Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > linux.kernel > #1456238 > unrolled thread

arm64: Implement IPI based TLB invalidation

Started byMatthias Brugger <mbrugger@suse.com>
First post2016-08-04 11:20 +0200
Last post2016-08-04 23:50 +0200
Articles 4 — 2 participants

Back to article view | Back to linux.kernel


Contents

  arm64: Implement IPI based TLB invalidation Matthias Brugger <mbrugger@suse.com> - 2016-08-04 11:20 +0200
    [PATCH 2/4] arm64: Implement IPI based TLB invalidation Matthias Brugger <mbrugger@suse.com> - 2016-08-04 11:20 +0200
    [PATCH 1/4] arm64: insn: Do not disable irqs during patching Matthias Brugger <mbrugger@suse.com> - 2016-08-04 11:20 +0200
    Re: [PATCH] arm64: Add workaround for Cavium erratum 26026 David Daney <ddaney.cavm@gmail.com> - 2016-08-04 23:50 +0200

#1456238 — arm64: Implement IPI based TLB invalidation

FromMatthias Brugger <mbrugger@suse.com>
Date2016-08-04 11:20 +0200
Subjectarm64: Implement IPI based TLB invalidation
Message-ID<s2mYN-6W6-7@gated-at.bofh.it>
This patch set adds a workaround for arm64 if the hardware didn't
implement the TLB invalidation via broadcast TLBI properly.

[toc] | [next] | [standalone]


#1456246 — [PATCH 2/4] arm64: Implement IPI based TLB invalidation

FromMatthias Brugger <mbrugger@suse.com>
Date2016-08-04 11:20 +0200
Subject[PATCH 2/4] arm64: Implement IPI based TLB invalidation
Message-ID<s2mYO-6W6-43@gated-at.bofh.it>
In reply to#1456238
Hardware may lack a sane implementation of TLB invalidation
using broadcast TLBI command. Add a capability to enable TLB
invalidation using IPI.

Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: Robert Richter <rrichter@cavium.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Matthias Brugger <mbrugger@suse.com>
---
 arch/arm64/include/asm/cpufeature.h |  3 +-
 arch/arm64/include/asm/tlbflush.h   | 94 ++++++++++++++++++++++++++++++++-----
 arch/arm64/mm/flush.c               | 46 ++++++++++++++++++
 3 files changed, 129 insertions(+), 14 deletions(-)

diff --git a/arch/arm64/include/asm/cpufeature.h b/arch/arm64/include/asm/cpufeature.h
index 49dd1bd..c4bf72b 100644
--- a/arch/arm64/include/asm/cpufeature.h
+++ b/arch/arm64/include/asm/cpufeature.h
@@ -36,8 +36,9 @@
 #define ARM64_HAS_VIRT_HOST_EXTN		11
 #define ARM64_WORKAROUND_CAVIUM_27456		12
 #define ARM64_HAS_32BIT_EL0			13
+#define ARM64_HAS_NO_BCAST_TLBI			14
 
-#define ARM64_NCAPS				14
+#define ARM64_NCAPS				15
 
 #ifndef __ASSEMBLY__
 
diff --git a/arch/arm64/include/asm/tlbflush.h b/arch/arm64/include/asm/tlbflush.h
index b460ae2..edc5495 100644
--- a/arch/arm64/include/asm/tlbflush.h
+++ b/arch/arm64/include/asm/tlbflush.h
@@ -71,7 +71,10 @@ static inline void local_flush_tlb_all(void)
 	isb();
 }
 
-static inline void flush_tlb_all(void)
+void __flush_tlb_all_ipi(void);
+void __flush_tlb_mm_ipi(struct mm_struct *mm);
+
+static inline void __flush_tlb_all_tlbi(void)
 {
 	dsb(ishst);
 	asm("tlbi	vmalle1is");
@@ -79,7 +82,17 @@ static inline void flush_tlb_all(void)
 	isb();
 }
 
-static inline void flush_tlb_mm(struct mm_struct *mm)
+static inline void flush_tlb_all(void)
+{
+	if (cpus_have_cap(ARM64_HAS_NO_BCAST_TLBI)) {
+		__flush_tlb_all_ipi();
+		return;
+	}
+
+	__flush_tlb_all_tlbi();
+}
+
+static inline void __flush_tlb_mm_tlbi(struct mm_struct *mm)
 {
 	unsigned long asid = ASID(mm) << 48;
 
@@ -88,8 +101,18 @@ static inline void flush_tlb_mm(struct mm_struct *mm)
 	dsb(ish);
 }
 
-static inline void flush_tlb_page(struct vm_area_struct *vma,
-				  unsigned long uaddr)
+static inline void flush_tlb_mm(struct mm_struct *mm)
+{
+	if (cpus_have_cap(ARM64_HAS_NO_BCAST_TLBI)) {
+		__flush_tlb_mm_ipi(mm);
+		return;
+	}
+
+	__flush_tlb_mm_tlbi(mm);
+}
+
+static inline void __flush_tlb_page_tlbi(struct vm_area_struct *vma,
+		unsigned long uaddr)
 {
 	unsigned long addr = uaddr >> 12 | (ASID(vma->vm_mm) << 48);
 
@@ -98,15 +121,26 @@ static inline void flush_tlb_page(struct vm_area_struct *vma,
 	dsb(ish);
 }
 
+static inline void flush_tlb_page(struct vm_area_struct *vma,
+		unsigned long uaddr)
+{
+	if (cpus_have_cap(ARM64_HAS_NO_BCAST_TLBI)) {
+		__flush_tlb_mm_ipi(vma->vm_mm);
+		return;
+	}
+
+	__flush_tlb_page_tlbi(vma, uaddr);
+}
+
 /*
  * This is meant to avoid soft lock-ups on large TLB flushing ranges and not
  * necessarily a performance improvement.
  */
 #define MAX_TLB_RANGE	(1024UL << PAGE_SHIFT)
 
-static inline void __flush_tlb_range(struct vm_area_struct *vma,
-				     unsigned long start, unsigned long end,
-				     bool last_level)
+static inline void __flush_tlb_range_tlbi(struct vm_area_struct *vma,
+		unsigned long start, unsigned long end,
+		bool last_level)
 {
 	unsigned long asid = ASID(vma->vm_mm) << 48;
 	unsigned long addr;
@@ -129,13 +163,26 @@ static inline void __flush_tlb_range(struct vm_area_struct *vma,
 	dsb(ish);
 }
 
+static inline void __flush_tlb_range(struct vm_area_struct *vma,
+		unsigned long start, unsigned long end,
+		bool last_level)
+{
+	if (cpus_have_cap(ARM64_HAS_NO_BCAST_TLBI)) {
+		__flush_tlb_mm_ipi(vma->vm_mm);
+		return;
+	}
+
+	__flush_tlb_range_tlbi(vma, start, end, last_level);
+}
+
 static inline void flush_tlb_range(struct vm_area_struct *vma,
-				   unsigned long start, unsigned long end)
+		unsigned long start, unsigned long end)
 {
 	__flush_tlb_range(vma, start, end, false);
 }
 
-static inline void flush_tlb_kernel_range(unsigned long start, unsigned long end)
+static inline void __flush_tlb_kernel_range_tlbi(unsigned long start,
+		unsigned long end)
 {
 	unsigned long addr;
 
@@ -154,17 +201,38 @@ static inline void flush_tlb_kernel_range(unsigned long start, unsigned long end
 	isb();
 }
 
+static inline void flush_tlb_kernel_range(unsigned long start, unsigned long end)
+{
+	if (cpus_have_cap(ARM64_HAS_NO_BCAST_TLBI)) {
+		__flush_tlb_all_ipi();
+		return;
+	}
+
+	__flush_tlb_kernel_range_tlbi(start, end);
+}
+
+static inline void __flush_tlb_pgtable_tlbi(struct mm_struct *mm,
+		unsigned long uaddr)
+{
+	unsigned long addr = uaddr >> 12 | (ASID(mm) << 48);
+
+	asm("tlbi	vae1is, %0" : : "r" (addr));
+	dsb(ish);
+}
+
 /*
  * Used to invalidate the TLB (walk caches) corresponding to intermediate page
  * table levels (pgd/pud/pmd).
  */
 static inline void __flush_tlb_pgtable(struct mm_struct *mm,
-				       unsigned long uaddr)
+		unsigned long uaddr)
 {
-	unsigned long addr = uaddr >> 12 | (ASID(mm) << 48);
+	if (cpus_have_cap(ARM64_HAS_NO_BCAST_TLBI)) {
+		__flush_tlb_mm_ipi(mm);
+		return;
+	}
 
-	asm("tlbi	vae1is, %0" : : "r" (addr));
-	dsb(ish);
+	__flush_tlb_pgtable_tlbi(mm, uaddr);
 }
 
 #endif
diff --git a/arch/arm64/mm/flush.c b/arch/arm64/mm/flush.c
index 43a76b0..402036a 100644
--- a/arch/arm64/mm/flush.c
+++ b/arch/arm64/mm/flush.c
@@ -27,6 +27,24 @@
 
 #include "mm.h"
 
+static void flush_tlb_local(void *info)
+{
+	local_flush_tlb_all();
+}
+
+static void flush_tlb_mm_local(void *info)
+{
+	unsigned long asid = (unsigned long)info;
+
+	asm volatile("\n"
+	"  dsb     nshst\n"
+	"  tlbi    aside1, %0\n"
+	"  dsb     nsh\n"
+	"  isb     sy"
+	: : "r" (asid)
+	);
+}
+
 void flush_cache_range(struct vm_area_struct *vma, unsigned long start,
 		       unsigned long end)
 {
@@ -90,6 +108,34 @@ void flush_dcache_page(struct page *page)
 }
 EXPORT_SYMBOL(flush_dcache_page);
 
+void __flush_tlb_mm_ipi(struct mm_struct *mm)
+{
+	unsigned long asid;
+
+	if (!mm) {
+		flush_tlb_all();
+	} else {
+		asid = ASID(mm) << 48;
+		/* Make sure page table modifications are visible. */
+		dsb(ishst);
+		/* IPI to all CPUs to do local flush. */
+		on_each_cpu(flush_tlb_mm_local, (void *)asid, 1);
+	}
+}
+EXPORT_SYMBOL(__flush_tlb_mm_ipi);
+
+void __flush_tlb_all_ipi(void)
+{
+	/* Make sure page table modifications are visible. */
+	dsb(ishst);
+	if (num_online_cpus() <= 1)
+		local_flush_tlb_all();
+	else
+		/* IPI to all CPUs to do local flush. */
+		on_each_cpu(flush_tlb_local, NULL, 1);
+}
+EXPORT_SYMBOL(__flush_tlb_all_ipi);
+
 /*
  * Additional functions defined in assembly.
  */
-- 
2.6.6

[toc] | [prev] | [next] | [standalone]


#1456248 — [PATCH 1/4] arm64: insn: Do not disable irqs during patching

FromMatthias Brugger <mbrugger@suse.com>
Date2016-08-04 11:20 +0200
Subject[PATCH 1/4] arm64: insn: Do not disable irqs during patching
Message-ID<s2mYO-6W6-51@gated-at.bofh.it>
In reply to#1456238
From: Robert Richter <rrichter@cavium.com>

__aarch64_insn_write() is always called with interrupts enabled. Thus,
there is no need to use an irqsave variant for the spin lock.

This change should also address the fix of:

commit abffa6f3b157 ("arm64: convert patch_lock to raw lock")

We need to enable interrupts to allow cpu sync for code patching
using smp_call_function*().

Signed-off-by: Robert Richter <rrichter@cavium.com>
Signed-off-by: Matthias Brugger <mbrugger@suse.com>
---
 arch/arm64/kernel/insn.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/arch/arm64/kernel/insn.c b/arch/arm64/kernel/insn.c
index 63f9432..138bd8a 100644
--- a/arch/arm64/kernel/insn.c
+++ b/arch/arm64/kernel/insn.c
@@ -86,7 +86,7 @@ bool aarch64_insn_is_branch_imm(u32 insn)
 		aarch64_insn_is_bcond(insn));
 }
 
-static DEFINE_RAW_SPINLOCK(patch_lock);
+static DEFINE_SPINLOCK(patch_lock);
 
 static void __kprobes *patch_map(void *addr, int fixmap)
 {
@@ -129,16 +129,15 @@ int __kprobes aarch64_insn_read(void *addr, u32 *insnp)
 static int __kprobes __aarch64_insn_write(void *addr, u32 insn)
 {
 	void *waddr = addr;
-	unsigned long flags = 0;
 	int ret;
 
-	raw_spin_lock_irqsave(&patch_lock, flags);
+	spin_lock(&patch_lock);
 	waddr = patch_map(addr, FIX_TEXT_POKE0);
 
 	ret = probe_kernel_write(waddr, &insn, AARCH64_INSN_SIZE);
 
 	patch_unmap(FIX_TEXT_POKE0);
-	raw_spin_unlock_irqrestore(&patch_lock, flags);
+	spin_unlock(&patch_lock);
 
 	return ret;
 }
-- 
2.6.6

[toc] | [prev] | [next] | [standalone]


#1456748 — Re: [PATCH] arm64: Add workaround for Cavium erratum 26026

FromDavid Daney <ddaney.cavm@gmail.com>
Date2016-08-04 23:50 +0200
SubjectRe: [PATCH] arm64: Add workaround for Cavium erratum 26026
Message-ID<s2yGB-6Y1-7@gated-at.bofh.it>
In reply to#1456238
On 08/04/2016 01:57 PM, Robert Richter wrote:
> The patch below is on top of Matthias' patch series:
>
>   arm64: Implement IPI based TLB invalidation
>
> The series is used to enable a workaround for Cavium ThunderX pass 1.x
> systems.


Where are the rest of the patches in the series?  I would have expected 
something like "[PATCH 1/X] arm64:..." with X being greater than one.

>
> -Robert
>
>
>
>  From abb99ee83473d9ecffb4fdaae9c69435ca670bc8 Mon Sep 17 00:00:00 2001
> From: Robert Richter <rrichter@cavium.com>
> Date: Fri, 29 Jul 2016 09:40:04 +0200
> Subject: [PATCH] arm64: Add workaround for Cavium erratum 26026
>
> STX may return a wrong status value if the store was successful.
>
> This may happen on ThunderX T88 pass 1.x cpus if a broadcast TLBI is
> executed on another cpu in parallel to an STX. As a result atomic or
> non-blocking implementations can behave incorrectly. Use IPIs that
> call local TLBIs on other cpus to avoid this.
>
> Applies to ThunderX T88 pass 1.x cpus.
>
> Signed-off-by: Robert Richter <rrichter@cavium.com>
> ---
>   Documentation/arm64/silicon-errata.txt |  1 +
>   arch/arm64/Kconfig                     | 14 ++++++++++++++
>   arch/arm64/kernel/cpu_errata.c         |  8 ++++++++
>   3 files changed, 23 insertions(+)
>
> diff --git a/Documentation/arm64/silicon-errata.txt b/Documentation/arm64/silicon-errata.txt
> index 4da60b463995..0d870d934528 100644
> --- a/Documentation/arm64/silicon-errata.txt
> +++ b/Documentation/arm64/silicon-errata.txt
> @@ -58,5 +58,6 @@ stable kernels.
>   | Cavium         | ThunderX ITS    | #22375, #24313  | CAVIUM_ERRATUM_22375    |
>   | Cavium         | ThunderX ITS    | #23144          | CAVIUM_ERRATUM_23144    |
>   | Cavium         | ThunderX GICv3  | #23154          | CAVIUM_ERRATUM_23154    |
> +| Cavium         | ThunderX Core   | #26026          | CAVIUM_ERRATUM_26026    |
>   | Cavium         | ThunderX Core   | #27456          | CAVIUM_ERRATUM_27456    |
>   | Cavium         | ThunderX SMMUv2 | #27704          | N/A		       |
> diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
> index 5a0a691d4220..8cbd9043ec6f 100644
> --- a/arch/arm64/Kconfig
> +++ b/arch/arm64/Kconfig
> @@ -457,6 +457,20 @@ config CAVIUM_ERRATUM_23154
>
>   	  If unsure, say Y.
>
> +config CAVIUM_ERRATUM_26026
> +	bool "Cavium erratum 26026: STX may return wrong status value"
> +	default y
> +	help
> +	  STX may return a wrong status value if the store was
> +	  successful. This may happen on ThunderX T88 pass 1.x cpus if
> +	  a broadcast TLBI is executed on another cpu in parallel to
> +	  an STX. As a result atomic or non-blocking implementations
> +	  can behave incorrectly. Use IPIs that call local TLBIs on
> +	  other cpus to avoid this.
> +	  Applies to ThunderX T88 pass 1.x cpus.
> +
> +	  If unsure, say Y.
> +
>   config CAVIUM_ERRATUM_27456
>   	bool "Cavium erratum 27456: Broadcast TLBI instructions may cause icache corruption"
>   	default y
> diff --git a/arch/arm64/kernel/cpu_errata.c b/arch/arm64/kernel/cpu_errata.c
> index af716b65110d..1e1753a6408e 100644
> --- a/arch/arm64/kernel/cpu_errata.c
> +++ b/arch/arm64/kernel/cpu_errata.c
> @@ -90,6 +90,14 @@ const struct arm64_cpu_capabilities arm64_errata[] = {
>   		MIDR_RANGE(MIDR_THUNDERX, 0x00, 0x01),
>   	},
>   #endif
> +#ifdef CONFIG_CAVIUM_ERRATUM_26026
> +	{
> +	/* Cavium ThunderX, pass 1.x */
> +		.desc = "Cavium erratum 26026",
> +		.capability = ARM64_HAS_NO_BCAST_TLBI,
> +		MIDR_RANGE(MIDR_THUNDERX, 0x00, 0x01),
> +	},
> +#endif
>   #ifdef CONFIG_CAVIUM_ERRATUM_27456
>   	{
>   	/* Cavium ThunderX, T88 pass 1.x - 2.1 */
>

[toc] | [prev] | [standalone]


Back to top | Article view | linux.kernel


csiph-web