| Message ID | 1407221289-19390-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Headers | show |
"Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com> writes: > If we know that user address space has never executed on other cpus > we could use tlbiel. > > Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> Now checking against flush_hash_page, I am wondering whether I need to handle transcational memory in case of tlbiel ? Michael, can you let me know if this is needed ? -aneesh
On Wed, 2014-08-06 at 20:16 +0530, Aneesh Kumar K.V wrote: > "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com> writes: > > > If we know that user address space has never executed on other cpus > > we could use tlbiel. > > > > Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> > > Now checking against flush_hash_page, I am wondering whether I need to handle > transcational memory in case of tlbiel ? Michael, can you let me know if > this is needed ? Also, beware that we have code that sets "tlbiel" in the page definition array to 0 on all but 4k and 64k pages... Cheers, Ben.
On Wed, 2014-08-06 at 20:16 +0530, Aneesh Kumar K.V wrote: > "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com> writes: > > > If we know that user address space has never executed on other cpus > > we could use tlbiel. > > > > Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> > > Now checking against flush_hash_page, I am wondering whether I need to handle > transcational memory in case of tlbiel ? Michael, can you let me know if > this is needed ? Yeah, you do need to add something. local tlbie won't abort the transaction (broadcast will abort), so you need to do a explicit tabort if there is an active transaction, to make sure the transaction is doomed. Mikey
Benjamin Herrenschmidt <benh@kernel.crashing.org> writes: > On Wed, 2014-08-06 at 20:16 +0530, Aneesh Kumar K.V wrote: >> "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com> writes: >> >> > If we know that user address space has never executed on other cpus >> > we could use tlbiel. >> > >> > Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> >> >> Now checking against flush_hash_page, I am wondering whether I need to handle >> transcational memory in case of tlbiel ? Michael, can you let me know if >> this is needed ? > > Also, beware that we have code that sets "tlbiel" in the page definition > array to 0 on all but 4k and 64k pages... > Ok we use base page size to determine whether we can use tlbiel or not. static inline void tlbie(unsigned long vpn, int psize, int apsize, int ssize, int local) { .... if (use_local) use_local = mmu_psize_defs[psize].tlbiel; With that we should be ok, because we are using 4k and 64k base page size and we have /* * We don't know for sure what's up with tlbiel, so * for now we only set it for 4K and 64K pages */ if (base_idx == MMU_PAGE_4K || base_idx == MMU_PAGE_64K) def->tlbiel = 1; else def->tlbiel = 0; -aneesh
diff --git a/arch/powerpc/include/asm/machdep.h b/arch/powerpc/include/asm/machdep.h index 8dcb721d03d8..fe7a11ae9d9c 100644 --- a/arch/powerpc/include/asm/machdep.h +++ b/arch/powerpc/include/asm/machdep.h @@ -60,7 +60,7 @@ struct machdep_calls { void (*hugepage_invalidate)(unsigned long vsid, unsigned long addr, unsigned char *hpte_slot_array, - int psize, int ssize); + int psize, int ssize, int local); /* special for kexec, to be called in real mode, linear mapping is * destroyed as well */ void (*hpte_clear_all)(void); diff --git a/arch/powerpc/mm/hash_native_64.c b/arch/powerpc/mm/hash_native_64.c index afc0a8295f84..88de8a869c98 100644 --- a/arch/powerpc/mm/hash_native_64.c +++ b/arch/powerpc/mm/hash_native_64.c @@ -415,7 +415,7 @@ static void native_hpte_invalidate(unsigned long slot, unsigned long vpn, static void native_hugepage_invalidate(unsigned long vsid, unsigned long addr, unsigned char *hpte_slot_array, - int psize, int ssize) + int psize, int ssize, int local) { int i; struct hash_pte *hptep; @@ -461,7 +461,7 @@ static void native_hugepage_invalidate(unsigned long vsid, * instruction compares entry_VA in tlb with the VA specified * here */ - tlbie(vpn, psize, actual_psize, ssize, 0); + tlbie(vpn, psize, actual_psize, ssize, local); } local_irq_restore(flags); } diff --git a/arch/powerpc/mm/hugepage-hash64.c b/arch/powerpc/mm/hugepage-hash64.c index 5f5e6328c21c..3e87005cfb79 100644 --- a/arch/powerpc/mm/hugepage-hash64.c +++ b/arch/powerpc/mm/hugepage-hash64.c @@ -19,7 +19,8 @@ #include <asm/machdep.h> static void invalidate_old_hpte(unsigned long vsid, unsigned long addr, - pmd_t *pmdp, unsigned int psize, int ssize) + pmd_t *pmdp, unsigned int psize, + int ssize, int local) { int i, max_hpte_count, valid; unsigned long s_addr; @@ -38,7 +39,7 @@ static void invalidate_old_hpte(unsigned long vsid, unsigned long addr, if (ppc_md.hugepage_invalidate) return ppc_md.hugepage_invalidate(vsid, s_addr, hpte_slot_array, - psize, ssize); + psize, ssize, local); /* * No bluk hpte removal support, invalidate each entry */ @@ -64,7 +65,7 @@ static void invalidate_old_hpte(unsigned long vsid, unsigned long addr, slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; slot += hidx & _PTEIDX_GROUP_IX; ppc_md.hpte_invalidate(slot, vpn, psize, - MMU_PAGE_16M, ssize, 0); + MMU_PAGE_16M, ssize, local); } } @@ -145,7 +146,8 @@ int __hash_page_thp(unsigned long ea, unsigned long access, unsigned long vsid, * hash page table entries. */ if ((old_pmd & _PAGE_HASHPTE) && !(old_pmd & _PAGE_COMBO)) - invalidate_old_hpte(vsid, ea, pmdp, MMU_PAGE_64K, ssize); + invalidate_old_hpte(vsid, ea, pmdp, MMU_PAGE_64K, + ssize, local); } valid = hpte_valid(hpte_slot_array, index); diff --git a/arch/powerpc/mm/pgtable_64.c b/arch/powerpc/mm/pgtable_64.c index ecc4079ca56c..b1ea87b34f79 100644 --- a/arch/powerpc/mm/pgtable_64.c +++ b/arch/powerpc/mm/pgtable_64.c @@ -731,10 +731,11 @@ void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address, void hpte_do_hugepage_flush(struct mm_struct *mm, unsigned long addr, pmd_t *pmdp, unsigned long old_pmd) { - int ssize, i; + int ssize, i, local = 0; unsigned long s_addr; int max_hpte_count; unsigned int psize, valid; + const struct cpumask *tmp; unsigned char *hpte_slot_array; unsigned long hidx, vpn, vsid, hash, shift, slot; @@ -770,10 +771,14 @@ void hpte_do_hugepage_flush(struct mm_struct *mm, unsigned long addr, ssize = mmu_kernel_ssize; } + tmp = cpumask_of(smp_processor_id()); + if (cpumask_equal(mm_cpumask(mm), tmp)) + local = 1; + if (ppc_md.hugepage_invalidate) return ppc_md.hugepage_invalidate(vsid, s_addr, hpte_slot_array, - psize, ssize); + psize, ssize, local); /* * No bluk hpte removal support, invalidate each entry */ @@ -799,7 +804,7 @@ void hpte_do_hugepage_flush(struct mm_struct *mm, unsigned long addr, slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; slot += hidx & _PTEIDX_GROUP_IX; ppc_md.hpte_invalidate(slot, vpn, psize, - MMU_PAGE_16M, ssize, 0); + MMU_PAGE_16M, ssize, local); } } diff --git a/arch/powerpc/platforms/pseries/lpar.c b/arch/powerpc/platforms/pseries/lpar.c index ccf6f162f69c..c543c20ffabb 100644 --- a/arch/powerpc/platforms/pseries/lpar.c +++ b/arch/powerpc/platforms/pseries/lpar.c @@ -433,7 +433,7 @@ static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot, static void pSeries_lpar_hugepage_invalidate(unsigned long vsid, unsigned long addr, unsigned char *hpte_slot_array, - int psize, int ssize) + int psize, int ssize, int local) { int i, index = 0; unsigned long s_addr = addr;
If we know that user address space has never executed on other cpus we could use tlbiel. Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> --- arch/powerpc/include/asm/machdep.h | 2 +- arch/powerpc/mm/hash_native_64.c | 4 ++-- arch/powerpc/mm/hugepage-hash64.c | 10 ++++++---- arch/powerpc/mm/pgtable_64.c | 11 ++++++++--- arch/powerpc/platforms/pseries/lpar.c | 2 +- 5 files changed, 18 insertions(+), 11 deletions(-)