Index: head/sys/mips/nlm/cms.c =================================================================== --- head/sys/mips/nlm/cms.c (revision 343451) +++ head/sys/mips/nlm/cms.c (revision 343452) @@ -1,498 +1,498 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright 2003-2011 Netlogic Microsystems (Netlogic). 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 Netlogic Microsystems ``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 NETLOGIC 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. * * NETLOGIC_BSD */ #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 #define MSGRNG_NSTATIONS 1024 /* * Keep track of our message ring handler threads, each core has a * different message station. Ideally we will need to start a few * message handling threads every core, and wake them up depending on * load */ struct msgring_thread { struct thread *thread; /* msgring handler threads */ int needed; /* thread needs to wake up */ }; static struct msgring_thread msgring_threads[XLP_MAX_CORES * XLP_MAX_THREADS]; static struct proc *msgring_proc; /* all threads are under a proc */ /* * The device drivers can register a handler for the messages sent * from a station (corresponding to the device). */ struct tx_stn_handler { msgring_handler action; void *arg; }; static struct tx_stn_handler msgmap[MSGRNG_NSTATIONS]; static struct mtx msgmap_lock; uint32_t xlp_msg_thread_mask; static int xlp_msg_threads_per_core = XLP_MAX_THREADS; static void create_msgring_thread(int hwtid); static int msgring_process_fast_intr(void *arg); /* Debug counters */ static int msgring_nintr[XLP_MAX_CORES * XLP_MAX_THREADS]; static int msgring_wakeup_sleep[XLP_MAX_CORES * XLP_MAX_THREADS]; static int msgring_wakeup_nosleep[XLP_MAX_CORES * XLP_MAX_THREADS]; static int fmn_msgcount[XLP_MAX_CORES * XLP_MAX_THREADS][4]; static int fmn_loops[XLP_MAX_CORES * XLP_MAX_THREADS]; /* Whether polled driver implementation */ static int polled = 0; /* We do only i/o device credit setup here. CPU credit setup is now * moved to xlp_msgring_cpu_init() so that the credits get setup * only if the CPU exists. xlp_msgring_cpu_init() gets called from * platform_init_ap; and this makes it easy for us to setup CMS * credits for various types of XLP chips, with varying number of * cpu's and cores. */ static void xlp_cms_credit_setup(int credit) { uint64_t cmspcibase, cmsbase, pcibase; uint32_t devoffset; int dev, fn, maxqid; int src, qid, i; for (i = 0; i < XLP_MAX_NODES; i++) { cmspcibase = nlm_get_cms_pcibase(i); if (!nlm_dev_exists(XLP_IO_CMS_OFFSET(i))) continue; cmsbase = nlm_get_cms_regbase(i); maxqid = nlm_read_reg(cmspcibase, XLP_PCI_DEVINFO_REG0); for (dev = 0; dev < 8; dev++) { for (fn = 0; fn < 8; fn++) { devoffset = XLP_HDR_OFFSET(i, 0, dev, fn); if (nlm_dev_exists(devoffset) == 0) continue; pcibase = nlm_pcicfg_base(devoffset); src = nlm_qidstart(pcibase); if (src == 0) continue; #if 0 /* Debug */ printf("Setup CMS credits for queues "); printf("[%d to %d] from src %d\n", 0, maxqid, src); #endif for (qid = 0; qid < maxqid; qid++) nlm_cms_setup_credits(cmsbase, qid, src, credit); } } } } void xlp_msgring_cpu_init(int node, int cpu, int credit) { uint64_t cmspcibase = nlm_get_cms_pcibase(node); uint64_t cmsbase = nlm_get_cms_regbase(node); int qid, maxqid, src; maxqid = nlm_read_reg(cmspcibase, XLP_PCI_DEVINFO_REG0); /* cpu credit setup is done only from thread-0 of each core */ if((cpu % 4) == 0) { src = cpu << 2; /* each thread has 4 vc's */ for (qid = 0; qid < maxqid; qid++) nlm_cms_setup_credits(cmsbase, qid, src, credit); } } /* * Drain out max_messages for the buckets set in the bucket mask. * Use max_msgs = 0 to drain out all messages. */ int xlp_handle_msg_vc(u_int vcmask, int max_msgs) { struct nlm_fmn_msg msg; int srcid = 0, size = 0, code = 0; struct tx_stn_handler *he; uint32_t mflags, status; int n_msgs = 0, vc, m, hwtid; u_int msgmask; hwtid = nlm_cpuid(); for (;;) { /* check if VC empty */ mflags = nlm_save_flags_cop2(); status = nlm_read_c2_msgstatus1(); nlm_restore_flags(mflags); msgmask = ((status >> 24) & 0xf) ^ 0xf; msgmask &= vcmask; if (msgmask == 0) break; m = 0; for (vc = 0; vc < 4; vc++) { if ((msgmask & (1 << vc)) == 0) continue; mflags = nlm_save_flags_cop2(); status = nlm_fmn_msgrcv(vc, &srcid, &size, &code, &msg); nlm_restore_flags(mflags); if (status != 0) /* no msg or error */ continue; - if (srcid < 0 && srcid >= 1024) { + if (srcid < 0 || srcid >= 1024) { printf("[%s]: bad src id %d\n", __func__, srcid); continue; } he = &msgmap[srcid]; if(he->action != NULL) (he->action)(vc, size, code, srcid, &msg, he->arg); #if 0 else printf("[%s]: No Handler for msg from stn %d," " vc=%d, size=%d, msg0=%jx, droppinge\n", __func__, srcid, vc, size, (uintmax_t)msg.msg[0]); #endif fmn_msgcount[hwtid][vc] += 1; m++; /* msgs handled in this iter */ } if (m == 0) break; /* nothing done in this iter */ n_msgs += m; if (max_msgs > 0 && n_msgs >= max_msgs) break; } return (n_msgs); } static void xlp_discard_msg_vc(u_int vcmask) { struct nlm_fmn_msg msg; int srcid = 0, size = 0, code = 0, vc; uint32_t mflags, status; for (vc = 0; vc < 4; vc++) { for (;;) { mflags = nlm_save_flags_cop2(); status = nlm_fmn_msgrcv(vc, &srcid, &size, &code, &msg); nlm_restore_flags(mflags); /* break if there is no msg or error */ if (status != 0) break; } } } void xlp_cms_enable_intr(int node, int cpu, int type, int watermark) { uint64_t cmsbase; int i, qid; cmsbase = nlm_get_cms_regbase(node); for (i = 0; i < 4; i++) { qid = (i + (cpu * 4)) & 0x7f; nlm_cms_per_queue_level_intr(cmsbase, qid, type, watermark); nlm_cms_per_queue_timer_intr(cmsbase, qid, 0x1, 0); } } static int msgring_process_fast_intr(void *arg) { struct msgring_thread *mthd; struct thread *td; int cpu; cpu = nlm_cpuid(); mthd = &msgring_threads[cpu]; msgring_nintr[cpu]++; td = mthd->thread; /* clear pending interrupts */ nlm_write_c0_eirr(1ULL << IRQ_MSGRING); /* wake up the target thread */ mthd->needed = 1; thread_lock(td); if (TD_AWAITING_INTR(td)) { msgring_wakeup_sleep[cpu]++; TD_CLR_IWAIT(td); sched_add(td, SRQ_INTR); } else msgring_wakeup_nosleep[cpu]++; thread_unlock(td); return (FILTER_HANDLED); } static void msgring_process(void * arg) { volatile struct msgring_thread *mthd; struct thread *td; uint32_t mflags, msgstatus1; int hwtid, nmsgs; hwtid = (intptr_t)arg; mthd = &msgring_threads[hwtid]; td = mthd->thread; KASSERT(curthread == td, ("%s:msg_ithread and proc linkage out of sync", __func__)); /* First bind this thread to the right CPU */ thread_lock(td); sched_bind(td, xlp_hwtid_to_cpuid[hwtid]); thread_unlock(td); if (hwtid != nlm_cpuid()) printf("Misscheduled hwtid %d != cpuid %d\n", hwtid, nlm_cpuid()); xlp_discard_msg_vc(0xf); xlp_msgring_cpu_init(nlm_nodeid(), nlm_cpuid(), CMS_DEFAULT_CREDIT); if (polled == 0) { mflags = nlm_save_flags_cop2(); nlm_fmn_cpu_init(IRQ_MSGRING, 0, 0, 0, 0, 0); nlm_restore_flags(mflags); xlp_cms_enable_intr(nlm_nodeid(), nlm_cpuid(), 0x2, 0); /* clear pending interrupts. * they will get re-raised if still valid */ nlm_write_c0_eirr(1ULL << IRQ_MSGRING); } /* start processing messages */ for (;;) { atomic_store_rel_int(&mthd->needed, 0); nmsgs = xlp_handle_msg_vc(0xf, 0); /* sleep */ if (polled == 0) { /* clear VC-pend bits */ mflags = nlm_save_flags_cop2(); msgstatus1 = nlm_read_c2_msgstatus1(); msgstatus1 |= (0xf << 16); nlm_write_c2_msgstatus1(msgstatus1); nlm_restore_flags(mflags); thread_lock(td); if (mthd->needed) { thread_unlock(td); continue; } sched_class(td, PRI_ITHD); TD_SET_IWAIT(td); mi_switch(SW_VOL, NULL); thread_unlock(td); } else pause("wmsg", 1); fmn_loops[hwtid]++; } } static void create_msgring_thread(int hwtid) { struct msgring_thread *mthd; struct thread *td; int error; mthd = &msgring_threads[hwtid]; error = kproc_kthread_add(msgring_process, (void *)(uintptr_t)hwtid, &msgring_proc, &td, RFSTOPPED, 2, "msgrngproc", "msgthr%d", hwtid); if (error) panic("kproc_kthread_add() failed with %d", error); mthd->thread = td; thread_lock(td); sched_class(td, PRI_ITHD); sched_add(td, SRQ_INTR); thread_unlock(td); } int register_msgring_handler(int startb, int endb, msgring_handler action, void *arg) { int i; if (bootverbose) printf("Register handler %d-%d %p(%p)\n", startb, endb, action, arg); KASSERT(startb >= 0 && startb <= endb && endb < MSGRNG_NSTATIONS, ("Invalid value for bucket range %d,%d", startb, endb)); mtx_lock_spin(&msgmap_lock); for (i = startb; i <= endb; i++) { KASSERT(msgmap[i].action == NULL, ("Bucket %d already used [action %p]", i, msgmap[i].action)); msgmap[i].action = action; msgmap[i].arg = arg; } mtx_unlock_spin(&msgmap_lock); return (0); } /* * Initialize the messaging subsystem. * * Message Stations are shared among all threads in a cpu core, this * has to be called once from every core which is online. */ static void xlp_msgring_config(void *arg) { void *cookie; unsigned int thrmask, mask; int i; /* used polled handler for Ax silion */ if (nlm_is_xlp8xx_ax()) polled = 1; /* Don't poll on all threads, if polled */ if (polled) xlp_msg_threads_per_core -= 1; mtx_init(&msgmap_lock, "msgring", NULL, MTX_SPIN); if (xlp_threads_per_core < xlp_msg_threads_per_core) xlp_msg_threads_per_core = xlp_threads_per_core; thrmask = ((1 << xlp_msg_threads_per_core) - 1); mask = 0; for (i = 0; i < XLP_MAX_CORES; i++) { mask <<= XLP_MAX_THREADS; mask |= thrmask; } xlp_msg_thread_mask = xlp_hw_thread_mask & mask; #if 0 printf("CMS Message handler thread mask %#jx\n", (uintmax_t)xlp_msg_thread_mask); #endif xlp_cms_credit_setup(CMS_DEFAULT_CREDIT); create_msgring_thread(0); cpu_establish_hardintr("msgring", msgring_process_fast_intr, NULL, NULL, IRQ_MSGRING, INTR_TYPE_NET, &cookie); } /* * Start message ring processing threads on other CPUs, after SMP start */ static void start_msgring_threads(void *arg) { int hwt; for (hwt = 1; hwt < XLP_MAX_CORES * XLP_MAX_THREADS; hwt++) { if ((xlp_msg_thread_mask & (1 << hwt)) == 0) continue; create_msgring_thread(hwt); } } SYSINIT(xlp_msgring_config, SI_SUB_DRIVERS, SI_ORDER_FIRST, xlp_msgring_config, NULL); SYSINIT(start_msgring_threads, SI_SUB_SMP, SI_ORDER_MIDDLE, start_msgring_threads, NULL); /* * DEBUG support, XXX: static buffer, not locked */ static int sys_print_debug(SYSCTL_HANDLER_ARGS) { struct sbuf sb; int error, i; sbuf_new_for_sysctl(&sb, NULL, 64, req); sbuf_printf(&sb, "\nID vc0 vc1 vc2 vc3 loops\n"); for (i = 0; i < 32; i++) { if ((xlp_hw_thread_mask & (1 << i)) == 0) continue; sbuf_printf(&sb, "%2d: %8d %8d %8d %8d %8d\n", i, fmn_msgcount[i][0], fmn_msgcount[i][1], fmn_msgcount[i][2], fmn_msgcount[i][3], fmn_loops[i]); } error = sbuf_finish(&sb); sbuf_delete(&sb); return (error); } SYSCTL_PROC(_debug, OID_AUTO, msgring, CTLTYPE_STRING | CTLFLAG_RD, 0, 0, sys_print_debug, "A", "msgring debug info");