Date: Fri, 18 Nov 2011 13:15:12 +0530 From: "Jayachandran C." <jchandra@freebsd.org> To: freebsd-mips@freebsd.org Subject: pmap changes for mips SMP Message-ID: <CA%2B7sy7Aj89iTTD=7ywiatyBkBW2OjborNydZTf8dgXLNjY8G1w@mail.gmail.com>
index | next in thread | raw e-mail
I've committed this change to MIPS pmap.c, this optimization gives significant improvement in fork/exec performance on SMP systems with large number of CPUs. I have stress tested it on XLR/XLP, but let me know if this causes regressions on any other platforms. Thanks, JC. ---------- Forwarded message ---------- From: Jayachandran C. <jchandra@freebsd.org> Date: Thu, Nov 17, 2011 at 6:44 PM Subject: svn commit: r227623 - head/sys/mips/mips Author: jchandra Date: Thu Nov 17 13:14:59 2011 New Revision: 227623 URL: http://svn.freebsd.org/changeset/base/227623 Log: Do pmap update only on active CPUs. The pmap update_page/invalidate_page/invalidate_all operations has to be done only on active cpus. In the simplest case, if the process is not active on any other CPUs, we can just do the operation on the current CPU. This change replaces the call to smp_rendezvous() for these operations with smp_rendezvous_cpus() in case there more than one active CPU, or with a direct function call if there is just one active CPU. This change give significant performance increase in fork/exec benchmarks on XLR/XLS/XLP with 32 cpus. Reviewed by: alc Modified: head/sys/mips/mips/pmap.c Modified: head/sys/mips/mips/pmap.c ============================================================================== --- head/sys/mips/mips/pmap.c Thu Nov 17 13:14:07 2011 (r227622) +++ head/sys/mips/mips/pmap.c Thu Nov 17 13:14:59 2011 (r227623) @@ -181,7 +181,6 @@ static pt_entry_t init_pte_prot(vm_offse #ifdef SMP static void pmap_invalidate_page_action(void *arg); -static void pmap_invalidate_all_action(void *arg); static void pmap_update_page_action(void *arg); #endif @@ -622,119 +621,94 @@ pmap_init(void) * Low level helper routines..... ***************************************************/ +#ifdef SMP static __inline void -pmap_invalidate_all_local(pmap_t pmap) +pmap_call_on_active_cpus(pmap_t pmap, void (*fn)(void *), void *arg) { - u_int cpuid; + int cpuid, cpu, self; + cpuset_t active_cpus; + sched_pin(); + if (is_kernel_pmap(pmap)) { + smp_rendezvous(NULL, fn, NULL, arg); + goto out; + } + /* Force ASID update on inactive CPUs */ + CPU_FOREACH(cpu) { + if (!CPU_ISSET(cpu, &pmap->pm_active)) + pmap->pm_asid[cpu].gen = 0; + } cpuid = PCPU_GET(cpuid); + /* + * XXX: barrier/locking for active? + * + * Take a snapshot of active here, any further changes are ignored. + * tlb update/invalidate should be harmless on inactive CPUs + */ + active_cpus = pmap->pm_active; + self = CPU_ISSET(cpuid, &active_cpus); + CPU_CLR(cpuid, &active_cpus); + /* Optimize for the case where this cpu is the only active one */ + if (CPU_EMPTY(&active_cpus)) { + if (self) + fn(arg); + } else { + if (self) + CPU_SET(cpuid, &active_cpus); + smp_rendezvous_cpus(active_cpus, NULL, fn, NULL, arg); + } +out: + sched_unpin(); +} +#else /* !SMP */ +static __inline void +pmap_call_on_active_cpus(pmap_t pmap, void (*fn)(void *), void *arg) +{ + int cpuid; - if (pmap == kernel_pmap) { - tlb_invalidate_all(); + if (is_kernel_pmap(pmap)) { + fn(arg); return; } - if (CPU_ISSET(cpuid, &pmap->pm_active)) - tlb_invalidate_all_user(pmap); - else + cpuid = PCPU_GET(cpuid); + if (!CPU_ISSET(cpuid, &pmap->pm_active)) pmap->pm_asid[cpuid].gen = 0; + else + fn(arg); } +#endif /* SMP */ -#ifdef SMP static void pmap_invalidate_all(pmap_t pmap) { - smp_rendezvous(0, pmap_invalidate_all_action, 0, pmap); + pmap_call_on_active_cpus(pmap, + (void (*)(void *))tlb_invalidate_all_user, pmap); } -static void -pmap_invalidate_all_action(void *arg) -{ - - pmap_invalidate_all_local((pmap_t)arg); -} -#else -static void -pmap_invalidate_all(pmap_t pmap) -{ - - pmap_invalidate_all_local(pmap); -} -#endif - -static __inline void -pmap_invalidate_page_local(pmap_t pmap, vm_offset_t va) -{ - u_int cpuid; - - cpuid = PCPU_GET(cpuid); - - if (is_kernel_pmap(pmap)) { - tlb_invalidate_address(pmap, va); - return; - } - if (pmap->pm_asid[cpuid].gen != PCPU_GET(asid_generation)) - return; - else if (!CPU_ISSET(cpuid, &pmap->pm_active)) { - pmap->pm_asid[cpuid].gen = 0; - return; - } - tlb_invalidate_address(pmap, va); -} - -#ifdef SMP struct pmap_invalidate_page_arg { pmap_t pmap; vm_offset_t va; }; static void -pmap_invalidate_page(pmap_t pmap, vm_offset_t va) -{ - struct pmap_invalidate_page_arg arg; - - arg.pmap = pmap; - arg.va = va; - smp_rendezvous(0, pmap_invalidate_page_action, 0, &arg); -} - -static void pmap_invalidate_page_action(void *arg) { struct pmap_invalidate_page_arg *p = arg; - pmap_invalidate_page_local(p->pmap, p->va); + tlb_invalidate_address(p->pmap, p->va); } -#else + static void pmap_invalidate_page(pmap_t pmap, vm_offset_t va) { + struct pmap_invalidate_page_arg arg; - pmap_invalidate_page_local(pmap, va); -} -#endif - -static __inline void -pmap_update_page_local(pmap_t pmap, vm_offset_t va, pt_entry_t pte) -{ - u_int cpuid; - - cpuid = PCPU_GET(cpuid); - - if (is_kernel_pmap(pmap)) { - tlb_update(pmap, va, pte); - return; - } - if (pmap->pm_asid[cpuid].gen != PCPU_GET(asid_generation)) - return; - else if (!CPU_ISSET(cpuid, &pmap->pm_active)) { - pmap->pm_asid[cpuid].gen = 0; - return; - } - tlb_update(pmap, va, pte); + arg.pmap = pmap; + arg.va = va; + pmap_call_on_active_cpus(pmap, pmap_invalidate_page_action, &arg); } -#ifdef SMP struct pmap_update_page_arg { pmap_t pmap; vm_offset_t va; @@ -742,31 +716,23 @@ struct pmap_update_page_arg { }; static void -pmap_update_page(pmap_t pmap, vm_offset_t va, pt_entry_t pte) -{ - struct pmap_update_page_arg arg; - - arg.pmap = pmap; - arg.va = va; - arg.pte = pte; - smp_rendezvous(0, pmap_update_page_action, 0, &arg); -} - -static void pmap_update_page_action(void *arg) { struct pmap_update_page_arg *p = arg; - pmap_update_page_local(p->pmap, p->va, p->pte); + tlb_update(p->pmap, p->va, p->pte); } -#else + static void pmap_update_page(pmap_t pmap, vm_offset_t va, pt_entry_t pte) { + struct pmap_update_page_arg arg; - pmap_update_page_local(pmap, va, pte); + arg.pmap = pmap; + arg.va = va; + arg.pte = pte; + pmap_call_on_active_cpus(pmap, pmap_update_page_action, &arg); } -#endif /* * Routine: pmap_extract @@ -3213,7 +3179,7 @@ pmap_emulate_modified(pmap_t pmap, vm_of #ifdef SMP /* It is possible that some other CPU changed m-bit */ if (!pte_test(pte, PTE_V) || pte_test(pte, PTE_D)) { - pmap_update_page_local(pmap, va, *pte); + tlb_update(pmap, va, *pte); PMAP_UNLOCK(pmap); return (0); } @@ -3227,7 +3193,7 @@ pmap_emulate_modified(pmap_t pmap, vm_of return (1); } pte_set(pte, PTE_D); - pmap_update_page_local(pmap, va, *pte); + tlb_update(pmap, va, *pte); pa = TLBLO_PTE_TO_PA(*pte); if (!page_is_managed(pa)) panic("pmap_emulate_modified: unmanaged page");help
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CA%2B7sy7Aj89iTTD=7ywiatyBkBW2OjborNydZTf8dgXLNjY8G1w>