diff options
author | Tony Luck <tony.luck@intel.com> | 2005-10-28 15:27:36 -0700 |
---|---|---|
committer | Tony Luck <tony.luck@intel.com> | 2005-10-28 15:27:36 -0700 |
commit | 8496f2a4513edbc0c1d40496f1b0868dfed36eb0 (patch) | |
tree | 18a772d7a4c42ec325539b52d2053dabc99c460a /arch/ia64 | |
parent | 2d8f6a521908e3563478347aecad4ab20dc48155 (diff) | |
parent | c1902aae322952f8726469a6657df7b9d5c794fe (diff) |
Pull fix-slow-tlb-purge into release branch
Diffstat (limited to 'arch/ia64')
-rw-r--r-- | arch/ia64/mm/tlb.c | 16 | ||||
-rw-r--r-- | arch/ia64/sn/kernel/sn2/sn2_smp.c | 31 |
2 files changed, 28 insertions, 19 deletions
diff --git a/arch/ia64/mm/tlb.c b/arch/ia64/mm/tlb.c index 987fb754d6ad..c93e0f2b5fea 100644 --- a/arch/ia64/mm/tlb.c +++ b/arch/ia64/mm/tlb.c @@ -87,10 +87,15 @@ wrap_mmu_context (struct mm_struct *mm) } void -ia64_global_tlb_purge (unsigned long start, unsigned long end, unsigned long nbits) +ia64_global_tlb_purge (struct mm_struct *mm, unsigned long start, unsigned long end, unsigned long nbits) { static DEFINE_SPINLOCK(ptcg_lock); + if (mm != current->active_mm) { + flush_tlb_all(); + return; + } + /* HW requires global serialization of ptc.ga. */ spin_lock(&ptcg_lock); { @@ -136,15 +141,12 @@ flush_tlb_range (struct vm_area_struct *vma, unsigned long start, unsigned long unsigned long size = end - start; unsigned long nbits; +#ifndef CONFIG_SMP if (mm != current->active_mm) { - /* this does happen, but perhaps it's not worth optimizing for? */ -#ifdef CONFIG_SMP - flush_tlb_all(); -#else mm->context = 0; -#endif return; } +#endif nbits = ia64_fls(size + 0xfff); while (unlikely (((1UL << nbits) & purge.mask) == 0) && (nbits < purge.max_bits)) @@ -154,7 +156,7 @@ flush_tlb_range (struct vm_area_struct *vma, unsigned long start, unsigned long start &= ~((1UL << nbits) - 1); # ifdef CONFIG_SMP - platform_global_tlb_purge(start, end, nbits); + platform_global_tlb_purge(mm, start, end, nbits); # else do { ia64_ptcl(start, (nbits<<2)); diff --git a/arch/ia64/sn/kernel/sn2/sn2_smp.c b/arch/ia64/sn/kernel/sn2/sn2_smp.c index 0a4ee50c302f..49b530c39a42 100644 --- a/arch/ia64/sn/kernel/sn2/sn2_smp.c +++ b/arch/ia64/sn/kernel/sn2/sn2_smp.c @@ -177,6 +177,7 @@ void sn_tlb_migrate_finish(struct mm_struct *mm) /** * sn2_global_tlb_purge - globally purge translation cache of virtual address range + * @mm: mm_struct containing virtual address range * @start: start of virtual address range * @end: end of virtual address range * @nbits: specifies number of bytes to purge per instruction (num = 1<<(nbits & 0xfc)) @@ -188,21 +189,22 @@ void sn_tlb_migrate_finish(struct mm_struct *mm) * - cpu_vm_mask is a bit mask that indicates which cpus have loaded the context. * - cpu_vm_mask is converted into a nodemask of the nodes containing the * cpus in cpu_vm_mask. - * - if only one bit is set in cpu_vm_mask & it is the current cpu, - * then only the local TLB needs to be flushed. This flushing can be done - * using ptc.l. This is the common case & avoids the global spinlock. + * - if only one bit is set in cpu_vm_mask & it is the current cpu & the + * process is purging its own virtual address range, then only the + * local TLB needs to be flushed. This flushing can be done using + * ptc.l. This is the common case & avoids the global spinlock. * - if multiple cpus have loaded the context, then flushing has to be * done with ptc.g/MMRs under protection of the global ptc_lock. */ void -sn2_global_tlb_purge(unsigned long start, unsigned long end, - unsigned long nbits) +sn2_global_tlb_purge(struct mm_struct *mm, unsigned long start, + unsigned long end, unsigned long nbits) { int i, opt, shub1, cnode, mynasid, cpu, lcpu = 0, nasid, flushed = 0; + int mymm = (mm == current->active_mm); volatile unsigned long *ptc0, *ptc1; - unsigned long itc, itc2, flags, data0 = 0, data1 = 0; - struct mm_struct *mm = current->active_mm; + unsigned long itc, itc2, flags, data0 = 0, data1 = 0, rr_value; short nasids[MAX_NUMNODES], nix; nodemask_t nodes_flushed; @@ -216,9 +218,12 @@ sn2_global_tlb_purge(unsigned long start, unsigned long end, i++; } + if (i == 0) + return; + preempt_disable(); - if (likely(i == 1 && lcpu == smp_processor_id())) { + if (likely(i == 1 && lcpu == smp_processor_id() && mymm)) { do { ia64_ptcl(start, nbits << 2); start += (1UL << nbits); @@ -229,7 +234,7 @@ sn2_global_tlb_purge(unsigned long start, unsigned long end, return; } - if (atomic_read(&mm->mm_users) == 1) { + if (atomic_read(&mm->mm_users) == 1 && mymm) { flush_tlb_mm(mm); __get_cpu_var(ptcstats).change_rid++; preempt_enable(); @@ -241,11 +246,13 @@ sn2_global_tlb_purge(unsigned long start, unsigned long end, for_each_node_mask(cnode, nodes_flushed) nasids[nix++] = cnodeid_to_nasid(cnode); + rr_value = (mm->context << 3) | REGION_NUMBER(start); + shub1 = is_shub1(); if (shub1) { data0 = (1UL << SH1_PTC_0_A_SHFT) | (nbits << SH1_PTC_0_PS_SHFT) | - ((ia64_get_rr(start) >> 8) << SH1_PTC_0_RID_SHFT) | + (rr_value << SH1_PTC_0_RID_SHFT) | (1UL << SH1_PTC_0_START_SHFT); ptc0 = (long *)GLOBAL_MMR_PHYS_ADDR(0, SH1_PTC_0); ptc1 = (long *)GLOBAL_MMR_PHYS_ADDR(0, SH1_PTC_1); @@ -254,7 +261,7 @@ sn2_global_tlb_purge(unsigned long start, unsigned long end, (nbits << SH2_PTC_PS_SHFT) | (1UL << SH2_PTC_START_SHFT); ptc0 = (long *)GLOBAL_MMR_PHYS_ADDR(0, SH2_PTC + - ((ia64_get_rr(start) >> 8) << SH2_PTC_RID_SHFT) ); + (rr_value << SH2_PTC_RID_SHFT)); ptc1 = NULL; } @@ -275,7 +282,7 @@ sn2_global_tlb_purge(unsigned long start, unsigned long end, data0 = (data0 & ~SH2_PTC_ADDR_MASK) | (start & SH2_PTC_ADDR_MASK); for (i = 0; i < nix; i++) { nasid = nasids[i]; - if ((!(sn2_ptctest & 3)) && unlikely(nasid == mynasid)) { + if ((!(sn2_ptctest & 3)) && unlikely(nasid == mynasid && mymm)) { ia64_ptcga(start, nbits << 2); ia64_srlz_i(); } else { |