opc); pci_nvme_status_genc(&compl.status, NVME_SC_INVALID_OPCODE); } sqhead = (sqhead + 1) % sq->size; if (NVME_COMPLETION_VALID(compl)) { pci_nvme_cq_update(sc, &sc->compl_queues[0], compl.cdw0, cmd->cid, 0, /* SQID */ compl.status); } } DPRINTF("setting sqhead %u", sqhead); sq->head = sqhead; if (cq->head != cq->tail) pci_generate_msix(sc->nsc_pi, 0); pthread_mutex_unlock(&sq->mtx); } /* * Update the Write and Read statistics reported in SMART data * * NVMe defines "data unit" as thousand's of 512 byte blocks and is rounded up. * E.g. 1 data unit is 1 - 1,000 512 byte blocks. 3 data units are 2,001 - 3,000 - * 512 byte blocks. Rounding up is acheived by initializing the remainder to 999. + * 512 byte blocks. Rounding up is achieved by initializing the remainder to 999. */ static void pci_nvme_stats_write_read_update(struct pci_nvme_softc *sc, uint8_t opc, size_t bytes, uint16_t status) { pthread_mutex_lock(&sc->mtx); switch (opc) { case NVME_OPC_WRITE: sc->write_commands++; if (status != NVME_SC_SUCCESS) break; sc->write_dunits_remainder += (bytes / 512); while (sc->write_dunits_remainder >= 1000) { sc->write_data_units++; sc->write_dunits_remainder -= 1000; } break; case NVME_OPC_READ: sc->read_commands++; if (status != NVME_SC_SUCCESS) break; sc->read_dunits_remainder += (bytes / 512); while (sc->read_dunits_remainder >= 1000) { sc->read_data_units++; sc->read_dunits_remainder -= 1000; } break; default: DPRINTF("%s: Invalid OPC 0x%02x for stats", __func__, opc); break; } pthread_mutex_unlock(&sc->mtx); } /* * Check if the combination of Starting LBA (slba) and number of blocks * exceeds the range of the underlying storage. * * Because NVMe specifies the SLBA in blocks as a uint64_t and blockif stores * the capacity in bytes as a uint64_t, care must be taken to avoid integer * overflow. */ static bool pci_nvme_out_of_range(struct pci_nvme_blockstore *nvstore, uint64_t slba, uint32_t nblocks) { size_t offset, bytes; /* Overflow check of multiplying Starting LBA by the sector size */ if (slba >> (64 - nvstore->sectsz_bits)) return (true); offset = slba << nvstore->sectsz_bits; bytes = nblocks << nvstore->sectsz_bits; /* Overflow check of Number of Logical Blocks */ if ((nvstore->size <= offset) || ((nvstore->size - offset) < bytes)) return (true); return (false); } static int pci_nvme_append_iov_req(struct pci_nvme_softc *sc __unused, struct pci_nvme_ioreq *req, uint64_t gpaddr, size_t size, uint64_t offset) { int iovidx; bool range_is_contiguous; if (req == NULL) return (-1); if (req->io_req.br_iovcnt == NVME_MAX_IOVEC) { return (-1); } /* * Minimize the number of IOVs by concatenating contiguous address * ranges. If the IOV count is zero, there is no previous range to * concatenate. */ if (req->io_req.br_iovcnt == 0) range_is_contiguous = false; else range_is_contiguous = (req->prev_gpaddr + req->prev_size) == gpaddr; if (range_is_contiguous) { iovidx = req->io_req.br_iovcnt - 1; req->io_req.br_iov[iovidx].iov_base = paddr_guest2host(req->sc->nsc_pi->pi_vmctx, req->prev_gpaddr, size); if (req->io_req.br_iov[iovidx].iov_base == NULL) return (-1); req->prev_size += size; req->io_req.br_resid += size; req->io_req.br_iov[iovidx].iov_len = req->prev_size; } else { iovidx = req->io_req.br_iovcnt; if (iovidx == 0) { req->io_req.br_offset = offset; req->io_req.br_resid = 0; req->io_req.br_param = req; } req->io_req.br_iov[iovidx].iov_base = paddr_guest2host(req->sc->nsc_pi->pi_vmctx, gpaddr, size); if (req->io_req.br_iov[iovidx].iov_base == NULL) return (-1); req->io_req.br_iov[iovidx].iov_len = size; req->prev_gpaddr = gpaddr; req->prev_size = size; req->io_req.br_resid += size; req->io_req.br_iovcnt++; } return (0); } static void pci_nvme_set_completion(struct pci_nvme_softc *sc, struct nvme_submission_queue *sq, int sqid, uint16_t cid, uint16_t status) { struct nvme_completion_queue *cq = &sc->compl_queues[sq->cqid]; DPRINTF("%s sqid %d cqid %u cid %u status: 0x%x 0x%x", __func__, sqid, sq->cqid, cid, NVME_STATUS_GET_SCT(status), NVME_STATUS_GET_SC(status)); pci_nvme_cq_update(sc, cq, 0, cid, sqid, status); if (cq->head != cq->tail) { if (cq->intr_en & NVME_CQ_INTEN) { pci_generate_msix(sc->nsc_pi, cq->intr_vec); } else { DPRINTF("%s: CQ%u interrupt disabled", __func__, sq->cqid); } } } static void pci_nvme_release_ioreq(struct pci_nvme_softc *sc, struct pci_nvme_ioreq *req) { req->sc = NULL; req->nvme_sq = NULL; req->sqid = 0; pthread_mutex_lock(&sc->mtx); STAILQ_INSERT_TAIL(&sc->ioreqs_free, req, link); sc->pending_ios--; /* when no more IO pending, can set to ready if device reset/enabled */ if (sc->pending_ios == 0 && NVME_CC_GET_EN(sc->regs.cc) && !(NVME_CSTS_GET_RDY(sc->regs.csts))) sc->regs.csts |= NVME_CSTS_RDY; pthread_mutex_unlock(&sc->mtx); sem_post(&sc->iosemlock); } static struct pci_nvme_ioreq * pci_nvme_get_ioreq(struct pci_nvme_softc *sc) { struct pci_nvme_ioreq *req = NULL; sem_wait(&sc->iosemlock); pthread_mutex_lock(&sc->mtx); req = STAILQ_FIRST(&sc->ioreqs_free); assert(req != NULL); STAILQ_REMOVE_HEAD(&sc->ioreqs_free, link); req->sc = sc; sc->pending_ios++; pthread_mutex_unlock(&sc->mtx); req->io_req.br_iovcnt = 0; req->io_req.br_offset = 0; req->io_req.br_resid = 0; req->io_req.br_param = req; req->prev_gpaddr = 0; req->prev_size = 0; return req; } static void pci_nvme_io_done(struct blockif_req *br, int err) { struct pci_nvme_ioreq *req = br->br_param; struct nvme_submission_queue *sq = req->nvme_sq; uint16_t code, status; DPRINTF("%s error %d %s", __func__, err, strerror(err)); /* TODO return correct error */ code = err ? NVME_SC_DATA_TRANSFER_ERROR : NVME_SC_SUCCESS; status = 0; pci_nvme_status_genc(&status, code); pci_nvme_set_completion(req->sc, sq, req->sqid, req->cid, status); pci_nvme_stats_write_read_update(req->sc, req->opc, req->bytes, status); pci_nvme_release_ioreq(req->sc, req); } /* * Implements the Flush command. The specification states: * If a volatile write cache is not present, Flush commands complete * successfully and have no effect * in the description of the Volatile Write Cache (VWC) field of the Identify * Controller data. Therefore, set status to Success if the command is * not supported (i.e. RAM or as indicated by the blockif). */ static bool nvme_opc_flush(struct pci_nvme_softc *sc __unused, struct nvme_command *cmd __unused, struct pci_nvme_blockstore *nvstore, struct pci_nvme_ioreq *req, uint16_t *status) { bool pending = false; if (nvstore->type == NVME_STOR_RAM) { pci_nvme_status_genc(status, NVME_SC_SUCCESS); } else { int err; req->io_req.br_callback = pci_nvme_io_done; err = blockif_flush(nvstore->ctx, &req->io_req); switch (err) { case 0: pending = true; break; case EOPNOTSUPP: pci_nvme_status_genc(status, NVME_SC_SUCCESS); break; default: pci_nvme_status_genc(status, NVME_SC_INTERNAL_DEVICE_ERROR); } } return (pending); } static uint16_t nvme_write_read_ram(struct pci_nvme_softc *sc, struct pci_nvme_blockstore *nvstore, uint64_t prp1, uint64_t prp2, size_t offset, uint64_t bytes, bool is_write) { uint8_t *buf = nvstore->ctx; enum nvme_copy_dir dir; uint16_t status; if (is_write) dir = NVME_COPY_TO_PRP; else dir = NVME_COPY_FROM_PRP; status = 0; if (nvme_prp_memcpy(sc->nsc_pi->pi_vmctx, prp1, prp2, buf + offset, bytes, dir)) pci_nvme_status_genc(&status, NVME_SC_DATA_TRANSFER_ERROR); else pci_nvme_status_genc(&status, NVME_SC_SUCCESS); return (status); } static uint16_t nvme_write_read_blockif(struct pci_nvme_softc *sc, struct pci_nvme_blockstore *nvstore, struct pci_nvme_ioreq *req, uint64_t prp1, uint64_t prp2, size_t offset, uint64_t bytes, bool is_write) { uint64_t size; int err; uint16_t status = NVME_NO_STATUS; size = MIN(PAGE_SIZE - (prp1 % PAGE_SIZE), bytes); if (pci_nvme_append_iov_req(sc, req, prp1, size, offset)) { err = -1; goto out; } offset += size; bytes -= size; if (bytes == 0) { ; } else if (bytes <= PAGE_SIZE) { size = bytes; if (pci_nvme_append_iov_req(sc, req, prp2, size, offset)) { err = -1; goto out; } } else { void *vmctx = sc->nsc_pi->pi_vmctx; uint64_t *prp_list = &prp2; uint64_t *last = prp_list; /* PRP2 is pointer to a physical region page list */ while (bytes) { /* Last entry in list points to the next list */ if ((prp_list == last) && (bytes > PAGE_SIZE)) { uint64_t prp = *prp_list; prp_list = paddr_guest2host(vmctx, prp, PAGE_SIZE - (prp % PAGE_SIZE)); if (prp_list == NULL) { err = -1; goto out; } last = prp_list + (NVME_PRP2_ITEMS - 1); } size = MIN(bytes, PAGE_SIZE); if (pci_nvme_append_iov_req(sc, req, *prp_list, size, offset)) { err = -1; goto out; } offset += size; bytes -= size; prp_list++; } } req->io_req.br_callback = pci_nvme_io_done; if (is_write) err = blockif_write(nvstore->ctx, &req->io_req); else err = blockif_read(nvstore->ctx, &req->io_req); out: if (err) pci_nvme_status_genc(&status, NVME_SC_DATA_TRANSFER_ERROR); return (status); } static bool nvme_opc_write_read(struct pci_nvme_softc *sc, struct nvme_command *cmd, struct pci_nvme_blockstore *nvstore, struct pci_nvme_ioreq *req, uint16_t *status) { uint64_t lba, nblocks, bytes; size_t offset; bool is_write = cmd->opc == NVME_OPC_WRITE; bool pending = false; lba = ((uint64_t)cmd->cdw11 << 32) | cmd->cdw10; nblocks = (cmd->cdw12 & 0xFFFF) + 1; bytes = nblocks << nvstore->sectsz_bits; if (bytes > NVME_MAX_DATA_SIZE) { WPRINTF("%s command would exceed MDTS", __func__); pci_nvme_status_genc(status, NVME_SC_INVALID_FIELD); goto out; } if (pci_nvme_out_of_range(nvstore, lba, nblocks)) { WPRINTF("%s command would exceed LBA range(slba=%#lx nblocks=%#lx)", __func__, lba, nblocks); pci_nvme_status_genc(status, NVME_SC_LBA_OUT_OF_RANGE); goto out; } offset = lba << nvstore->sectsz_bits; req->bytes = bytes; req->io_req.br_offset = lba; /* PRP bits 1:0 must be zero */ cmd->prp1 &= ~0x3UL; cmd->prp2 &= ~0x3UL; if (nvstore->type == NVME_STOR_RAM) { *status = nvme_write_read_ram(sc, nvstore, cmd->prp1, cmd->prp2, offset, bytes, is_write); } else { *status = nvme_write_read_blockif(sc, nvstore, req, cmd->prp1, cmd->prp2, offset, bytes, is_write); if (*status == NVME_NO_STATUS) pending = true; } out: if (!pending) pci_nvme_stats_write_read_update(sc, cmd->opc, bytes, *status); return (pending); } static void pci_nvme_dealloc_sm(struct blockif_req *br, int err) { struct pci_nvme_ioreq *req = br->br_param; struct pci_nvme_softc *sc = req->sc; bool done = true; uint16_t status; status = 0; if (err) { pci_nvme_status_genc(&status, NVME_SC_INTERNAL_DEVICE_ERROR); } else if ((req->prev_gpaddr + 1) == (req->prev_size)) { pci_nvme_status_genc(&status, NVME_SC_SUCCESS); } else { struct iovec *iov = req->io_req.br_iov; req->prev_gpaddr++; iov += req->prev_gpaddr; /* The iov_* values already include the sector size */ req->io_req.br_offset = (off_t)iov->iov_base; req->io_req.br_resid = iov->iov_len; if (blockif_delete(sc->nvstore.ctx, &req->io_req)) { pci_nvme_status_genc(&status, NVME_SC_INTERNAL_DEVICE_ERROR); } else done = false; } if (done) { pci_nvme_set_completion(sc, req->nvme_sq, req->sqid, req->cid, status); pci_nvme_release_ioreq(sc, req); } } static bool nvme_opc_dataset_mgmt(struct pci_nvme_softc *sc, struct nvme_command *cmd, struct pci_nvme_blockstore *nvstore, struct pci_nvme_ioreq *req, uint16_t *status) { struct nvme_dsm_range *range = NULL; uint32_t nr, r, non_zero, dr; int err; bool pending = false; if ((sc->ctrldata.oncs & NVME_ONCS_DSM) == 0) { pci_nvme_status_genc(status, NVME_SC_INVALID_OPCODE); goto out; } nr = cmd->cdw10 & 0xff; /* copy locally because a range entry could straddle PRPs */ range = calloc(1, NVME_MAX_DSM_TRIM); if (range == NULL) { pci_nvme_status_genc(status, NVME_SC_INTERNAL_DEVICE_ERROR); goto out; } nvme_prp_memcpy(sc->nsc_pi->pi_vmctx, cmd->prp1, cmd->prp2, (uint8_t *)range, NVME_MAX_DSM_TRIM, NVME_COPY_FROM_PRP); /* Check for invalid ranges and the number of non-zero lengths */ non_zero = 0; for (r = 0; r <= nr; r++) { if (pci_nvme_out_of_range(nvstore, range[r].starting_lba, range[r].length)) { pci_nvme_status_genc(status, NVME_SC_LBA_OUT_OF_RANGE); goto out; } if (range[r].length != 0) non_zero++; } if (cmd->cdw11 & NVME_DSM_ATTR_DEALLOCATE) { size_t offset, bytes; int sectsz_bits = sc->nvstore.sectsz_bits; /* * DSM calls are advisory only, and compliant controllers * may choose to take no actions (i.e. return Success). */ if (!nvstore->deallocate) { pci_nvme_status_genc(status, NVME_SC_SUCCESS); goto out; } /* If all ranges have a zero length, return Success */ if (non_zero == 0) { pci_nvme_status_genc(status, NVME_SC_SUCCESS); goto out; } if (req == NULL) { pci_nvme_status_genc(status, NVME_SC_INTERNAL_DEVICE_ERROR); goto out; } offset = range[0].starting_lba << sectsz_bits; bytes = range[0].length << sectsz_bits; /* * If the request is for more than a single range, store * the ranges in the br_iov. Optimize for the common case * of a single range. * * Note that NVMe Number of Ranges is a zero based value */ req->io_req.br_iovcnt = 0; req->io_req.br_offset = offset; req->io_req.br_resid = bytes; if (nr == 0) { req->io_req.br_callback = pci_nvme_io_done; } else { struct iovec *iov = req->io_req.br_iov; for (r = 0, dr = 0; r <= nr; r++) { offset = range[r].starting_lba << sectsz_bits; bytes = range[r].length << sectsz_bits; if (bytes == 0) continue; if ((nvstore->size - offset) < bytes) { pci_nvme_status_genc(status, NVME_SC_LBA_OUT_OF_RANGE); goto out; } iov[dr].iov_base = (void *)offset; iov[dr].iov_len = bytes; dr++; } req->io_req.br_callback = pci_nvme_dealloc_sm; /* * Use prev_gpaddr to track the current entry and * prev_size to track the number of entries */ req->prev_gpaddr = 0; req->prev_size = dr; } err = blockif_delete(nvstore->ctx, &req->io_req); if (err) pci_nvme_status_genc(status, NVME_SC_INTERNAL_DEVICE_ERROR); else pending = true; } out: free(range); return (pending); } static void pci_nvme_handle_io_cmd(struct pci_nvme_softc* sc, uint16_t idx) { struct nvme_submission_queue *sq; uint16_t status; uint16_t sqhead; /* handle all submissions up to sq->tail index */ sq = &sc->submit_queues[idx]; pthread_mutex_lock(&sq->mtx); sqhead = sq->head; DPRINTF("nvme_handle_io qid %u head %u tail %u cmdlist %p", idx, sqhead, sq->tail, sq->qbase); while (sqhead != atomic_load_acq_short(&sq->tail)) { struct nvme_command *cmd; struct pci_nvme_ioreq *req; uint32_t nsid; bool pending; pending = false; req = NULL; status = 0; cmd = &sq->qbase[sqhead]; sqhead = (sqhead + 1) % sq->size; nsid = le32toh(cmd->nsid); if ((nsid == 0) || (nsid > sc->ctrldata.nn)) { pci_nvme_status_genc(&status, NVME_SC_INVALID_NAMESPACE_OR_FORMAT); status |= NVME_STATUS_DNR_MASK << NVME_STATUS_DNR_SHIFT; goto complete; } req = pci_nvme_get_ioreq(sc); if (req == NULL) { pci_nvme_status_genc(&status, NVME_SC_INTERNAL_DEVICE_ERROR); WPRINTF("%s: unable to allocate IO req", __func__); goto complete; } req->nvme_sq = sq; req->sqid = idx; req->opc = cmd->opc; req->cid = cmd->cid; req->nsid = cmd->nsid; switch (cmd->opc) { case NVME_OPC_FLUSH: pending = nvme_opc_flush(sc, cmd, &sc->nvstore, req, &status); break; case NVME_OPC_WRITE: case NVME_OPC_READ: pending = nvme_opc_write_read(sc, cmd, &sc->nvstore, req, &status); break; case NVME_OPC_WRITE_ZEROES: /* TODO: write zeroes WPRINTF("%s write zeroes lba 0x%lx blocks %u", __func__, lba, cmd->cdw12 & 0xFFFF); */ pci_nvme_status_genc(&status, NVME_SC_SUCCESS); break; case NVME_OPC_DATASET_MANAGEMENT: pending = nvme_opc_dataset_mgmt(sc, cmd, &sc->nvstore, req, &status); break; default: WPRINTF("%s unhandled io command 0x%x", __func__, cmd->opc); pci_nvme_status_genc(&status, NVME_SC_INVALID_OPCODE); } complete: if (!pending) { pci_nvme_set_completion(sc, sq, idx, cmd->cid, status); if (req != NULL) pci_nvme_release_ioreq(sc, req); } } sq->head = sqhead; pthread_mutex_unlock(&sq->mtx); } static void pci_nvme_handle_doorbell(struct pci_nvme_softc* sc, uint64_t idx, int is_sq, uint64_t value) { DPRINTF("nvme doorbell %lu, %s, val 0x%lx", idx, is_sq ? "SQ" : "CQ", value & 0xFFFF); if (is_sq) { if (idx > sc->num_squeues) { WPRINTF("%s queue index %lu overflow from " "guest (max %u)", __func__, idx, sc->num_squeues); return; } atomic_store_short(&sc->submit_queues[idx].tail, (uint16_t)value); if (idx == 0) { pci_nvme_handle_admin_cmd(sc, value); } else { /* submission queue; handle new entries in SQ */ if (idx > sc->num_squeues) { WPRINTF("%s SQ index %lu overflow from " "guest (max %u)", __func__, idx, sc->num_squeues); return; } pci_nvme_handle_io_cmd(sc, (uint16_t)idx); } } else { if (idx > sc->num_cqueues) { WPRINTF("%s queue index %lu overflow from " "guest (max %u)", __func__, idx, sc->num_cqueues); return; } atomic_store_short(&sc->compl_queues[idx].head, (uint16_t)value); } } static void pci_nvme_bar0_reg_dumps(const char *func, uint64_t offset, int iswrite) { const char *s = iswrite ? "WRITE" : "READ"; switch (offset) { case NVME_CR_CAP_LOW: DPRINTF("%s %s NVME_CR_CAP_LOW", func, s); break; case NVME_CR_CAP_HI: DPRINTF("%s %s NVME_CR_CAP_HI", func, s); break; case NVME_CR_VS: DPRINTF("%s %s NVME_CR_VS", func, s); break; case NVME_CR_INTMS: DPRINTF("%s %s NVME_CR_INTMS", func, s); break; case NVME_CR_INTMC: DPRINTF("%s %s NVME_CR_INTMC", func, s); break; case NVME_CR_CC: DPRINTF("%s %s NVME_CR_CC", func, s); break; case NVME_CR_CSTS: DPRINTF("%s %s NVME_CR_CSTS", func, s); break; case NVME_CR_NSSR: DPRINTF("%s %s NVME_CR_NSSR", func, s); break; case NVME_CR_AQA: DPRINTF("%s %s NVME_CR_AQA", func, s); break; case NVME_CR_ASQ_LOW: DPRINTF("%s %s NVME_CR_ASQ_LOW", func, s); break; case NVME_CR_ASQ_HI: DPRINTF("%s %s NVME_CR_ASQ_HI", func, s); break; case NVME_CR_ACQ_LOW: DPRINTF("%s %s NVME_CR_ACQ_LOW", func, s); break; case NVME_CR_ACQ_HI: DPRINTF("%s %s NVME_CR_ACQ_HI", func, s); break; default: DPRINTF("unknown nvme bar-0 offset 0x%lx", offset); } } static void pci_nvme_write_bar_0(struct pci_nvme_softc *sc, uint64_t offset, int size, uint64_t value) { uint32_t ccreg; if (offset >= NVME_DOORBELL_OFFSET) { uint64_t belloffset = offset - NVME_DOORBELL_OFFSET; uint64_t idx = belloffset / 8; /* door bell size = 2*int */ int is_sq = (belloffset % 8) < 4; if ((sc->regs.csts & NVME_CSTS_RDY) == 0) { WPRINTF("doorbell write prior to RDY (offset=%#lx)\n", offset); return; } if (belloffset > ((sc->max_queues+1) * 8 - 4)) { WPRINTF("guest attempted an overflow write offset " "0x%lx, val 0x%lx in %s", offset, value, __func__); return; } if (is_sq) { if (sc->submit_queues[idx].qbase == NULL) return; } else if (sc->compl_queues[idx].qbase == NULL) return; pci_nvme_handle_doorbell(sc, idx, is_sq, value); return; } DPRINTF("nvme-write offset 0x%lx, size %d, value 0x%lx", offset, size, value); if (size != 4) { WPRINTF("guest wrote invalid size %d (offset 0x%lx, " "val 0x%lx) to bar0 in %s", size, offset, value, __func__); /* TODO: shutdown device */ return; } pci_nvme_bar0_reg_dumps(__func__, offset, 1); pthread_mutex_lock(&sc->mtx); switch (offset) { case NVME_CR_CAP_LOW: case NVME_CR_CAP_HI: /* readonly */ break; case NVME_CR_VS: /* readonly */ break; case NVME_CR_INTMS: /* MSI-X, so ignore */ break; case NVME_CR_INTMC: /* MSI-X, so ignore */ break; case NVME_CR_CC: ccreg = (uint32_t)value; DPRINTF("%s NVME_CR_CC en %x css %x shn %x iosqes %u " "iocqes %u", __func__, NVME_CC_GET_EN(ccreg), NVME_CC_GET_CSS(ccreg), NVME_CC_GET_SHN(ccreg), NVME_CC_GET_IOSQES(ccreg), NVME_CC_GET_IOCQES(ccreg)); if (NVME_CC_GET_SHN(ccreg)) { /* perform shutdown - flush out data to backend */ sc->regs.csts &= ~(NVME_CSTS_REG_SHST_MASK << NVME_CSTS_REG_SHST_SHIFT); sc->regs.csts |= NVME_SHST_COMPLETE << NVME_CSTS_REG_SHST_SHIFT; } if (NVME_CC_GET_EN(ccreg) != NVME_CC_GET_EN(sc->regs.cc)) { if (NVME_CC_GET_EN(ccreg) == 0) /* transition 1-> causes controller reset */ pci_nvme_reset_locked(sc); else pci_nvme_init_controller(sc); } /* Insert the iocqes, iosqes and en bits from the write */ sc->regs.cc &= ~NVME_CC_WRITE_MASK; sc->regs.cc |= ccreg & NVME_CC_WRITE_MASK; if (NVME_CC_GET_EN(ccreg) == 0) { /* Insert the ams, mps and css bit fields */ sc->regs.cc &= ~NVME_CC_NEN_WRITE_MASK; sc->regs.cc |= ccreg & NVME_CC_NEN_WRITE_MASK; sc->regs.csts &= ~NVME_CSTS_RDY; } else if ((sc->pending_ios == 0) && !(sc->regs.csts & NVME_CSTS_CFS)) { sc->regs.csts |= NVME_CSTS_RDY; } break; case NVME_CR_CSTS: break; case NVME_CR_NSSR: /* ignore writes; don't support subsystem reset */ break; case NVME_CR_AQA: sc->regs.aqa = (uint32_t)value; break; case NVME_CR_ASQ_LOW: sc->regs.asq = (sc->regs.asq & (0xFFFFFFFF00000000)) | (0xFFFFF000 & value); break; case NVME_CR_ASQ_HI: sc->regs.asq = (sc->regs.asq & (0x00000000FFFFFFFF)) | (value << 32); break; case NVME_CR_ACQ_LOW: sc->regs.acq = (sc->regs.acq & (0xFFFFFFFF00000000)) | (0xFFFFF000 & value); break; case NVME_CR_ACQ_HI: sc->regs.acq = (sc->regs.acq & (0x00000000FFFFFFFF)) | (value << 32); break; default: DPRINTF("%s unknown offset 0x%lx, value 0x%lx size %d", __func__, offset, value, size); } pthread_mutex_unlock(&sc->mtx); } static void pci_nvme_write(struct pci_devinst *pi, int baridx, uint64_t offset, int size, uint64_t value) { struct pci_nvme_softc* sc = pi->pi_arg; if (baridx == pci_msix_table_bar(pi) || baridx == pci_msix_pba_bar(pi)) { DPRINTF("nvme-write baridx %d, msix: off 0x%lx, size %d, " " value 0x%lx", baridx, offset, size, value); pci_emul_msix_twrite(pi, offset, size, value); return; } switch (baridx) { case 0: pci_nvme_write_bar_0(sc, offset, size, value); break; default: DPRINTF("%s unknown baridx %d, val 0x%lx", __func__, baridx, value); } } static uint64_t pci_nvme_read_bar_0(struct pci_nvme_softc* sc, uint64_t offset, int size) { uint64_t value; pci_nvme_bar0_reg_dumps(__func__, offset, 0); if (offset < NVME_DOORBELL_OFFSET) { void *p = &(sc->regs); pthread_mutex_lock(&sc->mtx); memcpy(&value, (void *)((uintptr_t)p + offset), size); pthread_mutex_unlock(&sc->mtx); } else { value = 0; WPRINTF("pci_nvme: read invalid offset %ld", offset); } switch (size) { case 1: value &= 0xFF; break; case 2: value &= 0xFFFF; break; case 4: value &= 0xFFFFFFFF; break; } DPRINTF(" nvme-read offset 0x%lx, size %d -> value 0x%x", offset, size, (uint32_t)value); return (value); } static uint64_t pci_nvme_read(struct pci_devinst *pi, int baridx, uint64_t offset, int size) { struct pci_nvme_softc* sc = pi->pi_arg; if (baridx == pci_msix_table_bar(pi) || baridx == pci_msix_pba_bar(pi)) { DPRINTF("nvme-read bar: %d, msix: regoff 0x%lx, size %d", baridx, offset, size); return pci_emul_msix_tread(pi, offset, size); } switch (baridx) { case 0: return pci_nvme_read_bar_0(sc, offset, size); default: DPRINTF("unknown bar %d, 0x%lx", baridx, offset); } return (0); } static int pci_nvme_parse_config(struct pci_nvme_softc *sc, nvlist_t *nvl) { char bident[sizeof("XXX:XXX")]; const char *value; uint32_t sectsz; sc->max_queues = NVME_QUEUES; sc->max_qentries = NVME_MAX_QENTRIES; sc->ioslots = NVME_IOSLOTS; sc->num_squeues = sc->max_queues; sc->num_cqueues = sc->max_queues; sc->dataset_management = NVME_DATASET_MANAGEMENT_AUTO; sectsz = 0; snprintf(sc->ctrldata.sn, sizeof(sc->ctrldata.sn), "NVME-%d-%d", sc->nsc_pi->pi_slot, sc->nsc_pi->pi_func); value = get_config_value_node(nvl, "maxq"); if (value != NULL) sc->max_queues = atoi(value); value = get_config_value_node(nvl, "qsz"); if (value != NULL) { sc->max_qentries = atoi(value); if (sc->max_qentries <= 0) { EPRINTLN("nvme: Invalid qsz option %d", sc->max_qentries); return (-1); } } value = get_config_value_node(nvl, "ioslots"); if (value != NULL) { sc->ioslots = atoi(value); if (sc->ioslots <= 0) { EPRINTLN("Invalid ioslots option %d", sc->ioslots); return (-1); } } value = get_config_value_node(nvl, "sectsz"); if (value != NULL) sectsz = atoi(value); value = get_config_value_node(nvl, "ser"); if (value != NULL) { /* * This field indicates the Product Serial Number in * 7-bit ASCII, unused bytes should be space characters. * Ref: NVMe v1.3c. */ cpywithpad((char *)sc->ctrldata.sn, sizeof(sc->ctrldata.sn), value, ' '); } value = get_config_value_node(nvl, "eui64"); if (value != NULL) sc->nvstore.eui64 = htobe64(strtoull(value, NULL, 0)); value = get_config_value_node(nvl, "dsm"); if (value != NULL) { if (strcmp(value, "auto") == 0) sc->dataset_management = NVME_DATASET_MANAGEMENT_AUTO; else if (strcmp(value, "enable") == 0) sc->dataset_management = NVME_DATASET_MANAGEMENT_ENABLE; else if (strcmp(value, "disable") == 0) sc->dataset_management = NVME_DATASET_MANAGEMENT_DISABLE; } value = get_config_value_node(nvl, "ram"); if (value != NULL) { uint64_t sz = strtoull(value, NULL, 10); sc->nvstore.type = NVME_STOR_RAM; sc->nvstore.size = sz * 1024 * 1024; sc->nvstore.ctx = calloc(1, sc->nvstore.size); sc->nvstore.sectsz = 4096; sc->nvstore.sectsz_bits = 12; if (sc->nvstore.ctx == NULL) { EPRINTLN("nvme: Unable to allocate RAM"); return (-1); } } else { snprintf(bident, sizeof(bident), "%u:%u", sc->nsc_pi->pi_slot, sc->nsc_pi->pi_func); sc->nvstore.ctx = blockif_open(nvl, bident); if (sc->nvstore.ctx == NULL) { EPRINTLN("nvme: Could not open backing file: %s", strerror(errno)); return (-1); } sc->nvstore.type = NVME_STOR_BLOCKIF; sc->nvstore.size = blockif_size(sc->nvstore.ctx); } if (sectsz == 512 || sectsz == 4096 || sectsz == 8192) sc->nvstore.sectsz = sectsz; else if (sc->nvstore.type != NVME_STOR_RAM) sc->nvstore.sectsz = blockif_sectsz(sc->nvstore.ctx); for (sc->nvstore.sectsz_bits = 9; (1U << sc->nvstore.sectsz_bits) < sc->nvstore.sectsz; sc->nvstore.sectsz_bits++); if (sc->max_queues <= 0 || sc->max_queues > NVME_QUEUES) sc->max_queues = NVME_QUEUES; return (0); } static void pci_nvme_resized(struct blockif_ctxt *bctxt __unused, void *arg, size_t new_size) { struct pci_nvme_softc *sc; struct pci_nvme_blockstore *nvstore; struct nvme_namespace_data *nd; sc = arg; nvstore = &sc->nvstore; nd = &sc->nsdata; nvstore->size = new_size; pci_nvme_init_nsdata_size(nvstore, nd); /* Add changed NSID to list */ sc->ns_log.ns[0] = 1; sc->ns_log.ns[1] = 0; pci_nvme_aen_post(sc, PCI_NVME_AE_TYPE_NOTICE, PCI_NVME_AEI_NOTICE_NS_ATTR_CHANGED); } static int pci_nvme_init(struct pci_devinst *pi, nvlist_t *nvl) { struct pci_nvme_softc *sc; uint32_t pci_membar_sz; int error; error = 0; sc = calloc(1, sizeof(struct pci_nvme_softc)); pi->pi_arg = sc; sc->nsc_pi = pi; error = pci_nvme_parse_config(sc, nvl); if (error < 0) goto done; else error = 0; STAILQ_INIT(&sc->ioreqs_free); sc->ioreqs = calloc(sc->ioslots, sizeof(struct pci_nvme_ioreq)); for (uint32_t i = 0; i < sc->ioslots; i++) { STAILQ_INSERT_TAIL(&sc->ioreqs_free, &sc->ioreqs[i], link); } pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0A0A); pci_set_cfgdata16(pi, PCIR_VENDOR, 0xFB5D); pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE); pci_set_cfgdata8(pi, PCIR_SUBCLASS, PCIS_STORAGE_NVM); pci_set_cfgdata8(pi, PCIR_PROGIF, PCIP_STORAGE_NVM_ENTERPRISE_NVMHCI_1_0); /* * Allocate size of NVMe registers + doorbell space for all queues. * * The specification requires a minimum memory I/O window size of 16K. * The Windows driver will refuse to start a device with a smaller * window. */ pci_membar_sz = sizeof(struct nvme_registers) + 2 * sizeof(uint32_t) * (sc->max_queues + 1); pci_membar_sz = MAX(pci_membar_sz, NVME_MMIO_SPACE_MIN); DPRINTF("nvme membar size: %u", pci_membar_sz); error = pci_emul_alloc_bar(pi, 0, PCIBAR_MEM64, pci_membar_sz); if (error) { WPRINTF("%s pci alloc mem bar failed", __func__); goto done; } error = pci_emul_add_msixcap(pi, sc->max_queues + 1, NVME_MSIX_BAR); if (error) { WPRINTF("%s pci add msixcap failed", __func__); goto done; } error = pci_emul_add_pciecap(pi, PCIEM_TYPE_ROOT_INT_EP); if (error) { WPRINTF("%s pci add Express capability failed", __func__); goto done; } pthread_mutex_init(&sc->mtx, NULL); sem_init(&sc->iosemlock, 0, sc->ioslots); blockif_register_resize_callback(sc->nvstore.ctx, pci_nvme_resized, sc); pci_nvme_init_queues(sc, sc->max_queues, sc->max_queues); /* * Controller data depends on Namespace data so initialize Namespace * data first. */ pci_nvme_init_nsdata(sc, &sc->nsdata, 1, &sc->nvstore); pci_nvme_init_ctrldata(sc); pci_nvme_init_logpages(sc); pci_nvme_init_features(sc); pci_nvme_aer_init(sc); pci_nvme_aen_init(sc); pci_nvme_reset(sc); pci_lintr_request(pi); done: return (error); } static int pci_nvme_legacy_config(nvlist_t *nvl, const char *opts) { char *cp, *ram; if (opts == NULL) return (0); if (strncmp(opts, "ram=", 4) == 0) { cp = strchr(opts, ','); if (cp == NULL) { set_config_value_node(nvl, "ram", opts + 4); return (0); } ram = strndup(opts + 4, cp - opts - 4); set_config_value_node(nvl, "ram", ram); free(ram); return (pci_parse_legacy_config(nvl, cp + 1)); } else return (blockif_legacy_config(nvl, opts)); } static const struct pci_devemu pci_de_nvme = { .pe_emu = "nvme", .pe_init = pci_nvme_init, .pe_legacy_config = pci_nvme_legacy_config, .pe_barwrite = pci_nvme_write, .pe_barread = pci_nvme_read }; PCI_EMUL_SET(pci_de_nvme); diff --git a/usr.sbin/bhyve/pci_virtio_net.c b/usr.sbin/bhyve/pci_virtio_net.c index 9af5854037fe..dd60c5a89523 100644 --- a/usr.sbin/bhyve/pci_virtio_net.c +++ b/usr.sbin/bhyve/pci_virtio_net.c @@ -1,820 +1,820 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include /* IFNAMSIZ */ #include #include #include #include #include #include #include #include #include #include #include #include #include "bhyverun.h" #include "config.h" #include "debug.h" #include "pci_emul.h" #include "mevent.h" #include "virtio.h" #include "net_utils.h" #include "net_backends.h" #include "iov.h" #define VTNET_RINGSZ 1024 #define VTNET_MAXSEGS 256 #define VTNET_MAX_PKT_LEN (65536 + 64) #define VTNET_MIN_MTU ETHERMIN #define VTNET_MAX_MTU 65535 #define VTNET_S_HOSTCAPS \ ( VIRTIO_NET_F_MAC | VIRTIO_NET_F_STATUS | \ VIRTIO_F_NOTIFY_ON_EMPTY | VIRTIO_RING_F_INDIRECT_DESC) /* * PCI config-space "registers" */ struct virtio_net_config { uint8_t mac[6]; uint16_t status; uint16_t max_virtqueue_pairs; uint16_t mtu; } __packed; /* * Queue definitions. */ #define VTNET_RXQ 0 #define VTNET_TXQ 1 #define VTNET_CTLQ 2 /* NB: not yet supported */ #define VTNET_MAXQ 3 /* * Debug printf */ static int pci_vtnet_debug; #define DPRINTF(params) if (pci_vtnet_debug) PRINTLN params #define WPRINTF(params) PRINTLN params /* * Per-device softc */ struct pci_vtnet_softc { struct virtio_softc vsc_vs; struct vqueue_info vsc_queues[VTNET_MAXQ - 1]; pthread_mutex_t vsc_mtx; net_backend_t *vsc_be; bool features_negotiated; /* protected by rx_mtx */ int resetting; /* protected by tx_mtx */ uint64_t vsc_features; /* negotiated features */ pthread_mutex_t rx_mtx; int rx_merge; /* merged rx bufs in use */ pthread_t tx_tid; pthread_mutex_t tx_mtx; pthread_cond_t tx_cond; int tx_in_progress; size_t vhdrlen; size_t be_vhdrlen; struct virtio_net_config vsc_config; struct virtio_consts vsc_consts; }; static void pci_vtnet_reset(void *); /* static void pci_vtnet_notify(void *, struct vqueue_info *); */ static int pci_vtnet_cfgread(void *, int, int, uint32_t *); static int pci_vtnet_cfgwrite(void *, int, int, uint32_t); static void pci_vtnet_neg_features(void *, uint64_t); #ifdef BHYVE_SNAPSHOT static void pci_vtnet_pause(void *); static void pci_vtnet_resume(void *); static int pci_vtnet_snapshot(void *, struct vm_snapshot_meta *); #endif static struct virtio_consts vtnet_vi_consts = { .vc_name = "vtnet", .vc_nvq = VTNET_MAXQ - 1, .vc_cfgsize = sizeof(struct virtio_net_config), .vc_reset = pci_vtnet_reset, .vc_cfgread = pci_vtnet_cfgread, .vc_cfgwrite = pci_vtnet_cfgwrite, .vc_apply_features = pci_vtnet_neg_features, .vc_hv_caps = VTNET_S_HOSTCAPS, #ifdef BHYVE_SNAPSHOT .vc_pause = pci_vtnet_pause, .vc_resume = pci_vtnet_resume, .vc_snapshot = pci_vtnet_snapshot, #endif }; static void pci_vtnet_reset(void *vsc) { struct pci_vtnet_softc *sc = vsc; DPRINTF(("vtnet: device reset requested !")); /* Acquire the RX lock to block RX processing. */ pthread_mutex_lock(&sc->rx_mtx); /* * Make sure receive operation is disabled at least until we * re-negotiate the features, since receive operation depends * on the value of sc->rx_merge and the header length, which * are both set in pci_vtnet_neg_features(). * Receive operation will be enabled again once the guest adds * the first receive buffers and kicks us. */ sc->features_negotiated = false; netbe_rx_disable(sc->vsc_be); /* Set sc->resetting and give a chance to the TX thread to stop. */ pthread_mutex_lock(&sc->tx_mtx); sc->resetting = 1; while (sc->tx_in_progress) { pthread_mutex_unlock(&sc->tx_mtx); usleep(10000); pthread_mutex_lock(&sc->tx_mtx); } /* * Now reset rings, MSI-X vectors, and negotiated capabilities. * Do that with the TX lock held, since we need to reset * sc->resetting. */ vi_reset_dev(&sc->vsc_vs); sc->resetting = 0; pthread_mutex_unlock(&sc->tx_mtx); pthread_mutex_unlock(&sc->rx_mtx); } static __inline struct iovec * iov_trim_hdr(struct iovec *iov, int *iovcnt, unsigned int hlen) { struct iovec *riov; if (iov[0].iov_len < hlen) { /* * Not enough header space in the first fragment. * That's not ok for us. */ return NULL; } iov[0].iov_len -= hlen; if (iov[0].iov_len == 0) { *iovcnt -= 1; if (*iovcnt == 0) { /* * Only space for the header. That's not * enough for us. */ return NULL; } riov = &iov[1]; } else { iov[0].iov_base = (void *)((uintptr_t)iov[0].iov_base + hlen); riov = &iov[0]; } return (riov); } struct virtio_mrg_rxbuf_info { uint16_t idx; uint16_t pad; uint32_t len; }; static void pci_vtnet_rx(struct pci_vtnet_softc *sc) { int prepend_hdr_len = sc->vhdrlen - sc->be_vhdrlen; struct virtio_mrg_rxbuf_info info[VTNET_MAXSEGS]; struct iovec iov[VTNET_MAXSEGS + 1]; struct vqueue_info *vq; struct vi_req req; vq = &sc->vsc_queues[VTNET_RXQ]; /* Features must be negotiated */ if (!sc->features_negotiated) { return; } for (;;) { struct virtio_net_rxhdr *hdr; uint32_t riov_bytes; struct iovec *riov; uint32_t ulen; int riov_len; int n_chains; ssize_t rlen; ssize_t plen; plen = netbe_peek_recvlen(sc->vsc_be); if (plen <= 0) { /* * No more packets (plen == 0), or backend errored * (plen < 0). Interrupt if needed and stop. */ vq_endchains(vq, /*used_all_avail=*/0); return; } plen += prepend_hdr_len; /* * Get a descriptor chain to store the next ingress * packet. In case of mergeable rx buffers, get as * many chains as necessary in order to make room * for plen bytes. */ riov_bytes = 0; riov_len = 0; riov = iov; n_chains = 0; do { int n = vq_getchain(vq, riov, VTNET_MAXSEGS - riov_len, &req); info[n_chains].idx = req.idx; if (n == 0) { /* * No rx buffers. Enable RX kicks and double * check. */ vq_kick_enable(vq); if (!vq_has_descs(vq)) { /* * Still no buffers. Return the unused * chains (if any), interrupt if needed * (including for NOTIFY_ON_EMPTY), and * disable the backend until the next * kick. */ vq_retchains(vq, n_chains); vq_endchains(vq, /*used_all_avail=*/1); netbe_rx_disable(sc->vsc_be); return; } /* More rx buffers found, so keep going. */ vq_kick_disable(vq); continue; } assert(n >= 1 && riov_len + n <= VTNET_MAXSEGS); riov_len += n; if (!sc->rx_merge) { n_chains = 1; break; } info[n_chains].len = (uint32_t)count_iov(riov, n); riov_bytes += info[n_chains].len; riov += n; n_chains++; } while (riov_bytes < plen && riov_len < VTNET_MAXSEGS); riov = iov; hdr = riov[0].iov_base; if (prepend_hdr_len > 0) { /* * The frontend uses a virtio-net header, but the * backend does not. We need to prepend a zeroed * header. */ riov = iov_trim_hdr(riov, &riov_len, prepend_hdr_len); if (riov == NULL) { /* * The first collected chain is nonsensical, * as it is not even enough to store the * virtio-net header. Just drop it. */ vq_relchain(vq, info[0].idx, 0); vq_retchains(vq, n_chains - 1); continue; } memset(hdr, 0, prepend_hdr_len); } rlen = netbe_recv(sc->vsc_be, riov, riov_len); if (rlen != plen - prepend_hdr_len) { /* * If this happens it means there is something * wrong with the backend (e.g., some other * process is stealing our packets). */ WPRINTF(("netbe_recv: expected %zd bytes, " "got %zd", plen - prepend_hdr_len, rlen)); vq_retchains(vq, n_chains); continue; } ulen = (uint32_t)plen; /* * Publish the used buffers to the guest, reporting the * number of bytes that we wrote. */ if (!sc->rx_merge) { vq_relchain(vq, info[0].idx, ulen); } else { uint32_t iolen; int i = 0; do { iolen = info[i].len; if (iolen > ulen) { iolen = ulen; } vq_relchain_prepare(vq, info[i].idx, iolen); ulen -= iolen; i++; } while (ulen > 0); hdr->vrh_bufs = i; vq_relchain_publish(vq); assert(i == n_chains); } } } /* * Called when there is read activity on the backend file descriptor. * Each buffer posted by the guest is assumed to be able to contain * an entire ethernet frame + rx header. */ static void pci_vtnet_rx_callback(int fd __unused, enum ev_type type __unused, void *param) { struct pci_vtnet_softc *sc = param; pthread_mutex_lock(&sc->rx_mtx); pci_vtnet_rx(sc); pthread_mutex_unlock(&sc->rx_mtx); } /* Called on RX kick. */ static void pci_vtnet_ping_rxq(void *vsc, struct vqueue_info *vq) { struct pci_vtnet_softc *sc = vsc; /* * A qnotify means that the rx process can now begin. * Enable RX only if features are negotiated. */ pthread_mutex_lock(&sc->rx_mtx); if (!sc->features_negotiated) { pthread_mutex_unlock(&sc->rx_mtx); return; } vq_kick_disable(vq); netbe_rx_enable(sc->vsc_be); pthread_mutex_unlock(&sc->rx_mtx); } /* TX virtqueue processing, called by the TX thread. */ static void pci_vtnet_proctx(struct pci_vtnet_softc *sc, struct vqueue_info *vq) { struct iovec iov[VTNET_MAXSEGS + 1]; struct iovec *siov = iov; struct vi_req req; ssize_t len; int n; /* * Obtain chain of descriptors. The first descriptor also * contains the virtio-net header. */ n = vq_getchain(vq, iov, VTNET_MAXSEGS, &req); assert(n >= 1 && n <= VTNET_MAXSEGS); if (sc->vhdrlen != sc->be_vhdrlen) { /* * The frontend uses a virtio-net header, but the backend * does not. We simply strip the header and ignore it, as * it should be zero-filled. */ siov = iov_trim_hdr(siov, &n, sc->vhdrlen); } if (siov == NULL) { /* The chain is nonsensical. Just drop it. */ len = 0; } else { len = netbe_send(sc->vsc_be, siov, n); if (len < 0) { /* * If send failed, report that 0 bytes * were read. */ len = 0; } } /* * Return the processed chain to the guest, reporting * the number of bytes that we read. */ vq_relchain(vq, req.idx, len); } /* Called on TX kick. */ static void pci_vtnet_ping_txq(void *vsc, struct vqueue_info *vq) { struct pci_vtnet_softc *sc = vsc; /* * Any ring entries to process? */ if (!vq_has_descs(vq)) return; /* Signal the tx thread for processing */ pthread_mutex_lock(&sc->tx_mtx); vq_kick_disable(vq); if (sc->tx_in_progress == 0) pthread_cond_signal(&sc->tx_cond); pthread_mutex_unlock(&sc->tx_mtx); } /* * Thread which will handle processing of TX desc */ static void * pci_vtnet_tx_thread(void *param) { struct pci_vtnet_softc *sc = param; struct vqueue_info *vq; int error; vq = &sc->vsc_queues[VTNET_TXQ]; /* * Let us wait till the tx queue pointers get initialised & * first tx signaled */ pthread_mutex_lock(&sc->tx_mtx); error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx); assert(error == 0); for (;;) { /* note - tx mutex is locked here */ while (sc->resetting || !vq_has_descs(vq)) { vq_kick_enable(vq); if (!sc->resetting && vq_has_descs(vq)) break; sc->tx_in_progress = 0; error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx); assert(error == 0); } vq_kick_disable(vq); sc->tx_in_progress = 1; pthread_mutex_unlock(&sc->tx_mtx); do { /* * Run through entries, placing them into * iovecs and sending when an end-of-packet * is found */ pci_vtnet_proctx(sc, vq); } while (vq_has_descs(vq)); /* * Generate an interrupt if needed. */ vq_endchains(vq, /*used_all_avail=*/1); pthread_mutex_lock(&sc->tx_mtx); } } #ifdef notyet static void pci_vtnet_ping_ctlq(void *vsc, struct vqueue_info *vq) { DPRINTF(("vtnet: control qnotify!")); } #endif static int pci_vtnet_init(struct pci_devinst *pi, nvlist_t *nvl) { struct pci_vtnet_softc *sc; const char *value; char tname[MAXCOMLEN + 1]; unsigned long mtu = ETHERMTU; int err; /* * Allocate data structures for further virtio initializations. * sc also contains a copy of vtnet_vi_consts, since capabilities * change depending on the backend. */ sc = calloc(1, sizeof(struct pci_vtnet_softc)); sc->vsc_consts = vtnet_vi_consts; pthread_mutex_init(&sc->vsc_mtx, NULL); sc->vsc_queues[VTNET_RXQ].vq_qsize = VTNET_RINGSZ; sc->vsc_queues[VTNET_RXQ].vq_notify = pci_vtnet_ping_rxq; sc->vsc_queues[VTNET_TXQ].vq_qsize = VTNET_RINGSZ; sc->vsc_queues[VTNET_TXQ].vq_notify = pci_vtnet_ping_txq; #ifdef notyet sc->vsc_queues[VTNET_CTLQ].vq_qsize = VTNET_RINGSZ; sc->vsc_queues[VTNET_CTLQ].vq_notify = pci_vtnet_ping_ctlq; #endif value = get_config_value_node(nvl, "mac"); if (value != NULL) { err = net_parsemac(value, sc->vsc_config.mac); if (err) { free(sc); return (err); } } else net_genmac(pi, sc->vsc_config.mac); value = get_config_value_node(nvl, "mtu"); if (value != NULL) { err = net_parsemtu(value, &mtu); if (err) { free(sc); return (err); } if (mtu < VTNET_MIN_MTU || mtu > VTNET_MAX_MTU) { err = EINVAL; errno = EINVAL; free(sc); return (err); } sc->vsc_consts.vc_hv_caps |= VIRTIO_NET_F_MTU; } sc->vsc_config.mtu = mtu; /* Permit interfaces without a configured backend. */ if (get_config_value_node(nvl, "backend") != NULL) { err = netbe_init(&sc->vsc_be, nvl, pci_vtnet_rx_callback, sc); if (err) { free(sc); return (err); } } sc->vsc_consts.vc_hv_caps |= VIRTIO_NET_F_MRG_RXBUF | netbe_get_cap(sc->vsc_be); /* * Since we do not actually support multiqueue, * set the maximum virtqueue pairs to 1. */ sc->vsc_config.max_virtqueue_pairs = 1; /* initialize config space */ pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_NET); pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR); pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_NETWORK); pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_ID_NETWORK); pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR); /* Link is always up. */ sc->vsc_config.status = 1; vi_softc_linkup(&sc->vsc_vs, &sc->vsc_consts, sc, pi, sc->vsc_queues); sc->vsc_vs.vs_mtx = &sc->vsc_mtx; /* use BAR 1 to map MSI-X table and PBA, if we're using MSI-X */ if (vi_intr_init(&sc->vsc_vs, 1, fbsdrun_virtio_msix())) { free(sc); return (1); } /* use BAR 0 to map config regs in IO space */ vi_set_io_bar(&sc->vsc_vs, 0); sc->resetting = 0; sc->rx_merge = 0; sc->vhdrlen = sizeof(struct virtio_net_rxhdr) - 2; pthread_mutex_init(&sc->rx_mtx, NULL); /* * Initialize tx semaphore & spawn TX processing thread. * As of now, only one thread for TX desc processing is * spawned. */ sc->tx_in_progress = 0; pthread_mutex_init(&sc->tx_mtx, NULL); pthread_cond_init(&sc->tx_cond, NULL); pthread_create(&sc->tx_tid, NULL, pci_vtnet_tx_thread, (void *)sc); snprintf(tname, sizeof(tname), "vtnet-%d:%d tx", pi->pi_slot, pi->pi_func); pthread_set_name_np(sc->tx_tid, tname); return (0); } static int pci_vtnet_cfgwrite(void *vsc, int offset, int size, uint32_t value) { struct pci_vtnet_softc *sc = vsc; void *ptr; if (offset < (int)sizeof(sc->vsc_config.mac)) { assert(offset + size <= (int)sizeof(sc->vsc_config.mac)); /* * The driver is allowed to change the MAC address */ ptr = &sc->vsc_config.mac[offset]; memcpy(ptr, &value, size); } else { /* silently ignore other writes */ DPRINTF(("vtnet: write to readonly reg %d", offset)); } return (0); } static int pci_vtnet_cfgread(void *vsc, int offset, int size, uint32_t *retval) { struct pci_vtnet_softc *sc = vsc; void *ptr; ptr = (uint8_t *)&sc->vsc_config + offset; memcpy(retval, ptr, size); return (0); } static void pci_vtnet_neg_features(void *vsc, uint64_t negotiated_features) { struct pci_vtnet_softc *sc = vsc; sc->vsc_features = negotiated_features; if (negotiated_features & VIRTIO_NET_F_MRG_RXBUF) { sc->vhdrlen = sizeof(struct virtio_net_rxhdr); sc->rx_merge = 1; } else { /* * Without mergeable rx buffers, virtio-net header is 2 * bytes shorter than sizeof(struct virtio_net_rxhdr). */ sc->vhdrlen = sizeof(struct virtio_net_rxhdr) - 2; sc->rx_merge = 0; } /* Tell the backend to enable some capabilities it has advertised. */ netbe_set_cap(sc->vsc_be, negotiated_features, sc->vhdrlen); sc->be_vhdrlen = netbe_get_vnet_hdr_len(sc->vsc_be); assert(sc->be_vhdrlen == 0 || sc->be_vhdrlen == sc->vhdrlen); pthread_mutex_lock(&sc->rx_mtx); sc->features_negotiated = true; pthread_mutex_unlock(&sc->rx_mtx); } #ifdef BHYVE_SNAPSHOT static void pci_vtnet_pause(void *vsc) { struct pci_vtnet_softc *sc = vsc; DPRINTF(("vtnet: device pause requested !\n")); /* Acquire the RX lock to block RX processing. */ pthread_mutex_lock(&sc->rx_mtx); /* Wait for the transmit thread to finish its processing. */ pthread_mutex_lock(&sc->tx_mtx); while (sc->tx_in_progress) { pthread_mutex_unlock(&sc->tx_mtx); usleep(10000); pthread_mutex_lock(&sc->tx_mtx); } } static void pci_vtnet_resume(void *vsc) { struct pci_vtnet_softc *sc = vsc; DPRINTF(("vtnet: device resume requested !\n")); pthread_mutex_unlock(&sc->tx_mtx); /* The RX lock should have been acquired in vtnet_pause. */ pthread_mutex_unlock(&sc->rx_mtx); } static int pci_vtnet_snapshot(void *vsc, struct vm_snapshot_meta *meta) { int ret; struct pci_vtnet_softc *sc = vsc; DPRINTF(("vtnet: device snapshot requested !\n")); /* * Queues and consts should have been saved by the more generic * vi_pci_snapshot function. We need to save only our features and * config. */ SNAPSHOT_VAR_OR_LEAVE(sc->vsc_features, meta, ret, done); SNAPSHOT_VAR_OR_LEAVE(sc->features_negotiated, meta, ret, done); - /* Force reapply negociated features at restore time */ + /* Force reapply negotiated features at restore time */ if (meta->op == VM_SNAPSHOT_RESTORE && sc->features_negotiated) { pci_vtnet_neg_features(sc, sc->vsc_features); netbe_rx_enable(sc->vsc_be); } SNAPSHOT_VAR_OR_LEAVE(sc->vsc_config, meta, ret, done); SNAPSHOT_VAR_OR_LEAVE(sc->rx_merge, meta, ret, done); SNAPSHOT_VAR_OR_LEAVE(sc->vhdrlen, meta, ret, done); SNAPSHOT_VAR_OR_LEAVE(sc->be_vhdrlen, meta, ret, done); done: return (ret); } #endif static const struct pci_devemu pci_de_vnet = { .pe_emu = "virtio-net", .pe_init = pci_vtnet_init, .pe_legacy_config = netbe_legacy_config, .pe_barwrite = vi_pci_write, .pe_barread = vi_pci_read, #ifdef BHYVE_SNAPSHOT .pe_snapshot = vi_pci_snapshot, .pe_pause = vi_pci_pause, .pe_resume = vi_pci_resume, #endif }; PCI_EMUL_SET(pci_de_vnet); diff --git a/usr.sbin/bhyve/usb_emul.h b/usr.sbin/bhyve/usb_emul.h index 1d713bc55cdc..b2122d6b7e13 100644 --- a/usr.sbin/bhyve/usb_emul.h +++ b/usr.sbin/bhyve/usb_emul.h @@ -1,159 +1,159 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2014 Leon Dang * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _USB_EMUL_H_ #define _USB_EMUL_H_ #include #include #include #include #define USB_MAX_XFER_BLOCKS 8 #define USB_XFER_OUT 0 #define USB_XFER_IN 1 struct usb_hci; struct usb_device_request; struct usb_data_xfer; struct vm_snapshot_meta; /* Device emulation handlers */ struct usb_devemu { const char *ue_emu; /* name of device emulation */ int ue_usbver; /* usb version: 2 or 3 */ int ue_usbspeed; /* usb device speed */ /* instance creation */ void *(*ue_init)(struct usb_hci *hci, nvlist_t *nvl); /* handlers */ int (*ue_request)(void *sc, struct usb_data_xfer *xfer); int (*ue_data)(void *sc, struct usb_data_xfer *xfer, int dir, int epctx); int (*ue_reset)(void *sc); int (*ue_remove)(void *sc); int (*ue_stop)(void *sc); int (*ue_snapshot)(void *scarg, struct vm_snapshot_meta *meta); }; #define USB_EMUL_SET(x) DATA_SET(usb_emu_set, x) /* * USB device events to notify HCI when state changes */ enum hci_usbev { USBDEV_ATTACH, USBDEV_RESET, USBDEV_STOP, USBDEV_REMOVE, }; /* usb controller, ie xhci, ehci */ struct usb_hci { int (*hci_intr)(struct usb_hci *hci, int epctx); int (*hci_event)(struct usb_hci *hci, enum hci_usbev evid, void *param); void *hci_sc; /* private softc for hci */ /* controller managed fields */ int hci_address; int hci_port; }; /* * Each xfer block is mapped to the hci transfer block. - * On input into the device handler, blen is set to the lenght of buf. + * On input into the device handler, blen is set to the length of buf. * The device handler is to update blen to reflect on the residual size * of the buffer, i.e. len(buf) - len(consumed). */ struct usb_data_xfer_block { void *buf; /* IN or OUT pointer */ int blen; /* in:len(buf), out:len(remaining) */ int bdone; /* bytes transferred */ uint32_t processed; /* device processed this + errcode */ void *hci_data; /* HCI private reference */ int ccs; uint32_t streamid; uint64_t trbnext; /* next TRB guest address */ }; struct usb_data_xfer { struct usb_data_xfer_block data[USB_MAX_XFER_BLOCKS]; struct usb_device_request *ureq; /* setup ctl request */ int ndata; /* # of data items */ int head; int tail; pthread_mutex_t mtx; }; enum USB_ERRCODE { USB_ACK, USB_NAK, USB_STALL, USB_NYET, USB_ERR, USB_SHORT }; #define USB_DATA_GET_ERRCODE(x) (x)->processed >> 8 #define USB_DATA_SET_ERRCODE(x,e) do { \ (x)->processed = ((x)->processed & 0xFF) | (e << 8); \ } while (0) #define USB_DATA_OK(x,i) ((x)->data[(i)].buf != NULL) #define USB_DATA_XFER_INIT(x) do { \ memset((x), 0, sizeof(*(x))); \ pthread_mutex_init(&((x)->mtx), NULL); \ } while (0) #define USB_DATA_XFER_RESET(x) do { \ memset((x)->data, 0, sizeof((x)->data)); \ (x)->ndata = 0; \ (x)->head = (x)->tail = 0; \ } while (0) #define USB_DATA_XFER_LOCK(x) do { \ pthread_mutex_lock(&((x)->mtx)); \ } while (0) #define USB_DATA_XFER_UNLOCK(x) do { \ pthread_mutex_unlock(&((x)->mtx)); \ } while (0) struct usb_devemu *usb_emu_finddev(const char *name); struct usb_data_xfer_block *usb_data_xfer_append(struct usb_data_xfer *xfer, void *buf, int blen, void *hci_data, int ccs); #endif /* _USB_EMUL_H_ */