Index: stable/12/sys/riscv/include/sbi.h =================================================================== --- stable/12/sys/riscv/include/sbi.h (revision 361085) +++ stable/12/sys/riscv/include/sbi.h (revision 361086) @@ -1,207 +1,242 @@ /*- * Copyright (c) 2016-2017 Ruslan Bukin * All rights reserved. * Copyright (c) 2019 Mitchell Horne * * Portions of this software were developed by SRI International and the * University of Cambridge Computer Laboratory under DARPA/AFRL contract * FA8750-10-C-0237 ("CTSRD"), as part of the DARPA CRASH research programme. * * Portions of this software were developed by the University of Cambridge * Computer Laboratory as part of the CTSRD Project, with support from the * UK Higher Education Innovation Fund (HEIF). * * 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 _MACHINE_SBI_H_ #define _MACHINE_SBI_H_ /* SBI Specification Version */ #define SBI_SPEC_VERS_MAJOR_OFFSET 24 #define SBI_SPEC_VERS_MAJOR_MASK (0x7F << SBI_SPEC_VERS_MAJOR_OFFSET) #define SBI_SPEC_VERS_MINOR_OFFSET 0 #define SBI_SPEC_VERS_MINOR_MASK (0xFFFFFF << SBI_SPEC_VERS_MINOR_OFFSET) /* SBI Implementation IDs */ #define SBI_IMPL_ID_BBL 0 #define SBI_IMPL_ID_OPENSBI 1 /* SBI Error Codes */ #define SBI_SUCCESS 0 #define SBI_ERR_FAILURE -1 #define SBI_ERR_NOT_SUPPORTED -2 #define SBI_ERR_INVALID_PARAM -3 #define SBI_ERR_DENIED -4 #define SBI_ERR_INVALID_ADDRESS -5 +#define SBI_ERR_ALREADY_AVAILABLE -6 /* SBI Base Extension */ #define SBI_EXT_ID_BASE 0x10 #define SBI_BASE_GET_SPEC_VERSION 0 #define SBI_BASE_GET_IMPL_ID 1 #define SBI_BASE_GET_IMPL_VERSION 2 #define SBI_BASE_PROBE_EXTENSION 3 #define SBI_BASE_GET_MVENDORID 4 #define SBI_BASE_GET_MARCHID 5 #define SBI_BASE_GET_MIMPID 6 +/* Hart State Management (HSM) Extension */ +#define SBI_EXT_ID_HSM 0x48534D +#define SBI_HSM_HART_START 0 +#define SBI_HSM_HART_STOP 1 +#define SBI_HSM_HART_STATUS 2 +#define SBI_HSM_STATUS_STARTED 0 +#define SBI_HSM_STATUS_STOPPED 1 +#define SBI_HSM_STATUS_START_PENDING 2 +#define SBI_HSM_STATUS_STOP_PENDING 3 + /* Legacy Extensions */ #define SBI_SET_TIMER 0 #define SBI_CONSOLE_PUTCHAR 1 #define SBI_CONSOLE_GETCHAR 2 #define SBI_CLEAR_IPI 3 #define SBI_SEND_IPI 4 #define SBI_REMOTE_FENCE_I 5 #define SBI_REMOTE_SFENCE_VMA 6 #define SBI_REMOTE_SFENCE_VMA_ASID 7 #define SBI_SHUTDOWN 8 #define SBI_CALL0(e, f) SBI_CALL4(e, f, 0, 0, 0, 0) #define SBI_CALL1(e, f, p1) SBI_CALL4(e, f, p1, 0, 0, 0) #define SBI_CALL2(e, f, p1, p2) SBI_CALL4(e, f, p1, p2, 0, 0) #define SBI_CALL3(e, f, p1, p2, p3) SBI_CALL4(e, f, p1, p2, p3, 0) #define SBI_CALL4(e, f, p1, p2, p3, p4) sbi_call(e, f, p1, p2, p3, p4) /* * Documentation available at * https://github.com/riscv/riscv-sbi-doc/blob/master/riscv-sbi.adoc */ struct sbi_ret { long error; long value; }; static __inline struct sbi_ret sbi_call(uint64_t arg7, uint64_t arg6, uint64_t arg0, uint64_t arg1, uint64_t arg2, uint64_t arg3) { struct sbi_ret ret; register uintptr_t a0 __asm ("a0") = (uintptr_t)(arg0); register uintptr_t a1 __asm ("a1") = (uintptr_t)(arg1); register uintptr_t a2 __asm ("a2") = (uintptr_t)(arg2); register uintptr_t a3 __asm ("a3") = (uintptr_t)(arg3); register uintptr_t a6 __asm ("a6") = (uintptr_t)(arg6); register uintptr_t a7 __asm ("a7") = (uintptr_t)(arg7); __asm __volatile( \ "ecall" \ :"+r"(a0), "+r"(a1) \ :"r"(a2), "r"(a3), "r"(a6), "r"(a7) \ :"memory"); ret.error = a0; ret.value = a1; return (ret); } /* Base extension functions and variables. */ extern u_long sbi_spec_version; extern u_long sbi_impl_id; extern u_long sbi_impl_version; static __inline long sbi_probe_extension(long id) { return (SBI_CALL1(SBI_EXT_ID_BASE, SBI_BASE_PROBE_EXTENSION, id).value); } + +/* Hart State Management extension functions. */ + +/* + * Start execution on the specified hart at physical address start_addr. The + * register a0 will contain the hart's ID, and a1 will contain the value of + * priv. + */ +int sbi_hsm_hart_start(u_long hart, u_long start_addr, u_long priv); + +/* + * Stop execution on the current hart. Interrupts should be disabled, or this + * function may return. + */ +void sbi_hsm_hart_stop(void); + +/* + * Get the execution status of the specified hart. The status will be one of: + * - SBI_HSM_STATUS_STARTED + * - SBI_HSM_STATUS_STOPPED + * - SBI_HSM_STATUS_START_PENDING + * - SBI_HSM_STATUS_STOP_PENDING + */ +int sbi_hsm_hart_status(u_long hart); /* Legacy extension functions. */ static __inline void sbi_console_putchar(int ch) { (void)SBI_CALL1(SBI_CONSOLE_PUTCHAR, 0, ch); } static __inline int sbi_console_getchar(void) { /* * XXX: The "error" is returned here because legacy SBI functions * continue to return their value in a0. */ return (SBI_CALL0(SBI_CONSOLE_GETCHAR, 0).error); } static __inline void sbi_set_timer(uint64_t val) { (void)SBI_CALL1(SBI_SET_TIMER, 0, val); } static __inline void sbi_shutdown(void) { (void)SBI_CALL0(SBI_SHUTDOWN, 0); } static __inline void sbi_clear_ipi(void) { (void)SBI_CALL0(SBI_CLEAR_IPI, 0); } static __inline void sbi_send_ipi(const unsigned long *hart_mask) { (void)SBI_CALL1(SBI_SEND_IPI, 0, (uint64_t)hart_mask); } static __inline void sbi_remote_fence_i(const unsigned long *hart_mask) { (void)SBI_CALL1(SBI_REMOTE_FENCE_I, 0, (uint64_t)hart_mask); } static __inline void sbi_remote_sfence_vma(const unsigned long *hart_mask, unsigned long start, unsigned long size) { (void)SBI_CALL3(SBI_REMOTE_SFENCE_VMA, 0, (uint64_t)hart_mask, start, size); } static __inline void sbi_remote_sfence_vma_asid(const unsigned long *hart_mask, unsigned long start, unsigned long size, unsigned long asid) { (void)SBI_CALL4(SBI_REMOTE_SFENCE_VMA_ASID, 0, (uint64_t)hart_mask, start, size, asid); } void sbi_print_version(void); void sbi_init(void); #endif /* !_MACHINE_SBI_H_ */ Index: stable/12/sys/riscv/riscv/locore.S =================================================================== --- stable/12/sys/riscv/riscv/locore.S (revision 361085) +++ stable/12/sys/riscv/riscv/locore.S (revision 361086) @@ -1,335 +1,345 @@ /*- * Copyright (c) 2015-2018 Ruslan Bukin * All rights reserved. * * Portions of this software were developed by SRI International and the * University of Cambridge Computer Laboratory under DARPA/AFRL contract * FA8750-10-C-0237 ("CTSRD"), as part of the DARPA CRASH research programme. * * Portions of this software were developed by the University of Cambridge * Computer Laboratory as part of the CTSRD Project, with support from the * UK Higher Education Innovation Fund (HEIF). * * 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. * * $FreeBSD$ */ #include "assym.inc" #include #include #include #include #include #include .globl kernbase .set kernbase, KERNBASE /* Trap entries */ .text /* Reset vector */ .text .globl _start _start: - /* Get the physical address kernel loaded to */ - lla t0, virt_map - ld t1, 0(t0) - sub t1, t1, t0 - li t2, KERNBASE - sub s9, t2, t1 /* s9 = physmem base */ - /* * a0 = hart id * a1 = dtbp */ /* Pick a hart to run the boot process. */ lla t0, hart_lottery li t1, 1 amoadd.w t0, t1, 0(t0) /* * We must jump to mpentry in the non-BSP case because the offset is * too large to fit in a 12-bit branch immediate. */ beqz t0, 1f j mpentry /* * Page tables */ 1: + /* Get the kernel's load address */ + jal get_physmem + /* Add L1 entry for kernel */ lla s1, pagetable_l1 lla s2, pagetable_l2 /* Link to next level PN */ srli s2, s2, PAGE_SHIFT li a5, KERNBASE srli a5, a5, L1_SHIFT /* >> L1_SHIFT */ andi a5, a5, 0x1ff /* & 0x1ff */ li t4, PTE_V slli t5, s2, PTE_PPN0_S /* (s2 << PTE_PPN0_S) */ or t6, t4, t5 /* Store L1 PTE entry to position */ li a6, PTE_SIZE mulw a5, a5, a6 add t0, s1, a5 sd t6, (t0) /* Level 2 superpages (512 x 2MiB) */ lla s1, pagetable_l2 srli t4, s9, 21 /* Div physmem base by 2 MiB */ li t2, 512 /* Build 512 entries */ add t3, t4, t2 li t5, 0 2: li t0, (PTE_KERN | PTE_X) slli t2, t4, PTE_PPN1_S /* << PTE_PPN1_S */ or t5, t0, t2 sd t5, (s1) /* Store PTE entry to position */ addi s1, s1, PTE_SIZE addi t4, t4, 1 bltu t4, t3, 2b /* Create an L1 page for early devmap */ lla s1, pagetable_l1 lla s2, pagetable_l2_devmap /* Link to next level PN */ srli s2, s2, PAGE_SHIFT li a5, (VM_MAX_KERNEL_ADDRESS - L2_SIZE) srli a5, a5, L1_SHIFT /* >> L1_SHIFT */ andi a5, a5, 0x1ff /* & 0x1ff */ li t4, PTE_V slli t5, s2, PTE_PPN0_S /* (s2 << PTE_PPN0_S) */ or t6, t4, t5 /* Store single level1 PTE entry to position */ li a6, PTE_SIZE mulw a5, a5, a6 add t0, s1, a5 sd t6, (t0) /* Create an L2 page superpage for DTB */ lla s1, pagetable_l2_devmap mv s2, a1 srli s2, s2, PAGE_SHIFT li t0, (PTE_KERN) slli t2, s2, PTE_PPN0_S /* << PTE_PPN0_S */ or t0, t0, t2 /* Store PTE entry to position */ li a6, PTE_SIZE li a5, 510 mulw a5, a5, a6 add t1, s1, a5 sd t0, (t1) /* Page tables END */ /* Setup supervisor trap vector */ lla t0, va sub t0, t0, s9 li t1, KERNBASE add t0, t0, t1 csrw stvec, t0 /* Set page tables base register */ lla s2, pagetable_l1 srli s2, s2, PAGE_SHIFT li t0, SATP_MODE_SV39 or s2, s2, t0 sfence.vma csrw satp, s2 .align 2 va: /* Setup supervisor trap vector */ la t0, cpu_exception_handler csrw stvec, t0 /* Ensure sscratch is zero */ li t0, 0 csrw sscratch, t0 /* Set the global pointer */ .option push .option norelax la gp, __global_pointer$ .option pop /* Initialize stack pointer */ la s3, initstack_end mv sp, s3 addi sp, sp, -PCB_SIZE /* Clear BSS */ la s0, _C_LABEL(__bss_start) la s1, _C_LABEL(_end) 1: sd zero, 0(s0) addi s0, s0, 8 bltu s0, s1, 1b #ifdef SMP /* Store boot hart id. */ la t0, boot_hart sw a0, 0(t0) #endif /* Fill riscv_bootparams */ addi sp, sp, -40 la t0, pagetable_l1 sd t0, 0(sp) /* kern_l1pt */ sd s9, 8(sp) /* kern_phys */ la t0, initstack sd t0, 16(sp) /* kern_stack */ li t0, (VM_MAX_KERNEL_ADDRESS - 2 * L2_SIZE) sd t0, 24(sp) /* dtbp_virt */ sd a1, 32(sp) /* dtbp_phys */ mv a0, sp call _C_LABEL(initriscv) /* Off we go */ call _C_LABEL(mi_startup) +/* + * Get the physical address the kernel is loaded to. Returned in s9. + */ +get_physmem: + lla t0, virt_map /* physical address of virt_map */ + ld t1, 0(t0) /* virtual address of virt_map */ + sub t1, t1, t0 /* calculate phys->virt delta */ + li t2, KERNBASE + sub s9, t2, t1 /* s9 = physmem base */ + ret + .align 4 initstack: .space (PAGE_SIZE * KSTACK_PAGES) initstack_end: ENTRY(sigcode) mv a0, sp addi a0, a0, SF_UC 1: li t0, SYS_sigreturn ecall /* sigreturn failed, exit */ li t0, SYS_exit ecall j 1b END(sigcode) /* This may be copied to the stack, keep it 16-byte aligned */ .align 3 esigcode: .data .align 3 .global szsigcode szsigcode: .quad esigcode - sigcode .align 12 pagetable_l1: .space PAGE_SIZE pagetable_l2: .space PAGE_SIZE pagetable_l2_devmap: .space PAGE_SIZE .align 3 virt_map: .quad virt_map hart_lottery: .space 4 .globl init_pt_va init_pt_va: .quad pagetable_l2 /* XXX: Keep page tables VA */ #ifndef SMP ENTRY(mpentry) 1: wfi j 1b END(mpentry) #else /* * mpentry(unsigned long) * * Called by a core when it is being brought online. */ ENTRY(mpentry) /* * Calculate the offset to __riscv_boot_ap * for the current core, cpuid is in a0. */ li t1, 4 mulw t1, t1, a0 /* Get the pointer */ lla t0, __riscv_boot_ap add t0, t0, t1 1: /* Wait the kernel to be ready */ lw t1, 0(t0) beqz t1, 1b /* Setup stack pointer */ lla t0, bootstack ld sp, 0(t0) + + /* Get the kernel's load address */ + jal get_physmem /* Setup supervisor trap vector */ lla t0, mpva sub t0, t0, s9 li t1, KERNBASE add t0, t0, t1 csrw stvec, t0 /* Set page tables base register */ lla s2, pagetable_l1 srli s2, s2, PAGE_SHIFT li t0, SATP_MODE_SV39 or s2, s2, t0 sfence.vma csrw satp, s2 .align 2 mpva: /* Setup supervisor trap vector */ la t0, cpu_exception_handler csrw stvec, t0 /* Ensure sscratch is zero */ li t0, 0 csrw sscratch, t0 /* Set the global pointer */ .option push .option norelax la gp, __global_pointer$ .option pop call init_secondary END(mpentry) #endif Index: stable/12/sys/riscv/riscv/machdep.c =================================================================== --- stable/12/sys/riscv/riscv/machdep.c (revision 361085) +++ stable/12/sys/riscv/riscv/machdep.c (revision 361086) @@ -1,936 +1,943 @@ /*- * Copyright (c) 2014 Andrew Turner * Copyright (c) 2015-2017 Ruslan Bukin * All rights reserved. * * Portions of this software were developed by SRI International and the * University of Cambridge Computer Laboratory under DARPA/AFRL contract * FA8750-10-C-0237 ("CTSRD"), as part of the DARPA CRASH research programme. * * Portions of this software were developed by the University of Cambridge * Computer Laboratory as part of the CTSRD Project, with support from the * UK Higher Education Innovation Fund (HEIF). * * 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 "opt_platform.h" #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef FPE #include #endif #ifdef FDT #include #include #endif static void get_fpcontext(struct thread *td, mcontext_t *mcp); static void set_fpcontext(struct thread *td, mcontext_t *mcp); struct pcpu __pcpu[MAXCPU]; static struct trapframe proc0_tf; vm_paddr_t phys_avail[PHYS_AVAIL_SIZE + 2]; vm_paddr_t dump_avail[PHYS_AVAIL_SIZE + 2]; int early_boot = 1; int cold = 1; long realmem = 0; long Maxmem = 0; #define DTB_SIZE_MAX (1024 * 1024) #define PHYSMAP_SIZE (2 * (VM_PHYSSEG_MAX - 1)) vm_paddr_t physmap[PHYSMAP_SIZE]; u_int physmap_idx; struct kva_md_info kmi; int64_t dcache_line_size; /* The minimum D cache line size */ int64_t icache_line_size; /* The minimum I cache line size */ int64_t idcache_line_size; /* The minimum cache line size */ uint32_t boot_hart; /* The hart we booted on. */ cpuset_t all_harts; extern int *end; static void cpu_startup(void *dummy) { sbi_print_version(); identify_cpu(); printf("real memory = %ju (%ju MB)\n", ptoa((uintmax_t)realmem), ptoa((uintmax_t)realmem) / (1024 * 1024)); /* * Display any holes after the first chunk of extended memory. */ if (bootverbose) { int indx; printf("Physical memory chunk(s):\n"); for (indx = 0; phys_avail[indx + 1] != 0; indx += 2) { vm_paddr_t size; size = phys_avail[indx + 1] - phys_avail[indx]; printf( "0x%016jx - 0x%016jx, %ju bytes (%ju pages)\n", (uintmax_t)phys_avail[indx], (uintmax_t)phys_avail[indx + 1] - 1, (uintmax_t)size, (uintmax_t)size / PAGE_SIZE); } } vm_ksubmap_init(&kmi); printf("avail memory = %ju (%ju MB)\n", ptoa((uintmax_t)vm_free_count()), ptoa((uintmax_t)vm_free_count()) / (1024 * 1024)); bufinit(); vm_pager_bufferinit(); } SYSINIT(cpu, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL); int cpu_idle_wakeup(int cpu) { return (0); } int fill_regs(struct thread *td, struct reg *regs) { struct trapframe *frame; frame = td->td_frame; regs->sepc = frame->tf_sepc; regs->sstatus = frame->tf_sstatus; regs->ra = frame->tf_ra; regs->sp = frame->tf_sp; regs->gp = frame->tf_gp; regs->tp = frame->tf_tp; memcpy(regs->t, frame->tf_t, sizeof(regs->t)); memcpy(regs->s, frame->tf_s, sizeof(regs->s)); memcpy(regs->a, frame->tf_a, sizeof(regs->a)); return (0); } int set_regs(struct thread *td, struct reg *regs) { struct trapframe *frame; frame = td->td_frame; frame->tf_sepc = regs->sepc; frame->tf_ra = regs->ra; frame->tf_sp = regs->sp; frame->tf_gp = regs->gp; frame->tf_tp = regs->tp; memcpy(frame->tf_t, regs->t, sizeof(frame->tf_t)); memcpy(frame->tf_s, regs->s, sizeof(frame->tf_s)); memcpy(frame->tf_a, regs->a, sizeof(frame->tf_a)); return (0); } int fill_fpregs(struct thread *td, struct fpreg *regs) { #ifdef FPE struct pcb *pcb; pcb = td->td_pcb; if ((pcb->pcb_fpflags & PCB_FP_STARTED) != 0) { /* * If we have just been running FPE instructions we will * need to save the state to memcpy it below. */ if (td == curthread) fpe_state_save(td); memcpy(regs->fp_x, pcb->pcb_x, sizeof(regs->fp_x)); regs->fp_fcsr = pcb->pcb_fcsr; } else #endif memset(regs, 0, sizeof(*regs)); return (0); } int set_fpregs(struct thread *td, struct fpreg *regs) { #ifdef FPE struct trapframe *frame; struct pcb *pcb; frame = td->td_frame; pcb = td->td_pcb; memcpy(pcb->pcb_x, regs->fp_x, sizeof(regs->fp_x)); pcb->pcb_fcsr = regs->fp_fcsr; pcb->pcb_fpflags |= PCB_FP_STARTED; frame->tf_sstatus &= ~SSTATUS_FS_MASK; frame->tf_sstatus |= SSTATUS_FS_CLEAN; #endif return (0); } int fill_dbregs(struct thread *td, struct dbreg *regs) { panic("fill_dbregs"); } int set_dbregs(struct thread *td, struct dbreg *regs) { panic("set_dbregs"); } int ptrace_set_pc(struct thread *td, u_long addr) { panic("ptrace_set_pc"); return (0); } int ptrace_single_step(struct thread *td) { /* TODO; */ return (0); } int ptrace_clear_single_step(struct thread *td) { /* TODO; */ return (0); } void exec_setregs(struct thread *td, struct image_params *imgp, u_long stack) { struct trapframe *tf; struct pcb *pcb; tf = td->td_frame; pcb = td->td_pcb; memset(tf, 0, sizeof(struct trapframe)); tf->tf_a[0] = stack; tf->tf_sp = STACKALIGN(stack); tf->tf_ra = imgp->entry_addr; tf->tf_sepc = imgp->entry_addr; pcb->pcb_fpflags &= ~PCB_FP_STARTED; } /* Sanity check these are the same size, they will be memcpy'd to and fro */ CTASSERT(sizeof(((struct trapframe *)0)->tf_a) == sizeof((struct gpregs *)0)->gp_a); CTASSERT(sizeof(((struct trapframe *)0)->tf_s) == sizeof((struct gpregs *)0)->gp_s); CTASSERT(sizeof(((struct trapframe *)0)->tf_t) == sizeof((struct gpregs *)0)->gp_t); CTASSERT(sizeof(((struct trapframe *)0)->tf_a) == sizeof((struct reg *)0)->a); CTASSERT(sizeof(((struct trapframe *)0)->tf_s) == sizeof((struct reg *)0)->s); CTASSERT(sizeof(((struct trapframe *)0)->tf_t) == sizeof((struct reg *)0)->t); /* Support for FDT configurations only. */ CTASSERT(FDT); int get_mcontext(struct thread *td, mcontext_t *mcp, int clear_ret) { struct trapframe *tf = td->td_frame; memcpy(mcp->mc_gpregs.gp_t, tf->tf_t, sizeof(mcp->mc_gpregs.gp_t)); memcpy(mcp->mc_gpregs.gp_s, tf->tf_s, sizeof(mcp->mc_gpregs.gp_s)); memcpy(mcp->mc_gpregs.gp_a, tf->tf_a, sizeof(mcp->mc_gpregs.gp_a)); if (clear_ret & GET_MC_CLEAR_RET) { mcp->mc_gpregs.gp_a[0] = 0; mcp->mc_gpregs.gp_t[0] = 0; /* clear syscall error */ } mcp->mc_gpregs.gp_ra = tf->tf_ra; mcp->mc_gpregs.gp_sp = tf->tf_sp; mcp->mc_gpregs.gp_gp = tf->tf_gp; mcp->mc_gpregs.gp_tp = tf->tf_tp; mcp->mc_gpregs.gp_sepc = tf->tf_sepc; mcp->mc_gpregs.gp_sstatus = tf->tf_sstatus; get_fpcontext(td, mcp); return (0); } int set_mcontext(struct thread *td, mcontext_t *mcp) { struct trapframe *tf; tf = td->td_frame; /* * Permit changes to the USTATUS bits of SSTATUS. * * Ignore writes to read-only bits (SD, XS). * * Ignore writes to the FS field as set_fpcontext() will set * it explicitly. */ if (((mcp->mc_gpregs.gp_sstatus ^ tf->tf_sstatus) & ~(SSTATUS_SD | SSTATUS_XS_MASK | SSTATUS_FS_MASK | SSTATUS_UPIE | SSTATUS_UIE)) != 0) return (EINVAL); memcpy(tf->tf_t, mcp->mc_gpregs.gp_t, sizeof(tf->tf_t)); memcpy(tf->tf_s, mcp->mc_gpregs.gp_s, sizeof(tf->tf_s)); memcpy(tf->tf_a, mcp->mc_gpregs.gp_a, sizeof(tf->tf_a)); tf->tf_ra = mcp->mc_gpregs.gp_ra; tf->tf_sp = mcp->mc_gpregs.gp_sp; tf->tf_gp = mcp->mc_gpregs.gp_gp; tf->tf_sepc = mcp->mc_gpregs.gp_sepc; tf->tf_sstatus = mcp->mc_gpregs.gp_sstatus; set_fpcontext(td, mcp); return (0); } static void get_fpcontext(struct thread *td, mcontext_t *mcp) { #ifdef FPE struct pcb *curpcb; critical_enter(); curpcb = curthread->td_pcb; KASSERT(td->td_pcb == curpcb, ("Invalid fpe pcb")); if ((curpcb->pcb_fpflags & PCB_FP_STARTED) != 0) { /* * If we have just been running FPE instructions we will * need to save the state to memcpy it below. */ fpe_state_save(td); KASSERT((curpcb->pcb_fpflags & ~PCB_FP_USERMASK) == 0, ("Non-userspace FPE flags set in get_fpcontext")); memcpy(mcp->mc_fpregs.fp_x, curpcb->pcb_x, sizeof(mcp->mc_fpregs)); mcp->mc_fpregs.fp_fcsr = curpcb->pcb_fcsr; mcp->mc_fpregs.fp_flags = curpcb->pcb_fpflags; mcp->mc_flags |= _MC_FP_VALID; } critical_exit(); #endif } static void set_fpcontext(struct thread *td, mcontext_t *mcp) { #ifdef FPE struct pcb *curpcb; #endif td->td_frame->tf_sstatus &= ~SSTATUS_FS_MASK; td->td_frame->tf_sstatus |= SSTATUS_FS_OFF; #ifdef FPE critical_enter(); if ((mcp->mc_flags & _MC_FP_VALID) != 0) { curpcb = curthread->td_pcb; /* FPE usage is enabled, override registers. */ memcpy(curpcb->pcb_x, mcp->mc_fpregs.fp_x, sizeof(mcp->mc_fpregs)); curpcb->pcb_fcsr = mcp->mc_fpregs.fp_fcsr; curpcb->pcb_fpflags = mcp->mc_fpregs.fp_flags & PCB_FP_USERMASK; td->td_frame->tf_sstatus |= SSTATUS_FS_CLEAN; } critical_exit(); #endif } void cpu_idle(int busy) { spinlock_enter(); if (!busy) cpu_idleclock(); if (!sched_runnable()) __asm __volatile( "fence \n" "wfi \n"); if (!busy) cpu_activeclock(); spinlock_exit(); } void cpu_halt(void) { + /* + * Try to power down using the HSM SBI extension and fall back to a + * simple wfi loop. + */ intr_disable(); + if (sbi_probe_extension(SBI_EXT_ID_HSM) != 0) + sbi_hsm_hart_stop(); for (;;) __asm __volatile("wfi"); + /* NOTREACHED */ } /* * Flush the D-cache for non-DMA I/O so that the I-cache can * be made coherent later. */ void cpu_flush_dcache(void *ptr, size_t len) { /* TBD */ } /* Get current clock frequency for the given CPU ID. */ int cpu_est_clockrate(int cpu_id, uint64_t *rate) { panic("cpu_est_clockrate"); } void cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size) { } void spinlock_enter(void) { struct thread *td; register_t reg; td = curthread; if (td->td_md.md_spinlock_count == 0) { reg = intr_disable(); td->td_md.md_spinlock_count = 1; td->td_md.md_saved_sstatus_ie = reg; } else td->td_md.md_spinlock_count++; critical_enter(); } void spinlock_exit(void) { struct thread *td; register_t sstatus_ie; td = curthread; critical_exit(); sstatus_ie = td->td_md.md_saved_sstatus_ie; td->td_md.md_spinlock_count--; if (td->td_md.md_spinlock_count == 0) intr_restore(sstatus_ie); } #ifndef _SYS_SYSPROTO_H_ struct sigreturn_args { ucontext_t *ucp; }; #endif int sys_sigreturn(struct thread *td, struct sigreturn_args *uap) { ucontext_t uc; int error; if (copyin(uap->sigcntxp, &uc, sizeof(uc))) return (EFAULT); error = set_mcontext(td, &uc.uc_mcontext); if (error != 0) return (error); /* Restore signal mask. */ kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0); return (EJUSTRETURN); } /* * Construct a PCB from a trapframe. This is called from kdb_trap() where * we want to start a backtrace from the function that caused us to enter * the debugger. We have the context in the trapframe, but base the trace * on the PCB. The PCB doesn't have to be perfect, as long as it contains * enough for a backtrace. */ void makectx(struct trapframe *tf, struct pcb *pcb) { memcpy(pcb->pcb_s, tf->tf_s, sizeof(tf->tf_s)); pcb->pcb_ra = tf->tf_sepc; pcb->pcb_sp = tf->tf_sp; pcb->pcb_gp = tf->tf_gp; pcb->pcb_tp = tf->tf_tp; } void sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct sigframe *fp, frame; struct sysentvec *sysent; struct trapframe *tf; struct sigacts *psp; struct thread *td; struct proc *p; int onstack; int sig; td = curthread; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); sig = ksi->ksi_signo; psp = p->p_sigacts; mtx_assert(&psp->ps_mtx, MA_OWNED); tf = td->td_frame; onstack = sigonstack(tf->tf_sp); CTR4(KTR_SIG, "sendsig: td=%p (%s) catcher=%p sig=%d", td, p->p_comm, catcher, sig); /* Allocate and validate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) != 0 && !onstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct sigframe *)((uintptr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size); } else { fp = (struct sigframe *)td->td_frame->tf_sp; } /* Make room, keeping the stack aligned */ fp--; fp = (struct sigframe *)STACKALIGN(fp); /* Fill in the frame to copy out */ bzero(&frame, sizeof(frame)); get_mcontext(td, &frame.sf_uc.uc_mcontext, 0); frame.sf_si = ksi->ksi_info; frame.sf_uc.uc_sigmask = *mask; frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((onstack) ? SS_ONSTACK : 0) : SS_DISABLE; frame.sf_uc.uc_stack = td->td_sigstk; mtx_unlock(&psp->ps_mtx); PROC_UNLOCK(td->td_proc); /* Copy the sigframe out to the user's stack. */ if (copyout(&frame, fp, sizeof(*fp)) != 0) { /* Process has trashed its stack. Kill it. */ CTR2(KTR_SIG, "sendsig: sigexit td=%p fp=%p", td, fp); PROC_LOCK(p); sigexit(td, SIGILL); } tf->tf_a[0] = sig; tf->tf_a[1] = (register_t)&fp->sf_si; tf->tf_a[2] = (register_t)&fp->sf_uc; tf->tf_sepc = (register_t)catcher; tf->tf_sp = (register_t)fp; sysent = p->p_sysent; if (sysent->sv_sigcode_base != 0) tf->tf_ra = (register_t)sysent->sv_sigcode_base; else tf->tf_ra = (register_t)(sysent->sv_psstrings - *(sysent->sv_szsigcode)); CTR3(KTR_SIG, "sendsig: return td=%p pc=%#x sp=%#x", td, tf->tf_sepc, tf->tf_sp); PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } static void init_proc0(vm_offset_t kstack) { struct pcpu *pcpup; pcpup = &__pcpu[0]; proc_linkup0(&proc0, &thread0); thread0.td_kstack = kstack; thread0.td_kstack_pages = KSTACK_PAGES; thread0.td_pcb = (struct pcb *)(thread0.td_kstack + thread0.td_kstack_pages * PAGE_SIZE) - 1; thread0.td_pcb->pcb_fpflags = 0; thread0.td_frame = &proc0_tf; pcpup->pc_curpcb = thread0.td_pcb; } static int add_physmap_entry(uint64_t base, uint64_t length, vm_paddr_t *physmap, u_int *physmap_idxp) { u_int i, insert_idx, _physmap_idx; _physmap_idx = *physmap_idxp; if (length == 0) return (1); /* * Find insertion point while checking for overlap. Start off by * assuming the new entry will be added to the end. */ insert_idx = _physmap_idx; for (i = 0; i <= _physmap_idx; i += 2) { if (base < physmap[i + 1]) { if (base + length <= physmap[i]) { insert_idx = i; break; } if (boothowto & RB_VERBOSE) printf( "Overlapping memory regions, ignoring second region\n"); return (1); } } /* See if we can prepend to the next entry. */ if (insert_idx <= _physmap_idx && base + length == physmap[insert_idx]) { physmap[insert_idx] = base; return (1); } /* See if we can append to the previous entry. */ if (insert_idx > 0 && base == physmap[insert_idx - 1]) { physmap[insert_idx - 1] += length; return (1); } _physmap_idx += 2; *physmap_idxp = _physmap_idx; if (_physmap_idx == PHYSMAP_SIZE) { printf( "Too many segments in the physical address map, giving up\n"); return (0); } /* * Move the last 'N' entries down to make room for the new * entry if needed. */ for (i = _physmap_idx; i > insert_idx; i -= 2) { physmap[i] = physmap[i - 2]; physmap[i + 1] = physmap[i - 1]; } /* Insert the new entry. */ physmap[insert_idx] = base; physmap[insert_idx + 1] = base + length; printf("physmap[%d] = 0x%016lx\n", insert_idx, base); printf("physmap[%d] = 0x%016lx\n", insert_idx + 1, base + length); return (1); } #ifdef FDT static void try_load_dtb(caddr_t kmdp, vm_offset_t dtbp) { #if defined(FDT_DTB_STATIC) dtbp = (vm_offset_t)&fdt_static_dtb; #endif if (dtbp == (vm_offset_t)NULL) { printf("ERROR loading DTB\n"); return; } if (OF_install(OFW_FDT, 0) == FALSE) panic("Cannot install FDT"); if (OF_init((void *)dtbp) != 0) panic("OF_init failed with the found device tree"); } #endif static void cache_setup(void) { /* TODO */ dcache_line_size = 0; icache_line_size = 0; idcache_line_size = 0; } /* * Fake up a boot descriptor table. * RISCVTODO: This needs to be done via loader (when it's available). */ vm_offset_t fake_preload_metadata(struct riscv_bootparams *rvbp __unused) { static uint32_t fake_preload[35]; #ifdef DDB vm_offset_t zstart = 0, zend = 0; #endif vm_offset_t lastaddr; int i; i = 0; fake_preload[i++] = MODINFO_NAME; fake_preload[i++] = strlen("kernel") + 1; strcpy((char*)&fake_preload[i++], "kernel"); i += 1; fake_preload[i++] = MODINFO_TYPE; fake_preload[i++] = strlen("elf64 kernel") + 1; strcpy((char*)&fake_preload[i++], "elf64 kernel"); i += 3; fake_preload[i++] = MODINFO_ADDR; fake_preload[i++] = sizeof(vm_offset_t); *(vm_offset_t *)&fake_preload[i++] = (vm_offset_t)(KERNBASE + KERNENTRY); i += 1; fake_preload[i++] = MODINFO_SIZE; fake_preload[i++] = sizeof(vm_offset_t); fake_preload[i++] = (vm_offset_t)&end - (vm_offset_t)(KERNBASE + KERNENTRY); i += 1; #ifdef DDB #if 0 /* RISCVTODO */ if (*(uint32_t *)KERNVIRTADDR == MAGIC_TRAMP_NUMBER) { fake_preload[i++] = MODINFO_METADATA|MODINFOMD_SSYM; fake_preload[i++] = sizeof(vm_offset_t); fake_preload[i++] = *(uint32_t *)(KERNVIRTADDR + 4); fake_preload[i++] = MODINFO_METADATA|MODINFOMD_ESYM; fake_preload[i++] = sizeof(vm_offset_t); fake_preload[i++] = *(uint32_t *)(KERNVIRTADDR + 8); lastaddr = *(uint32_t *)(KERNVIRTADDR + 8); zend = lastaddr; zstart = *(uint32_t *)(KERNVIRTADDR + 4); db_fetch_ksymtab(zstart, zend); } else #endif #endif lastaddr = (vm_offset_t)&end; fake_preload[i++] = 0; fake_preload[i] = 0; preload_metadata = (void *)fake_preload; return (lastaddr); } void initriscv(struct riscv_bootparams *rvbp) { struct mem_region mem_regions[FDT_MEM_REGIONS]; struct pcpu *pcpup; vm_offset_t rstart, rend; vm_offset_t s, e; int mem_regions_sz; vm_offset_t lastaddr; vm_size_t kernlen; caddr_t kmdp; int i; TSRAW(&thread0, TS_ENTER, __func__, NULL); /* Set the pcpu data, this is needed by pmap_bootstrap */ pcpup = &__pcpu[0]; pcpu_init(pcpup, 0, sizeof(struct pcpu)); pcpup->pc_hart = boot_hart; /* Set the pcpu pointer */ __asm __volatile("mv tp, %0" :: "r"(pcpup)); PCPU_SET(curthread, &thread0); /* Initialize SBI interface. */ sbi_init(); /* Set the module data location */ lastaddr = fake_preload_metadata(rvbp); /* Find the kernel address */ kmdp = preload_search_by_type("elf kernel"); if (kmdp == NULL) kmdp = preload_search_by_type("elf64 kernel"); boothowto = RB_VERBOSE | RB_SINGLE; boothowto = RB_VERBOSE; kern_envp = NULL; #ifdef FDT try_load_dtb(kmdp, rvbp->dtbp_virt); #endif /* Load the physical memory ranges */ physmap_idx = 0; #ifdef FDT /* Grab physical memory regions information from device tree. */ if (fdt_get_mem_regions(mem_regions, &mem_regions_sz, NULL) != 0) panic("Cannot get physical memory regions"); s = rvbp->dtbp_phys; e = s + DTB_SIZE_MAX; for (i = 0; i < mem_regions_sz; i++) { rstart = mem_regions[i].mr_start; rend = (mem_regions[i].mr_start + mem_regions[i].mr_size); if ((rstart < s) && (rend > e)) { /* Exclude DTB region. */ add_physmap_entry(rstart, (s - rstart), physmap, &physmap_idx); add_physmap_entry(e, (rend - e), physmap, &physmap_idx); } else { add_physmap_entry(mem_regions[i].mr_start, mem_regions[i].mr_size, physmap, &physmap_idx); } } #endif /* Do basic tuning, hz etc */ init_param1(); cache_setup(); /* Bootstrap enough of pmap to enter the kernel proper */ kernlen = (lastaddr - KERNBASE); pmap_bootstrap(rvbp->kern_l1pt, mem_regions[0].mr_start, kernlen); cninit(); init_proc0(rvbp->kern_stack); msgbufinit(msgbufp, msgbufsize); mutex_init(); init_param2(physmem); kdb_init(); early_boot = 0; TSEXIT(); } #undef bzero void bzero(void *buf, size_t len) { uint8_t *p; p = buf; while(len-- > 0) *p++ = 0; } Index: stable/12/sys/riscv/riscv/mp_machdep.c =================================================================== --- stable/12/sys/riscv/riscv/mp_machdep.c (revision 361085) +++ stable/12/sys/riscv/riscv/mp_machdep.c (revision 361086) @@ -1,534 +1,554 @@ /*- * Copyright (c) 2015 The FreeBSD Foundation * Copyright (c) 2016 Ruslan Bukin * All rights reserved. * * Portions of this software were developed by Andrew Turner under * sponsorship from the FreeBSD Foundation. * * Portions of this software were developed by SRI International and the * University of Cambridge Computer Laboratory under DARPA/AFRL contract * FA8750-10-C-0237 ("CTSRD"), as part of the DARPA CRASH research programme. * * Portions of this software were developed by the University of Cambridge * Computer Laboratory as part of the CTSRD Project, with support from the * UK Higher Education Innovation Fund (HEIF). * * 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 "opt_kstack_pages.h" #include "opt_platform.h" #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef FDT #include #include #endif boolean_t ofw_cpu_reg(phandle_t node, u_int, cell_t *); uint32_t __riscv_boot_ap[MAXCPU]; static enum { CPUS_UNKNOWN, #ifdef FDT CPUS_FDT, #endif } cpu_enum_method; static device_identify_t riscv64_cpu_identify; static device_probe_t riscv64_cpu_probe; static device_attach_t riscv64_cpu_attach; static int ipi_handler(void *); struct pcb stoppcbs[MAXCPU]; extern uint32_t boot_hart; extern cpuset_t all_harts; #ifdef INVARIANTS static uint32_t cpu_reg[MAXCPU][2]; #endif static device_t cpu_list[MAXCPU]; +void mpentry(u_long hartid); void init_secondary(uint64_t); static struct mtx ap_boot_mtx; /* Stacks for AP initialization, discarded once idle threads are started. */ void *bootstack; static void *bootstacks[MAXCPU]; /* Count of started APs, used to synchronize access to bootstack. */ static volatile int aps_started; /* Set to 1 once we're ready to let the APs out of the pen. */ static volatile int aps_ready; /* Temporary variables for init_secondary() */ void *dpcpu[MAXCPU - 1]; static device_method_t riscv64_cpu_methods[] = { /* Device interface */ DEVMETHOD(device_identify, riscv64_cpu_identify), DEVMETHOD(device_probe, riscv64_cpu_probe), DEVMETHOD(device_attach, riscv64_cpu_attach), DEVMETHOD_END }; static devclass_t riscv64_cpu_devclass; static driver_t riscv64_cpu_driver = { "riscv64_cpu", riscv64_cpu_methods, 0 }; DRIVER_MODULE(riscv64_cpu, cpu, riscv64_cpu_driver, riscv64_cpu_devclass, 0, 0); static void riscv64_cpu_identify(driver_t *driver, device_t parent) { if (device_find_child(parent, "riscv64_cpu", -1) != NULL) return; if (BUS_ADD_CHILD(parent, 0, "riscv64_cpu", -1) == NULL) device_printf(parent, "add child failed\n"); } static int riscv64_cpu_probe(device_t dev) { u_int cpuid; cpuid = device_get_unit(dev); if (cpuid >= MAXCPU || cpuid > mp_maxid) return (EINVAL); device_quiet(dev); return (0); } static int riscv64_cpu_attach(device_t dev) { const uint32_t *reg; size_t reg_size; u_int cpuid; int i; cpuid = device_get_unit(dev); if (cpuid >= MAXCPU || cpuid > mp_maxid) return (EINVAL); KASSERT(cpu_list[cpuid] == NULL, ("Already have cpu %u", cpuid)); reg = cpu_get_cpuid(dev, ®_size); if (reg == NULL) return (EINVAL); if (bootverbose) { device_printf(dev, "register <"); for (i = 0; i < reg_size; i++) printf("%s%x", (i == 0) ? "" : " ", reg[i]); printf(">\n"); } /* Set the device to start it later */ cpu_list[cpuid] = dev; return (0); } static void release_aps(void *dummy __unused) { cpuset_t mask; int cpu, i; if (mp_ncpus == 1) return; /* Setup the IPI handler */ riscv_setup_ipihandler(ipi_handler); atomic_store_rel_int(&aps_ready, 1); /* Wake up the other CPUs */ mask = all_harts; CPU_CLR(boot_hart, &mask); printf("Release APs\n"); sbi_send_ipi(mask.__bits); for (i = 0; i < 2000; i++) { if (smp_started) { for (cpu = 0; cpu <= mp_maxid; cpu++) { if (CPU_ABSENT(cpu)) continue; } return; } DELAY(1000); } printf("APs not started\n"); } SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, release_aps, NULL); void init_secondary(uint64_t hart) { struct pcpu *pcpup; u_int cpuid; /* Renumber this cpu */ cpuid = hart; if (cpuid < boot_hart) cpuid += mp_maxid + 1; cpuid -= boot_hart; /* Setup the pcpu pointer */ pcpup = &__pcpu[cpuid]; __asm __volatile("mv tp, %0" :: "r"(pcpup)); /* Workaround: make sure wfi doesn't halt the hart */ csr_set(sie, SIE_SSIE); csr_set(sip, SIE_SSIE); /* Signal the BSP and spin until it has released all APs. */ atomic_add_int(&aps_started, 1); while (!atomic_load_int(&aps_ready)) __asm __volatile("wfi"); /* Initialize curthread */ KASSERT(PCPU_GET(idlethread) != NULL, ("no idle thread")); pcpup->pc_curthread = pcpup->pc_idlethread; /* * Identify current CPU. This is necessary to setup * affinity registers and to provide support for * runtime chip identification. */ identify_cpu(); /* Enable software interrupts */ riscv_unmask_ipi(); #ifndef EARLY_AP_STARTUP /* Start per-CPU event timers. */ cpu_initclocks_ap(); #endif /* Enable external (PLIC) interrupts */ csr_set(sie, SIE_SEIE); /* Activate process 0's pmap. */ pmap_activate_boot(vmspace_pmap(proc0.p_vmspace)); mtx_lock_spin(&ap_boot_mtx); atomic_add_rel_32(&smp_cpus, 1); if (smp_cpus == mp_ncpus) { /* enable IPI's, tlb shootdown, freezes etc */ atomic_store_rel_int(&smp_started, 1); } mtx_unlock_spin(&ap_boot_mtx); /* * Assert that smp_after_idle_runnable condition is reasonable. */ MPASS(PCPU_GET(curpcb) == NULL); /* Enter the scheduler */ sched_throw(NULL); panic("scheduler returned us to init_secondary"); /* NOTREACHED */ } static void smp_after_idle_runnable(void *arg __unused) { struct pcpu *pc; int cpu; - for (cpu = 1; cpu < mp_ncpus; cpu++) { + for (cpu = 1; cpu <= mp_maxid; cpu++) { if (bootstacks[cpu] != NULL) { pc = pcpu_find(cpu); while ((void *)atomic_load_ptr(&pc->pc_curpcb) == NULL) cpu_spinwait(); kmem_free((vm_offset_t)bootstacks[cpu], PAGE_SIZE); } } } SYSINIT(smp_after_idle_runnable, SI_SUB_SMP, SI_ORDER_ANY, smp_after_idle_runnable, NULL); static int ipi_handler(void *arg) { u_int ipi_bitmap; u_int cpu, ipi; int bit; sbi_clear_ipi(); cpu = PCPU_GET(cpuid); mb(); ipi_bitmap = atomic_readandclear_int(PCPU_PTR(pending_ipis)); if (ipi_bitmap == 0) return (FILTER_HANDLED); while ((bit = ffs(ipi_bitmap))) { bit = (bit - 1); ipi = (1 << bit); ipi_bitmap &= ~ipi; mb(); switch (ipi) { case IPI_AST: CTR0(KTR_SMP, "IPI_AST"); break; case IPI_PREEMPT: CTR1(KTR_SMP, "%s: IPI_PREEMPT", __func__); sched_preempt(curthread); break; case IPI_RENDEZVOUS: CTR0(KTR_SMP, "IPI_RENDEZVOUS"); smp_rendezvous_action(); break; case IPI_STOP: case IPI_STOP_HARD: CTR0(KTR_SMP, (ipi == IPI_STOP) ? "IPI_STOP" : "IPI_STOP_HARD"); savectx(&stoppcbs[cpu]); /* Indicate we are stopped */ CPU_SET_ATOMIC(cpu, &stopped_cpus); /* Wait for restart */ while (!CPU_ISSET(cpu, &started_cpus)) cpu_spinwait(); CPU_CLR_ATOMIC(cpu, &started_cpus); CPU_CLR_ATOMIC(cpu, &stopped_cpus); CTR0(KTR_SMP, "IPI_STOP (restart)"); /* * The kernel debugger might have set a breakpoint, * so flush the instruction cache. */ fence_i(); break; case IPI_HARDCLOCK: CTR1(KTR_SMP, "%s: IPI_HARDCLOCK", __func__); hardclockintr(); break; default: panic("Unknown IPI %#0x on cpu %d", ipi, curcpu); } } return (FILTER_HANDLED); } struct cpu_group * cpu_topo(void) { return (smp_topo_none()); } /* Determine if we running MP machine */ int cpu_mp_probe(void) { return (mp_ncpus > 1); } #ifdef FDT static boolean_t cpu_init_fdt(u_int id, phandle_t node, u_int addr_size, pcell_t *reg) { struct pcpu *pcpup; + vm_paddr_t start_addr; uint64_t hart; u_int cpuid; int naps; + int error; /* Check if this hart supports MMU. */ if (OF_getproplen(node, "mmu-type") < 0) return (0); KASSERT(id < MAXCPU, ("Too many CPUs")); KASSERT(addr_size == 1 || addr_size == 2, ("Invalid register size")); #ifdef INVARIANTS cpu_reg[id][0] = reg[0]; if (addr_size == 2) cpu_reg[id][1] = reg[1]; #endif hart = reg[0]; if (addr_size == 2) { hart <<= 32; hart |= reg[1]; } KASSERT(hart < MAXCPU, ("Too many harts.")); /* We are already running on this cpu */ if (hart == boot_hart) return (1); /* * Rotate the CPU IDs to put the boot CPU as CPU 0. * We keep the other CPUs ordered. */ cpuid = hart; if (cpuid < boot_hart) cpuid += mp_maxid + 1; cpuid -= boot_hart; /* Check if we are able to start this cpu */ if (cpuid > mp_maxid) return (0); + + /* + * Depending on the SBI implementation, APs are waiting either in + * locore.S or to be activated explicitly, via SBI call. + */ + if (sbi_probe_extension(SBI_EXT_ID_HSM) != 0) { + start_addr = pmap_kextract((vm_offset_t)mpentry); + error = sbi_hsm_hart_start(hart, start_addr, 0); + if (error != 0) { + mp_ncpus--; + + /* Send a warning to the user and continue. */ + printf("AP %u (hart %lu) failed to start, error %d\n", + cpuid, hart, error); + return (0); + } + } pcpup = &__pcpu[cpuid]; pcpu_init(pcpup, cpuid, sizeof(struct pcpu)); pcpup->pc_hart = hart; dpcpu[cpuid - 1] = (void *)kmem_malloc(DPCPU_SIZE, M_WAITOK | M_ZERO); dpcpu_init(dpcpu[cpuid - 1], cpuid); bootstacks[cpuid] = (void *)kmem_malloc(PAGE_SIZE, M_WAITOK | M_ZERO); naps = atomic_load_int(&aps_started); bootstack = (char *)bootstacks[cpuid] + PAGE_SIZE; printf("Starting CPU %u (hart %lx)\n", cpuid, hart); atomic_store_32(&__riscv_boot_ap[hart], 1); /* Wait for the AP to switch to its boot stack. */ while (atomic_load_int(&aps_started) < naps + 1) cpu_spinwait(); CPU_SET(cpuid, &all_cpus); CPU_SET(hart, &all_harts); return (1); } #endif /* Initialize and fire up non-boot processors */ void cpu_mp_start(void) { mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN); CPU_SET(0, &all_cpus); CPU_SET(boot_hart, &all_harts); switch(cpu_enum_method) { #ifdef FDT case CPUS_FDT: ofw_cpu_early_foreach(cpu_init_fdt, true); break; #endif case CPUS_UNKNOWN: break; } } /* Introduce rest of cores to the world */ void cpu_mp_announce(void) { } static boolean_t cpu_check_mmu(u_int id, phandle_t node, u_int addr_size, pcell_t *reg) { /* Check if this hart supports MMU. */ if (OF_getproplen(node, "mmu-type") < 0) return (0); return (1); } void cpu_mp_setmaxid(void) { #ifdef FDT int cores; cores = ofw_cpu_early_foreach(cpu_check_mmu, true); if (cores > 0) { cores = MIN(cores, MAXCPU); if (bootverbose) printf("Found %d CPUs in the device tree\n", cores); mp_ncpus = cores; mp_maxid = cores - 1; cpu_enum_method = CPUS_FDT; return; } #endif if (bootverbose) printf("No CPU data, limiting to 1 core\n"); mp_ncpus = 1; mp_maxid = 0; } Index: stable/12/sys/riscv/riscv/sbi.c =================================================================== --- stable/12/sys/riscv/riscv/sbi.c (revision 361085) +++ stable/12/sys/riscv/riscv/sbi.c (revision 361086) @@ -1,164 +1,189 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2019 Mitchell Horne * * 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 /* SBI Implementation-Specific Definitions */ #define OPENSBI_VERSION_MAJOR_OFFSET 16 #define OPENSBI_VERSION_MINOR_MASK 0xFFFF u_long sbi_spec_version; u_long sbi_impl_id; u_long sbi_impl_version; static struct sbi_ret sbi_get_spec_version(void) { return (SBI_CALL0(SBI_EXT_ID_BASE, SBI_BASE_GET_SPEC_VERSION)); } static struct sbi_ret sbi_get_impl_id(void) { return (SBI_CALL0(SBI_EXT_ID_BASE, SBI_BASE_GET_IMPL_ID)); } static struct sbi_ret sbi_get_impl_version(void) { return (SBI_CALL0(SBI_EXT_ID_BASE, SBI_BASE_GET_IMPL_VERSION)); } static struct sbi_ret sbi_get_mvendorid(void) { return (SBI_CALL0(SBI_EXT_ID_BASE, SBI_BASE_GET_MVENDORID)); } static struct sbi_ret sbi_get_marchid(void) { return (SBI_CALL0(SBI_EXT_ID_BASE, SBI_BASE_GET_MARCHID)); } static struct sbi_ret sbi_get_mimpid(void) { return (SBI_CALL0(SBI_EXT_ID_BASE, SBI_BASE_GET_MIMPID)); } void sbi_print_version(void) { u_int major; u_int minor; /* For legacy SBI implementations. */ if (sbi_spec_version == 0) { printf("SBI: Unknown (Legacy) Implementation\n"); printf("SBI Specification Version: 0.1\n"); return; } switch (sbi_impl_id) { case (SBI_IMPL_ID_BBL): printf("SBI: Berkely Boot Loader %u\n", sbi_impl_version); break; case (SBI_IMPL_ID_OPENSBI): major = sbi_impl_version >> OPENSBI_VERSION_MAJOR_OFFSET; minor = sbi_impl_version & OPENSBI_VERSION_MINOR_MASK; printf("SBI: OpenSBI v%u.%u\n", major, minor); break; default: printf("SBI: Unrecognized Implementation: %u\n", sbi_impl_id); break; } major = (sbi_spec_version & SBI_SPEC_VERS_MAJOR_MASK) >> SBI_SPEC_VERS_MAJOR_OFFSET; minor = (sbi_spec_version & SBI_SPEC_VERS_MINOR_MASK); printf("SBI Specification Version: %u.%u\n", major, minor); } +int +sbi_hsm_hart_start(u_long hart, u_long start_addr, u_long priv) +{ + struct sbi_ret ret; + + ret = SBI_CALL3(SBI_EXT_ID_HSM, SBI_HSM_HART_START, hart, start_addr, priv); + return (ret.error != 0 ? (int)ret.error : 0); +} + +void +sbi_hsm_hart_stop(void) +{ + (void)SBI_CALL0(SBI_EXT_ID_HSM, SBI_HSM_HART_STOP); +} + +int +sbi_hsm_hart_status(u_long hart) +{ + struct sbi_ret ret; + + ret = SBI_CALL1(SBI_EXT_ID_HSM, SBI_HSM_HART_STATUS, hart); + + return (ret.error != 0 ? (int)ret.error : (int)ret.value); +} + void sbi_init(void) { struct sbi_ret sret; /* * Get the spec version. For legacy SBI implementations this will * return an error, otherwise it is guaranteed to succeed. */ sret = sbi_get_spec_version(); if (sret.error != 0) { /* We are running a legacy SBI implementation. */ sbi_spec_version = 0; return; } /* Set the SBI implementation info. */ sbi_spec_version = sret.value; sbi_impl_id = sbi_get_impl_id().value; sbi_impl_version = sbi_get_impl_version().value; /* Set the hardware implementation info. */ mvendorid = sbi_get_mvendorid().value; marchid = sbi_get_marchid().value; mimpid = sbi_get_mimpid().value; /* * Probe for legacy extensions. Currently we rely on all of them * to be implemented, but this is not guaranteed by the spec. */ KASSERT(sbi_probe_extension(SBI_SET_TIMER) != 0, ("SBI doesn't implement sbi_set_timer()")); KASSERT(sbi_probe_extension(SBI_CONSOLE_PUTCHAR) != 0, ("SBI doesn't implement sbi_console_putchar()")); KASSERT(sbi_probe_extension(SBI_CONSOLE_GETCHAR) != 0, ("SBI doesn't implement sbi_console_getchar()")); KASSERT(sbi_probe_extension(SBI_CLEAR_IPI) != 0, ("SBI doesn't implement sbi_clear_ipi()")); KASSERT(sbi_probe_extension(SBI_SEND_IPI) != 0, ("SBI doesn't implement sbi_send_ipi()")); KASSERT(sbi_probe_extension(SBI_REMOTE_FENCE_I) != 0, ("SBI doesn't implement sbi_remote_fence_i()")); KASSERT(sbi_probe_extension(SBI_REMOTE_SFENCE_VMA) != 0, ("SBI doesn't implement sbi_remote_sfence_vma()")); KASSERT(sbi_probe_extension(SBI_REMOTE_SFENCE_VMA_ASID) != 0, ("SBI doesn't implement sbi_remote_sfence_vma_asid()")); KASSERT(sbi_probe_extension(SBI_SHUTDOWN) != 0, ("SBI doesn't implement sbi_shutdown()")); }