diff mbox series

[V5,2/3] riscv: Add ASID-based tlbflushing methods

Message ID 1622970249-50770-4-git-send-email-guoren@kernel.org (mailing list archive)
State New, archived
Headers show
Series riscv: Add DMA_COHERENT support for Allwinner D1 | expand

Commit Message

Guo Ren June 6, 2021, 9:03 a.m. UTC 
From: Guo Ren <guoren@linux.alibaba.com>

Implement optimized version of the tlb flushing routines for systems
using ASIDs. These are behind the use_asid_allocator static branch to
not affect existing systems not using ASIDs.

Signed-off-by: Guo Ren <guoren@linux.alibaba.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>
Cc: Palmer Dabbelt <palmerdabbelt@google.com>
Cc: Christoph Hellwig <hch@lst.de>
---
 arch/riscv/include/asm/mmu_context.h |  2 ++
 arch/riscv/include/asm/tlbflush.h    | 22 +++++++++++++++++
 arch/riscv/mm/context.c              |  2 +-
 arch/riscv/mm/tlbflush.c             | 46 +++++++++++++++++++++++++++++++++---
 4 files changed, 68 insertions(+), 4 deletions(-)

Comments

Christoph Hellwig June 6, 2021, 2:38 p.m. UTC | #1 
On Sun, Jun 06, 2021 at 09:03:58AM +0000, guoren@kernel.org wrote:
> +static inline void local_flush_tlb_all_asid(unsigned long asid)
> +{
> +	__asm__ __volatile__ ("sfence.vma x0, %0"
> +			:
> +			: "r" (asid)
> +			: "memory");
> +}
> +
> +static inline void local_flush_tlb_range_asid(unsigned long start,
> +				unsigned long size, unsigned long asid)
> +{
> +	unsigned long tmp, end = ALIGN(start + size, PAGE_SIZE);
> +
> +	for (tmp = start & PAGE_MASK; tmp < end; tmp += PAGE_SIZE) {
> +		__asm__ __volatile__ ("sfence.vma %0, %1"
> +				:
> +				: "r" (tmp), "r" (asid)
> +				: "memory");
> +	}

No need to expose these in a header.

> +static void __sbi_tlb_flush_range_asid(struct cpumask *cmask,
> +				       unsigned long start,
> +				       unsigned long size,
> +				       unsigned long asid)
> +{
> +	struct cpumask hmask;
> +	unsigned int cpuid;
> +
> +	if (cpumask_empty(cmask))
> +		return;
> +
> +	cpuid = get_cpu();
> +
> +	if (cpumask_any_but(cmask, cpuid) >= nr_cpu_ids) {
> +		if (size == -1)
> +			local_flush_tlb_all_asid(asid);
> +		else
> +			local_flush_tlb_range_asid(start, size, asid);
> +	} else {
> +		riscv_cpuid_to_hartid_mask(cmask, &hmask);
> +		sbi_remote_sfence_vma_asid(cpumask_bits(&hmask),
> +					   start, size, asid);
> +	}
> +
> +	put_cpu();
> +}

Still no need to duplicate most of this logic.  Also please document
why this uses a different tradeoff for the flush all logic compared
to the non-ASID path.

> +
>  void flush_tlb_mm(struct mm_struct *mm)
>  {
> -	__sbi_tlb_flush_range(mm_cpumask(mm), 0, -1);
> +	if (static_branch_unlikely(&use_asid_allocator))
> +		__sbi_tlb_flush_range_asid(mm_cpumask(mm), 0, -1,
> +					   atomic_long_read(&mm->context.id));
> +	else
> +		__sbi_tlb_flush_range(mm_cpumask(mm), 0, -1);
>  }
>  
>  void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr)
>  {
> -	__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), addr, PAGE_SIZE);
> +	if (static_branch_unlikely(&use_asid_allocator))
> +		__sbi_tlb_flush_range_asid(mm_cpumask(vma->vm_mm), addr, PAGE_SIZE,
> +					   atomic_long_read(&vma->vm_mm->context.id));
> +	else
> +		__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), addr, PAGE_SIZE);
>  }
>  
>  void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
>  		     unsigned long end)
>  {
> -	__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), start, end - start);
> +	if (static_branch_unlikely(&use_asid_allocator))
> +		__sbi_tlb_flush_range_asid(mm_cpumask(vma->vm_mm), start, end - start,
> +					   atomic_long_read(&vma->vm_mm->context.id));
> +	else
> +		__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), start, end - start);

Various overly long lines (which are trivially avoided when doing the
right thing from the beginning).
diff mbox series

Patch

diff --git a/arch/riscv/include/asm/mmu_context.h b/arch/riscv/include/asm/mmu_context.h
index b065941..7030837 100644
--- a/arch/riscv/include/asm/mmu_context.h
+++ b/arch/riscv/include/asm/mmu_context.h
@@ -33,6 +33,8 @@  static inline int init_new_context(struct task_struct *tsk,
 	return 0;
 }
 
+DECLARE_STATIC_KEY_FALSE(use_asid_allocator);
+
 #include <asm-generic/mmu_context.h>
 
 #endif /* _ASM_RISCV_MMU_CONTEXT_H */
diff --git a/arch/riscv/include/asm/tlbflush.h b/arch/riscv/include/asm/tlbflush.h
index c84218a..894cf75 100644
--- a/arch/riscv/include/asm/tlbflush.h
+++ b/arch/riscv/include/asm/tlbflush.h
@@ -22,9 +22,31 @@  static inline void local_flush_tlb_page(unsigned long addr)
 {
 	ALT_FLUSH_TLB_PAGE(__asm__ __volatile__ ("sfence.vma %0" : : "r" (addr) : "memory"));
 }
+
+static inline void local_flush_tlb_all_asid(unsigned long asid)
+{
+	__asm__ __volatile__ ("sfence.vma x0, %0"
+			:
+			: "r" (asid)
+			: "memory");
+}
+
+static inline void local_flush_tlb_range_asid(unsigned long start,
+				unsigned long size, unsigned long asid)
+{
+	unsigned long tmp, end = ALIGN(start + size, PAGE_SIZE);
+
+	for (tmp = start & PAGE_MASK; tmp < end; tmp += PAGE_SIZE) {
+		__asm__ __volatile__ ("sfence.vma %0, %1"
+				:
+				: "r" (tmp), "r" (asid)
+				: "memory");
+	}
+}
 #else /* CONFIG_MMU */
 #define local_flush_tlb_all()			do { } while (0)
 #define local_flush_tlb_page(addr)		do { } while (0)
+#define local_flush_tlb_range_asid(addr)	do { } while (0)
 #endif /* CONFIG_MMU */
 
 #if defined(CONFIG_SMP) && defined(CONFIG_MMU)
diff --git a/arch/riscv/mm/context.c b/arch/riscv/mm/context.c
index 68aa312..45c1b04 100644
--- a/arch/riscv/mm/context.c
+++ b/arch/riscv/mm/context.c
@@ -18,7 +18,7 @@ 
 
 #ifdef CONFIG_MMU
 
-static DEFINE_STATIC_KEY_FALSE(use_asid_allocator);
+DEFINE_STATIC_KEY_FALSE(use_asid_allocator);
 
 static unsigned long asid_bits;
 static unsigned long num_asids;
diff --git a/arch/riscv/mm/tlbflush.c b/arch/riscv/mm/tlbflush.c
index 720b443..87b4e52 100644
--- a/arch/riscv/mm/tlbflush.c
+++ b/arch/riscv/mm/tlbflush.c
@@ -4,6 +4,7 @@ 
 #include <linux/smp.h>
 #include <linux/sched.h>
 #include <asm/sbi.h>
+#include <asm/mmu_context.h>
 
 void flush_tlb_all(void)
 {
@@ -39,18 +40,57 @@  static void __sbi_tlb_flush_range(struct cpumask *cmask, unsigned long start,
 	put_cpu();
 }
 
+static void __sbi_tlb_flush_range_asid(struct cpumask *cmask,
+				       unsigned long start,
+				       unsigned long size,
+				       unsigned long asid)
+{
+	struct cpumask hmask;
+	unsigned int cpuid;
+
+	if (cpumask_empty(cmask))
+		return;
+
+	cpuid = get_cpu();
+
+	if (cpumask_any_but(cmask, cpuid) >= nr_cpu_ids) {
+		if (size == -1)
+			local_flush_tlb_all_asid(asid);
+		else
+			local_flush_tlb_range_asid(start, size, asid);
+	} else {
+		riscv_cpuid_to_hartid_mask(cmask, &hmask);
+		sbi_remote_sfence_vma_asid(cpumask_bits(&hmask),
+					   start, size, asid);
+	}
+
+	put_cpu();
+}
+
 void flush_tlb_mm(struct mm_struct *mm)
 {
-	__sbi_tlb_flush_range(mm_cpumask(mm), 0, -1);
+	if (static_branch_unlikely(&use_asid_allocator))
+		__sbi_tlb_flush_range_asid(mm_cpumask(mm), 0, -1,
+					   atomic_long_read(&mm->context.id));
+	else
+		__sbi_tlb_flush_range(mm_cpumask(mm), 0, -1);
 }
 
 void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr)
 {
-	__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), addr, PAGE_SIZE);
+	if (static_branch_unlikely(&use_asid_allocator))
+		__sbi_tlb_flush_range_asid(mm_cpumask(vma->vm_mm), addr, PAGE_SIZE,
+					   atomic_long_read(&vma->vm_mm->context.id));
+	else
+		__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), addr, PAGE_SIZE);
 }
 
 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
 		     unsigned long end)
 {
-	__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), start, end - start);
+	if (static_branch_unlikely(&use_asid_allocator))
+		__sbi_tlb_flush_range_asid(mm_cpumask(vma->vm_mm), start, end - start,
+					   atomic_long_read(&vma->vm_mm->context.id));
+	else
+		__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), start, end - start);
 }