From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: From: Leonardo Bras Subject: [PATCH v6 03/11] powerpc/mm: Adds arch-specificic functions to track lockless pgtable walks Date: Thu, 6 Feb 2020 00:08:52 -0300 Message-ID: <20200206030900.147032-4-leonardo@linux.ibm.com> In-Reply-To: <20200206030900.147032-1-leonardo@linux.ibm.com> References: <20200206030900.147032-1-leonardo@linux.ibm.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit To: Benjamin Herrenschmidt , Paul Mackerras , Michael Ellerman , Arnd Bergmann , Andrew Morton , "Aneesh Kumar K.V" , Nicholas Piggin , Christophe Leroy , Steven Price , Robin Murphy , Leonardo Bras , Mahesh Salgaonkar , Balbir Singh , Reza Arbab , Thomas Gleixner , Allison Randal , Greg Kroah-Hartman , Mike Rapoport , Michal Suchanek Cc: linuxppc-dev@lists.ozlabs.org, linux-kernel@vger.kernel.org, kvm-ppc@vger.kernel.org, linux-arch@vger.kernel.org, linux-mm@kvack.org List-ID: Message-ID: <20200206030852.9DM-Xt2FBimT8jeAHg5YbPlKWWbGfJH92t8zWgQsCyo@z> On powerpc, we need to do some lockless pagetable walks from functions that already have disabled interrupts, specially from real mode with MSR[EE=0]. In these contexts, disabling/enabling interrupts can be very troubling. So, this arch-specific implementation features functions with an extra argument that allows interrupt enable/disable to be skipped: __begin_lockless_pgtbl_walk() and __end_lockless_pgtbl_walk(). Functions similar to the generic ones are also exported, by calling the above functions with parameter {en,dis}able_irq = true. Signed-off-by: Leonardo Bras --- arch/powerpc/include/asm/book3s/64/pgtable.h | 6 ++ arch/powerpc/mm/book3s64/pgtable.c | 86 +++++++++++++++++++- 2 files changed, 91 insertions(+), 1 deletion(-) diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h b/arch/powerpc/include/asm/book3s/64/pgtable.h index 201a69e6a355..78f6ffb1bb3e 100644 --- a/arch/powerpc/include/asm/book3s/64/pgtable.h +++ b/arch/powerpc/include/asm/book3s/64/pgtable.h @@ -1375,5 +1375,11 @@ static inline bool pgd_is_leaf(pgd_t pgd) return !!(pgd_raw(pgd) & cpu_to_be64(_PAGE_PTE)); } +#define __HAVE_ARCH_LOCKLESS_PGTBL_WALK_CONTROL +unsigned long begin_lockless_pgtbl_walk(void); +unsigned long __begin_lockless_pgtbl_walk(bool disable_irq); +void end_lockless_pgtbl_walk(unsigned long irq_mask); +void __end_lockless_pgtbl_walk(unsigned long irq_mask, bool enable_irq); + #endif /* __ASSEMBLY__ */ #endif /* _ASM_POWERPC_BOOK3S_64_PGTABLE_H_ */ diff --git a/arch/powerpc/mm/book3s64/pgtable.c b/arch/powerpc/mm/book3s64/pgtable.c index 2bf7e1b4fd82..535613030363 100644 --- a/arch/powerpc/mm/book3s64/pgtable.c +++ b/arch/powerpc/mm/book3s64/pgtable.c @@ -82,6 +82,7 @@ static void do_nothing(void *unused) { } + /* * Serialize against find_current_mm_pte which does lock-less * lookup in page tables with local interrupts disabled. For huge pages @@ -98,6 +99,89 @@ void serialize_against_pte_lookup(struct mm_struct *mm) smp_call_function_many(mm_cpumask(mm), do_nothing, NULL, 1); } +/* begin_lockless_pgtbl_walk: Must be inserted before a function call that does + * lockless pagetable walks, such as __find_linux_pte(). + * This version allows setting disable_irq=false, so irqs are not touched, which + * is quite useful for running when ints are already disabled (like real-mode) + */ +inline +unsigned long __begin_lockless_pgtbl_walk(bool disable_irq) +{ + unsigned long irq_mask = 0; + + /* + * Interrupts must be disabled during the lockless page table walk. + * That's because the deleting or splitting involves flushing TLBs, + * which in turn issues interrupts, that will block when disabled. + * + * When this function is called from realmode with MSR[EE=0], + * it's not needed to touch irq, since it's already disabled. + */ + if (disable_irq) + local_irq_save(irq_mask); + + /* + * This memory barrier pairs with any code that is either trying to + * delete page tables, or split huge pages. Without this barrier, + * the page tables could be read speculatively outside of interrupt + * disabling or reference counting. + */ + smp_mb(); + + return irq_mask; +} +EXPORT_SYMBOL(__begin_lockless_pgtbl_walk); + +/* begin_lockless_pgtbl_walk: Must be inserted before a function call that does + * lockless pagetable walks, such as __find_linux_pte(). + * This version is used by generic code, and always assume irqs will be disabled + */ +unsigned long begin_lockless_pgtbl_walk(void) +{ + return __begin_lockless_pgtbl_walk(true); +} +EXPORT_SYMBOL(begin_lockless_pgtbl_walk); + +/* + * __end_lockless_pgtbl_walk: Must be inserted after the last use of a pointer + * returned by a lockless pagetable walk, such as __find_linux_pte() + * This version allows setting enable_irq=false, so irqs are not touched, which + * is quite useful for running when ints are already disabled (like real-mode) + */ +inline void __end_lockless_pgtbl_walk(unsigned long irq_mask, bool enable_irq) +{ + /* + * This memory barrier pairs with any code that is either trying to + * delete page tables, or split huge pages. Without this barrier, + * the page tables could be read speculatively outside of interrupt + * disabling or reference counting. + */ + smp_mb(); + + /* + * Interrupts must be disabled during the lockless page table walk. + * That's because the deleting or splitting involves flushing TLBs, + * which in turn issues interrupts, that will block when disabled. + * + * When this function is called from realmode with MSR[EE=0], + * it's not needed to touch irq, since it's already disabled. + */ + if (enable_irq) + local_irq_restore(irq_mask); +} +EXPORT_SYMBOL(__end_lockless_pgtbl_walk); + +/* + * end_lockless_pgtbl_walk: Must be inserted after the last use of a pointer + * returned by a lockless pagetable walk, such as __find_linux_pte() + * This version is used by generic code, and always assume irqs will be enabled + */ +void end_lockless_pgtbl_walk(unsigned long irq_mask) +{ + __end_lockless_pgtbl_walk(irq_mask, true); +} +EXPORT_SYMBOL(end_lockless_pgtbl_walk); + /* * We use this to invalidate a pmdp entry before switching from a * hugepte to regular pmd entry. @@ -487,7 +571,7 @@ static int __init setup_disable_tlbie(char *str) tlbie_capable = false; tlbie_enabled = false; - return 1; + return 1; } __setup("disable_tlbie", setup_disable_tlbie); -- 2.24.1