* [PATCH v4 1/6] powerpc: Allow flush_icache_range to work across ranges >4GB
2019-09-26 4:54 [PATCH v4 0/6] powerpc: convert cache asm to C Alastair D'Silva
@ 2019-09-26 4:54 ` Alastair D'Silva
2019-09-26 4:54 ` [PATCH v4 2/6] powerpc: Allow 64bit VDSO __kernel_sync_dicache " Alastair D'Silva
` (4 subsequent siblings)
5 siblings, 0 replies; 9+ messages in thread
From: Alastair D'Silva @ 2019-09-26 4:54 UTC (permalink / raw)
To: alastair
Cc: Michal Hocko, Greg Kroah-Hartman, David Hildenbrand, linux-kernel,
stable, Mike Rapoport, Qian Cai, Nicholas Piggin, Paul Mackerras,
Thomas Gleixner, linuxppc-dev, Andrew Morton, Allison Randal
From: Alastair D'Silva <alastair@d-silva.org>
When calling flush_icache_range with a size >4GB, we were masking
off the upper 32 bits, so we would incorrectly flush a range smaller
than intended.
This patch replaces the 32 bit shifts with 64 bit ones, so that
the full size is accounted for.
Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
Cc: stable@vger.kernel.org
---
arch/powerpc/kernel/misc_64.S | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/powerpc/kernel/misc_64.S b/arch/powerpc/kernel/misc_64.S
index b55a7b4cb543..9bc0aa9aeb65 100644
--- a/arch/powerpc/kernel/misc_64.S
+++ b/arch/powerpc/kernel/misc_64.S
@@ -82,7 +82,7 @@ END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
subf r8,r6,r4 /* compute length */
add r8,r8,r5 /* ensure we get enough */
lwz r9,DCACHEL1LOGBLOCKSIZE(r10) /* Get log-2 of cache block size */
- srw. r8,r8,r9 /* compute line count */
+ srd. r8,r8,r9 /* compute line count */
beqlr /* nothing to do? */
mtctr r8
1: dcbst 0,r6
@@ -98,7 +98,7 @@ END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
subf r8,r6,r4 /* compute length */
add r8,r8,r5
lwz r9,ICACHEL1LOGBLOCKSIZE(r10) /* Get log-2 of Icache block size */
- srw. r8,r8,r9 /* compute line count */
+ srd. r8,r8,r9 /* compute line count */
beqlr /* nothing to do? */
mtctr r8
2: icbi 0,r6
--
2.21.0
^ permalink raw reply related [flat|nested] 9+ messages in thread* [PATCH v4 2/6] powerpc: Allow 64bit VDSO __kernel_sync_dicache to work across ranges >4GB
2019-09-26 4:54 [PATCH v4 0/6] powerpc: convert cache asm to C Alastair D'Silva
2019-09-26 4:54 ` [PATCH v4 1/6] powerpc: Allow flush_icache_range to work across ranges >4GB Alastair D'Silva
@ 2019-09-26 4:54 ` Alastair D'Silva
2019-09-26 4:54 ` [PATCH v4 3/6] powerpc: define helpers to get L1 icache sizes Alastair D'Silva
` (3 subsequent siblings)
5 siblings, 0 replies; 9+ messages in thread
From: Alastair D'Silva @ 2019-09-26 4:54 UTC (permalink / raw)
To: alastair
Cc: Greg Kroah-Hartman, David Hildenbrand, linux-kernel, stable,
Mike Rapoport, Paul Mackerras, Nicholas Piggin, Qian Cai,
Thomas Gleixner, linuxppc-dev, Andrew Morton, Allison Randal
From: Alastair D'Silva <alastair@d-silva.org>
When calling __kernel_sync_dicache with a size >4GB, we were masking
off the upper 32 bits, so we would incorrectly flush a range smaller
than intended.
This patch replaces the 32 bit shifts with 64 bit ones, so that
the full size is accounted for.
Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
Cc: stable@vger.kernel.org
---
arch/powerpc/kernel/vdso64/cacheflush.S | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/powerpc/kernel/vdso64/cacheflush.S b/arch/powerpc/kernel/vdso64/cacheflush.S
index 3f92561a64c4..526f5ba2593e 100644
--- a/arch/powerpc/kernel/vdso64/cacheflush.S
+++ b/arch/powerpc/kernel/vdso64/cacheflush.S
@@ -35,7 +35,7 @@ V_FUNCTION_BEGIN(__kernel_sync_dicache)
subf r8,r6,r4 /* compute length */
add r8,r8,r5 /* ensure we get enough */
lwz r9,CFG_DCACHE_LOGBLOCKSZ(r10)
- srw. r8,r8,r9 /* compute line count */
+ srd. r8,r8,r9 /* compute line count */
crclr cr0*4+so
beqlr /* nothing to do? */
mtctr r8
@@ -52,7 +52,7 @@ V_FUNCTION_BEGIN(__kernel_sync_dicache)
subf r8,r6,r4 /* compute length */
add r8,r8,r5
lwz r9,CFG_ICACHE_LOGBLOCKSZ(r10)
- srw. r8,r8,r9 /* compute line count */
+ srd. r8,r8,r9 /* compute line count */
crclr cr0*4+so
beqlr /* nothing to do? */
mtctr r8
--
2.21.0
^ permalink raw reply related [flat|nested] 9+ messages in thread* [PATCH v4 3/6] powerpc: define helpers to get L1 icache sizes
2019-09-26 4:54 [PATCH v4 0/6] powerpc: convert cache asm to C Alastair D'Silva
2019-09-26 4:54 ` [PATCH v4 1/6] powerpc: Allow flush_icache_range to work across ranges >4GB Alastair D'Silva
2019-09-26 4:54 ` [PATCH v4 2/6] powerpc: Allow 64bit VDSO __kernel_sync_dicache " Alastair D'Silva
@ 2019-09-26 4:54 ` Alastair D'Silva
2019-09-26 4:54 ` [PATCH v4 4/6] powerpc: Convert flush_icache_range & friends to C Alastair D'Silva
` (2 subsequent siblings)
5 siblings, 0 replies; 9+ messages in thread
From: Alastair D'Silva @ 2019-09-26 4:54 UTC (permalink / raw)
To: alastair
Cc: Greg Kroah-Hartman, David Hildenbrand, linux-kernel,
Nicholas Piggin, Mike Rapoport, Qian Cai, Paul Mackerras,
Thomas Gleixner, linuxppc-dev, Andrew Morton, Allison Randal
From: Alastair D'Silva <alastair@d-silva.org>
This patch adds helpers to retrieve icache sizes, and renames the existing
helpers to make it clear that they are for dcache.
Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
---
arch/powerpc/include/asm/cache.h | 29 +++++++++++++++++++++++----
arch/powerpc/include/asm/cacheflush.h | 12 +++++------
2 files changed, 31 insertions(+), 10 deletions(-)
diff --git a/arch/powerpc/include/asm/cache.h b/arch/powerpc/include/asm/cache.h
index b3388d95f451..f852d5cd746c 100644
--- a/arch/powerpc/include/asm/cache.h
+++ b/arch/powerpc/include/asm/cache.h
@@ -55,25 +55,46 @@ struct ppc64_caches {
extern struct ppc64_caches ppc64_caches;
-static inline u32 l1_cache_shift(void)
+static inline u32 l1_dcache_shift(void)
{
return ppc64_caches.l1d.log_block_size;
}
-static inline u32 l1_cache_bytes(void)
+static inline u32 l1_dcache_bytes(void)
{
return ppc64_caches.l1d.block_size;
}
+
+static inline u32 l1_icache_shift(void)
+{
+ return ppc64_caches.l1i.log_block_size;
+}
+
+static inline u32 l1_icache_bytes(void)
+{
+ return ppc64_caches.l1i.block_size;
+}
#else
-static inline u32 l1_cache_shift(void)
+static inline u32 l1_dcache_shift(void)
{
return L1_CACHE_SHIFT;
}
-static inline u32 l1_cache_bytes(void)
+static inline u32 l1_dcache_bytes(void)
{
return L1_CACHE_BYTES;
}
+
+static inline u32 l1_icache_shift(void)
+{
+ return L1_CACHE_SHIFT;
+}
+
+static inline u32 l1_icache_bytes(void)
+{
+ return L1_CACHE_BYTES;
+}
+
#endif
#endif /* ! __ASSEMBLY__ */
diff --git a/arch/powerpc/include/asm/cacheflush.h b/arch/powerpc/include/asm/cacheflush.h
index eef388f2659f..ed57843ef452 100644
--- a/arch/powerpc/include/asm/cacheflush.h
+++ b/arch/powerpc/include/asm/cacheflush.h
@@ -63,8 +63,8 @@ static inline void __flush_dcache_icache_phys(unsigned long physaddr)
*/
static inline void flush_dcache_range(unsigned long start, unsigned long stop)
{
- unsigned long shift = l1_cache_shift();
- unsigned long bytes = l1_cache_bytes();
+ unsigned long shift = l1_dcache_shift();
+ unsigned long bytes = l1_dcache_bytes();
void *addr = (void *)(start & ~(bytes - 1));
unsigned long size = stop - (unsigned long)addr + (bytes - 1);
unsigned long i;
@@ -89,8 +89,8 @@ static inline void flush_dcache_range(unsigned long start, unsigned long stop)
*/
static inline void clean_dcache_range(unsigned long start, unsigned long stop)
{
- unsigned long shift = l1_cache_shift();
- unsigned long bytes = l1_cache_bytes();
+ unsigned long shift = l1_dcache_shift();
+ unsigned long bytes = l1_dcache_bytes();
void *addr = (void *)(start & ~(bytes - 1));
unsigned long size = stop - (unsigned long)addr + (bytes - 1);
unsigned long i;
@@ -108,8 +108,8 @@ static inline void clean_dcache_range(unsigned long start, unsigned long stop)
static inline void invalidate_dcache_range(unsigned long start,
unsigned long stop)
{
- unsigned long shift = l1_cache_shift();
- unsigned long bytes = l1_cache_bytes();
+ unsigned long shift = l1_dcache_shift();
+ unsigned long bytes = l1_dcache_bytes();
void *addr = (void *)(start & ~(bytes - 1));
unsigned long size = stop - (unsigned long)addr + (bytes - 1);
unsigned long i;
--
2.21.0
^ permalink raw reply related [flat|nested] 9+ messages in thread* [PATCH v4 4/6] powerpc: Convert flush_icache_range & friends to C
2019-09-26 4:54 [PATCH v4 0/6] powerpc: convert cache asm to C Alastair D'Silva
` (2 preceding siblings ...)
2019-09-26 4:54 ` [PATCH v4 3/6] powerpc: define helpers to get L1 icache sizes Alastair D'Silva
@ 2019-09-26 4:54 ` Alastair D'Silva
2019-09-26 4:54 ` [PATCH v4 5/6] powerpc: Chunk calls to flush_dcache_range in arch_*_memory Alastair D'Silva
2019-09-26 4:54 ` [PATCH v4 6/6] powerpc: Don't flush caches when adding memory Alastair D'Silva
5 siblings, 0 replies; 9+ messages in thread
From: Alastair D'Silva @ 2019-09-26 4:54 UTC (permalink / raw)
To: alastair
Cc: Greg Kroah-Hartman, David Hildenbrand, linux-kernel,
Nicholas Piggin, Mike Rapoport, Qian Cai, Paul Mackerras,
Thomas Gleixner, linuxppc-dev, Andrew Morton, Allison Randal
From: Alastair D'Silva <alastair@d-silva.org>
Similar to commit 22e9c88d486a
("powerpc/64: reuse PPC32 static inline flush_dcache_range()")
this patch converts the following ASM symbols to C:
flush_icache_range()
__flush_dcache_icache()
__flush_dcache_icache_phys()
This was done as we discovered a long-standing bug where the length of the
range was truncated due to using a 32 bit shift instead of a 64 bit one.
By converting these functions to C, it becomes easier to maintain.
flush_dcache_icache_phys() retains a critical assembler section as we must
ensure there are no memory accesses while the data MMU is disabled
(authored by Christophe Leroy). Since this has no external callers, it has
also been made static, allowing the compiler to inline it within
flush_dcache_icache_page().
Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
arch/powerpc/include/asm/cache.h | 26 ++---
arch/powerpc/include/asm/cacheflush.h | 24 ++--
arch/powerpc/kernel/misc_32.S | 117 --------------------
arch/powerpc/kernel/misc_64.S | 102 -----------------
arch/powerpc/mm/mem.c | 151 +++++++++++++++++++++++++-
5 files changed, 172 insertions(+), 248 deletions(-)
diff --git a/arch/powerpc/include/asm/cache.h b/arch/powerpc/include/asm/cache.h
index f852d5cd746c..e33451a9c4e6 100644
--- a/arch/powerpc/include/asm/cache.h
+++ b/arch/powerpc/include/asm/cache.h
@@ -98,20 +98,7 @@ static inline u32 l1_icache_bytes(void)
#endif
#endif /* ! __ASSEMBLY__ */
-#if defined(__ASSEMBLY__)
-/*
- * For a snooping icache, we still need a dummy icbi to purge all the
- * prefetched instructions from the ifetch buffers. We also need a sync
- * before the icbi to order the the actual stores to memory that might
- * have modified instructions with the icbi.
- */
-#define PURGE_PREFETCHED_INS \
- sync; \
- icbi 0,r3; \
- sync; \
- isync
-
-#else
+#if !defined(__ASSEMBLY__)
#define __read_mostly __attribute__((__section__(".data..read_mostly")))
#ifdef CONFIG_PPC_BOOK3S_32
@@ -145,6 +132,17 @@ static inline void dcbst(void *addr)
{
__asm__ __volatile__ ("dcbst %y0" : : "Z"(*(u8 *)addr) : "memory");
}
+
+static inline void icbi(void *addr)
+{
+ asm volatile ("icbi 0, %0" : : "r"(addr) : "memory");
+}
+
+static inline void iccci(void *addr)
+{
+ asm volatile ("iccci 0, %0" : : "r"(addr) : "memory");
+}
+
#endif /* !__ASSEMBLY__ */
#endif /* __KERNEL__ */
#endif /* _ASM_POWERPC_CACHE_H */
diff --git a/arch/powerpc/include/asm/cacheflush.h b/arch/powerpc/include/asm/cacheflush.h
index ed57843ef452..4a1c9f0200e1 100644
--- a/arch/powerpc/include/asm/cacheflush.h
+++ b/arch/powerpc/include/asm/cacheflush.h
@@ -42,24 +42,20 @@ extern void flush_dcache_page(struct page *page);
#define flush_dcache_mmap_lock(mapping) do { } while (0)
#define flush_dcache_mmap_unlock(mapping) do { } while (0)
-extern void flush_icache_range(unsigned long, unsigned long);
+void flush_icache_range(unsigned long start, unsigned long stop);
extern void flush_icache_user_range(struct vm_area_struct *vma,
struct page *page, unsigned long addr,
int len);
-extern void __flush_dcache_icache(void *page_va);
extern void flush_dcache_icache_page(struct page *page);
-#if defined(CONFIG_PPC32) && !defined(CONFIG_BOOKE)
-extern void __flush_dcache_icache_phys(unsigned long physaddr);
-#else
-static inline void __flush_dcache_icache_phys(unsigned long physaddr)
-{
- BUG();
-}
-#endif
-
-/*
- * Write any modified data cache blocks out to memory and invalidate them.
- * Does not invalidate the corresponding instruction cache blocks.
+void __flush_dcache_icache(void *page);
+
+/**
+ * flush_dcache_range(): Write any modified data cache blocks out to memory and
+ * invalidate them. Does not invalidate the corresponding instruction cache
+ * blocks.
+ *
+ * @start: the start address
+ * @stop: the stop address (exclusive)
*/
static inline void flush_dcache_range(unsigned long start, unsigned long stop)
{
diff --git a/arch/powerpc/kernel/misc_32.S b/arch/powerpc/kernel/misc_32.S
index fe4bd321730e..12b95e6799d4 100644
--- a/arch/powerpc/kernel/misc_32.S
+++ b/arch/powerpc/kernel/misc_32.S
@@ -318,123 +318,6 @@ END_FTR_SECTION_IFSET(CPU_FTR_UNIFIED_ID_CACHE)
EXPORT_SYMBOL(flush_instruction_cache)
#endif /* CONFIG_PPC_8xx */
-/*
- * Write any modified data cache blocks out to memory
- * and invalidate the corresponding instruction cache blocks.
- * This is a no-op on the 601.
- *
- * flush_icache_range(unsigned long start, unsigned long stop)
- */
-_GLOBAL(flush_icache_range)
-BEGIN_FTR_SECTION
- PURGE_PREFETCHED_INS
- blr /* for 601, do nothing */
-END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
- rlwinm r3,r3,0,0,31 - L1_CACHE_SHIFT
- subf r4,r3,r4
- addi r4,r4,L1_CACHE_BYTES - 1
- srwi. r4,r4,L1_CACHE_SHIFT
- beqlr
- mtctr r4
- mr r6,r3
-1: dcbst 0,r3
- addi r3,r3,L1_CACHE_BYTES
- bdnz 1b
- sync /* wait for dcbst's to get to ram */
-#ifndef CONFIG_44x
- mtctr r4
-2: icbi 0,r6
- addi r6,r6,L1_CACHE_BYTES
- bdnz 2b
-#else
- /* Flash invalidate on 44x because we are passed kmapped addresses and
- this doesn't work for userspace pages due to the virtually tagged
- icache. Sigh. */
- iccci 0, r0
-#endif
- sync /* additional sync needed on g4 */
- isync
- blr
-_ASM_NOKPROBE_SYMBOL(flush_icache_range)
-EXPORT_SYMBOL(flush_icache_range)
-
-/*
- * Flush a particular page from the data cache to RAM.
- * Note: this is necessary because the instruction cache does *not*
- * snoop from the data cache.
- * This is a no-op on the 601 which has a unified cache.
- *
- * void __flush_dcache_icache(void *page)
- */
-_GLOBAL(__flush_dcache_icache)
-BEGIN_FTR_SECTION
- PURGE_PREFETCHED_INS
- blr
-END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
- rlwinm r3,r3,0,0,31-PAGE_SHIFT /* Get page base address */
- li r4,PAGE_SIZE/L1_CACHE_BYTES /* Number of lines in a page */
- mtctr r4
- mr r6,r3
-0: dcbst 0,r3 /* Write line to ram */
- addi r3,r3,L1_CACHE_BYTES
- bdnz 0b
- sync
-#ifdef CONFIG_44x
- /* We don't flush the icache on 44x. Those have a virtual icache
- * and we don't have access to the virtual address here (it's
- * not the page vaddr but where it's mapped in user space). The
- * flushing of the icache on these is handled elsewhere, when
- * a change in the address space occurs, before returning to
- * user space
- */
-BEGIN_MMU_FTR_SECTION
- blr
-END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_44x)
-#endif /* CONFIG_44x */
- mtctr r4
-1: icbi 0,r6
- addi r6,r6,L1_CACHE_BYTES
- bdnz 1b
- sync
- isync
- blr
-
-#ifndef CONFIG_BOOKE
-/*
- * Flush a particular page from the data cache to RAM, identified
- * by its physical address. We turn off the MMU so we can just use
- * the physical address (this may be a highmem page without a kernel
- * mapping).
- *
- * void __flush_dcache_icache_phys(unsigned long physaddr)
- */
-_GLOBAL(__flush_dcache_icache_phys)
-BEGIN_FTR_SECTION
- PURGE_PREFETCHED_INS
- blr /* for 601, do nothing */
-END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
- mfmsr r10
- rlwinm r0,r10,0,28,26 /* clear DR */
- mtmsr r0
- isync
- rlwinm r3,r3,0,0,31-PAGE_SHIFT /* Get page base address */
- li r4,PAGE_SIZE/L1_CACHE_BYTES /* Number of lines in a page */
- mtctr r4
- mr r6,r3
-0: dcbst 0,r3 /* Write line to ram */
- addi r3,r3,L1_CACHE_BYTES
- bdnz 0b
- sync
- mtctr r4
-1: icbi 0,r6
- addi r6,r6,L1_CACHE_BYTES
- bdnz 1b
- sync
- mtmsr r10 /* restore DR */
- isync
- blr
-#endif /* CONFIG_BOOKE */
-
/*
* Copy a whole page. We use the dcbz instruction on the destination
* to reduce memory traffic (it eliminates the unnecessary reads of
diff --git a/arch/powerpc/kernel/misc_64.S b/arch/powerpc/kernel/misc_64.S
index 9bc0aa9aeb65..ff20c253f273 100644
--- a/arch/powerpc/kernel/misc_64.S
+++ b/arch/powerpc/kernel/misc_64.S
@@ -49,108 +49,6 @@ _GLOBAL(call_do_irq)
mtlr r0
blr
- .section ".toc","aw"
-PPC64_CACHES:
- .tc ppc64_caches[TC],ppc64_caches
- .section ".text"
-
-/*
- * Write any modified data cache blocks out to memory
- * and invalidate the corresponding instruction cache blocks.
- *
- * flush_icache_range(unsigned long start, unsigned long stop)
- *
- * flush all bytes from start through stop-1 inclusive
- */
-
-_GLOBAL_TOC(flush_icache_range)
-BEGIN_FTR_SECTION
- PURGE_PREFETCHED_INS
- blr
-END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
-/*
- * Flush the data cache to memory
- *
- * Different systems have different cache line sizes
- * and in some cases i-cache and d-cache line sizes differ from
- * each other.
- */
- ld r10,PPC64_CACHES@toc(r2)
- lwz r7,DCACHEL1BLOCKSIZE(r10)/* Get cache block size */
- addi r5,r7,-1
- andc r6,r3,r5 /* round low to line bdy */
- subf r8,r6,r4 /* compute length */
- add r8,r8,r5 /* ensure we get enough */
- lwz r9,DCACHEL1LOGBLOCKSIZE(r10) /* Get log-2 of cache block size */
- srd. r8,r8,r9 /* compute line count */
- beqlr /* nothing to do? */
- mtctr r8
-1: dcbst 0,r6
- add r6,r6,r7
- bdnz 1b
- sync
-
-/* Now invalidate the instruction cache */
-
- lwz r7,ICACHEL1BLOCKSIZE(r10) /* Get Icache block size */
- addi r5,r7,-1
- andc r6,r3,r5 /* round low to line bdy */
- subf r8,r6,r4 /* compute length */
- add r8,r8,r5
- lwz r9,ICACHEL1LOGBLOCKSIZE(r10) /* Get log-2 of Icache block size */
- srd. r8,r8,r9 /* compute line count */
- beqlr /* nothing to do? */
- mtctr r8
-2: icbi 0,r6
- add r6,r6,r7
- bdnz 2b
- isync
- blr
-_ASM_NOKPROBE_SYMBOL(flush_icache_range)
-EXPORT_SYMBOL(flush_icache_range)
-
-/*
- * Flush a particular page from the data cache to RAM.
- * Note: this is necessary because the instruction cache does *not*
- * snoop from the data cache.
- *
- * void __flush_dcache_icache(void *page)
- */
-_GLOBAL(__flush_dcache_icache)
-/*
- * Flush the data cache to memory
- *
- * Different systems have different cache line sizes
- */
-
-BEGIN_FTR_SECTION
- PURGE_PREFETCHED_INS
- blr
-END_FTR_SECTION_IFSET(CPU_FTR_COHERENT_ICACHE)
-
-/* Flush the dcache */
- ld r7,PPC64_CACHES@toc(r2)
- clrrdi r3,r3,PAGE_SHIFT /* Page align */
- lwz r4,DCACHEL1BLOCKSPERPAGE(r7) /* Get # dcache blocks per page */
- lwz r5,DCACHEL1BLOCKSIZE(r7) /* Get dcache block size */
- mr r6,r3
- mtctr r4
-0: dcbst 0,r6
- add r6,r6,r5
- bdnz 0b
- sync
-
-/* Now invalidate the icache */
-
- lwz r4,ICACHEL1BLOCKSPERPAGE(r7) /* Get # icache blocks per page */
- lwz r5,ICACHEL1BLOCKSIZE(r7) /* Get icache block size */
- mtctr r4
-1: icbi 0,r3
- add r3,r3,r5
- bdnz 1b
- isync
- blr
-
_GLOBAL(__bswapdi2)
EXPORT_SYMBOL(__bswapdi2)
srdi r8,r3,32
diff --git a/arch/powerpc/mm/mem.c b/arch/powerpc/mm/mem.c
index 9191a66b3bc5..cff947cb2a84 100644
--- a/arch/powerpc/mm/mem.c
+++ b/arch/powerpc/mm/mem.c
@@ -321,6 +321,120 @@ void free_initmem(void)
free_initmem_default(POISON_FREE_INITMEM);
}
+/**
+ * flush_coherent_icache() - if a CPU has a coherent icache, flush it
+ * @addr: The base address to use (can be any valid address, the whole cache will be flushed)
+ * Return true if the cache was flushed, false otherwise
+ */
+static inline bool flush_coherent_icache(unsigned long addr)
+{
+ /*
+ * For a snooping icache, we still need a dummy icbi to purge all the
+ * prefetched instructions from the ifetch buffers. We also need a sync
+ * before the icbi to order the the actual stores to memory that might
+ * have modified instructions with the icbi.
+ */
+ if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE)) {
+ mb(); /* sync */
+ icbi((void *)addr);
+ mb(); /* sync */
+ isync();
+ return true;
+ }
+
+ return false;
+}
+
+/**
+ * invalidate_icache_range() - Flush the icache by issuing icbi across an address range
+ * @start: the start address
+ * @stop: the stop address (exclusive)
+ */
+static void invalidate_icache_range(unsigned long start, unsigned long stop)
+{
+ unsigned long shift = l1_icache_shift();
+ unsigned long bytes = l1_icache_bytes();
+ char *addr = (char *)(start & ~(bytes - 1));
+ unsigned long size = stop - (unsigned long)addr + (bytes - 1);
+ unsigned long i;
+
+ for (i = 0; i < size >> shift; i++, addr += bytes)
+ icbi(addr);
+
+ mb(); /* sync */
+ isync();
+}
+
+/**
+ * flush_icache_range: Write any modified data cache blocks out to memory
+ * and invalidate the corresponding blocks in the instruction cache
+ *
+ * Generic code will call this after writing memory, before executing from it.
+ *
+ * @start: the start address
+ * @stop: the stop address (exclusive)
+ */
+void flush_icache_range(unsigned long start, unsigned long stop)
+{
+ if (flush_coherent_icache(start))
+ return;
+
+ clean_dcache_range(start, stop);
+
+ if (IS_ENABLED(CONFIG_44x)) {
+ /*
+ * Flash invalidate on 44x because we are passed kmapped
+ * addresses and this doesn't work for userspace pages due to
+ * the virtually tagged icache.
+ */
+ iccci((void *)start);
+ mb(); /* sync */
+ isync();
+ } else
+ invalidate_icache_range(start, stop);
+}
+EXPORT_SYMBOL(flush_icache_range);
+
+#if !defined(CONFIG_PPC_8xx) && !defined(CONFIG_PPC64)
+/**
+ * flush_dcache_icache_phys() - Flush a page by it's physical address
+ * @physaddr: the physical address of the page
+ */
+static void flush_dcache_icache_phys(unsigned long physaddr)
+{
+ unsigned long bytes = l1_dcache_bytes();
+ unsigned long nb = PAGE_SIZE / bytes;
+ unsigned long addr = physaddr & PAGE_MASK;
+ unsigned long msr, msr0;
+ unsigned long loop1 = addr, loop2 = addr;
+
+ msr0 = mfmsr();
+ msr = msr0 & ~MSR_DR;
+ /*
+ * This must remain as ASM to prevent potential memory accesses
+ * while the data MMU is disabled
+ */
+ asm volatile(
+ " mtctr %2;\n"
+ " mtmsr %3;\n"
+ " isync;\n"
+ "0: dcbst 0, %0;\n"
+ " addi %0, %0, %4;\n"
+ " bdnz 0b;\n"
+ " sync;\n"
+ " mtctr %2;\n"
+ "1: icbi 0, %1;\n"
+ " addi %1, %1, %4;\n"
+ " bdnz 1b;\n"
+ " sync;\n"
+ " mtmsr %5;\n"
+ " isync;\n"
+ : "+&r" (loop1), "+&r" (loop2)
+ : "r" (nb), "r" (msr), "i" (bytes), "r" (msr0)
+ : "ctr", "memory");
+}
+#endif // !defined(CONFIG_PPC_8xx) && !defined(CONFIG_PPC64)
+
/*
* This is called when a page has been modified by the kernel.
* It just marks the page as not i-cache clean. We do the i-cache
@@ -353,12 +467,47 @@ void flush_dcache_icache_page(struct page *page)
__flush_dcache_icache(start);
kunmap_atomic(start);
} else {
- __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
+ unsigned long addr = page_to_pfn(page) << PAGE_SHIFT;
+
+ if (flush_coherent_icache((void *)addr))
+ return;
+ flush_dcache_icache_phys(addr);
}
#endif
}
EXPORT_SYMBOL(flush_dcache_icache_page);
+/**
+ * __flush_dcache_icache(): Flush a particular page from the data cache to RAM.
+ * Note: this is necessary because the instruction cache does *not*
+ * snoop from the data cache.
+ *
+ * @page: the address of the page to flush
+ */
+void __flush_dcache_icache(void *page)
+{
+ u64 addr = (u64)page;
+
+ if (flush_coherent_icache(addr))
+ return;
+
+ clean_dcache_range(addr, addr + PAGE_SIZE);
+
+ /*
+ * We don't flush the icache on 44x. Those have a virtual icache and we
+ * don't have access to the virtual address here (it's not the page
+ * vaddr but where it's mapped in user space). The flushing of the
+ * icache on these is handled elsewhere, when a change in the address
+ * space occurs, before returning to user space.
+ */
+
+ if (cpu_has_feature(MMU_FTR_TYPE_44x))
+ return;
+
+ invalidate_icache_range(addr, addr + PAGE_SIZE);
+}
+EXPORT_SYMBOL(__flush_dcache_icache);
+
void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
{
clear_page(page);
--
2.21.0
^ permalink raw reply related [flat|nested] 9+ messages in thread* [PATCH v4 5/6] powerpc: Chunk calls to flush_dcache_range in arch_*_memory
2019-09-26 4:54 [PATCH v4 0/6] powerpc: convert cache asm to C Alastair D'Silva
` (3 preceding siblings ...)
2019-09-26 4:54 ` [PATCH v4 4/6] powerpc: Convert flush_icache_range & friends to C Alastair D'Silva
@ 2019-09-26 4:54 ` Alastair D'Silva
2019-09-27 6:37 ` Mark Marshall
2019-09-26 4:54 ` [PATCH v4 6/6] powerpc: Don't flush caches when adding memory Alastair D'Silva
5 siblings, 1 reply; 9+ messages in thread
From: Alastair D'Silva @ 2019-09-26 4:54 UTC (permalink / raw)
To: alastair
Cc: Michal Hocko, David Hildenbrand, linux-kernel, Nicholas Piggin,
Mike Rapoport, Qian Cai, Greg Kroah-Hartman, Paul Mackerras,
Thomas Gleixner, linuxppc-dev, Andrew Morton, Allison Randal
From: Alastair D'Silva <alastair@d-silva.org>
When presented with large amounts of memory being hotplugged
(in my test case, ~890GB), the call to flush_dcache_range takes
a while (~50 seconds), triggering RCU stalls.
This patch breaks up the call into 1GB chunks, calling
cond_resched() inbetween to allow the scheduler to run.
Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
---
arch/powerpc/mm/mem.c | 27 +++++++++++++++++++++++++--
1 file changed, 25 insertions(+), 2 deletions(-)
diff --git a/arch/powerpc/mm/mem.c b/arch/powerpc/mm/mem.c
index cff947cb2a84..a2758e473d58 100644
--- a/arch/powerpc/mm/mem.c
+++ b/arch/powerpc/mm/mem.c
@@ -104,6 +104,27 @@ int __weak remove_section_mapping(unsigned long start, unsigned long end)
return -ENODEV;
}
+#define FLUSH_CHUNK_SIZE SZ_1G
+/**
+ * flush_dcache_range_chunked(): Write any modified data cache blocks out to
+ * memory and invalidate them, in chunks of up to FLUSH_CHUNK_SIZE
+ * Does not invalidate the corresponding instruction cache blocks.
+ *
+ * @start: the start address
+ * @stop: the stop address (exclusive)
+ * @chunk: the max size of the chunks
+ */
+static void flush_dcache_range_chunked(unsigned long start, unsigned long stop,
+ unsigned long chunk)
+{
+ unsigned long i;
+
+ for (i = start; i < stop; i += FLUSH_CHUNK_SIZE) {
+ flush_dcache_range(i, min(stop, start + FLUSH_CHUNK_SIZE));
+ cond_resched();
+ }
+}
+
int __ref arch_add_memory(int nid, u64 start, u64 size,
struct mhp_restrictions *restrictions)
{
@@ -120,7 +141,8 @@ int __ref arch_add_memory(int nid, u64 start, u64 size,
start, start + size, rc);
return -EFAULT;
}
- flush_dcache_range(start, start + size);
+
+ flush_dcache_range_chunked(start, start + size, FLUSH_CHUNK_SIZE);
return __add_pages(nid, start_pfn, nr_pages, restrictions);
}
@@ -137,7 +159,8 @@ void __ref arch_remove_memory(int nid, u64 start, u64 size,
/* Remove htab bolted mappings for this section of memory */
start = (unsigned long)__va(start);
- flush_dcache_range(start, start + size);
+ flush_dcache_range_chunked(start, start + size, FLUSH_CHUNK_SIZE);
+
ret = remove_section_mapping(start, start + size);
WARN_ON_ONCE(ret);
--
2.21.0
^ permalink raw reply related [flat|nested] 9+ messages in thread* Re: [PATCH v4 5/6] powerpc: Chunk calls to flush_dcache_range in arch_*_memory
2019-09-26 4:54 ` [PATCH v4 5/6] powerpc: Chunk calls to flush_dcache_range in arch_*_memory Alastair D'Silva
@ 2019-09-27 6:37 ` Mark Marshall
2019-09-27 6:46 ` Alastair D'Silva
0 siblings, 1 reply; 9+ messages in thread
From: Mark Marshall @ 2019-09-27 6:37 UTC (permalink / raw)
To: Alastair D'Silva
Cc: Michal Hocko, Greg Kroah-Hartman, alastair, David Hildenbrand,
linux-kernel, Nicholas Piggin, Mike Rapoport, Qian Cai,
Paul Mackerras, Thomas Gleixner, linuxppc-dev, Andrew Morton,
Allison Randal
Comment below...
On Thu, 26 Sep 2019 at 12:18, Alastair D'Silva <alastair@au1.ibm.com> wrote:
>
> From: Alastair D'Silva <alastair@d-silva.org>
>
> When presented with large amounts of memory being hotplugged
> (in my test case, ~890GB), the call to flush_dcache_range takes
> a while (~50 seconds), triggering RCU stalls.
>
> This patch breaks up the call into 1GB chunks, calling
> cond_resched() inbetween to allow the scheduler to run.
>
> Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
> ---
> arch/powerpc/mm/mem.c | 27 +++++++++++++++++++++++++--
> 1 file changed, 25 insertions(+), 2 deletions(-)
>
> diff --git a/arch/powerpc/mm/mem.c b/arch/powerpc/mm/mem.c
> index cff947cb2a84..a2758e473d58 100644
> --- a/arch/powerpc/mm/mem.c
> +++ b/arch/powerpc/mm/mem.c
> @@ -104,6 +104,27 @@ int __weak remove_section_mapping(unsigned long start, unsigned long end)
> return -ENODEV;
> }
>
> +#define FLUSH_CHUNK_SIZE SZ_1G
> +/**
> + * flush_dcache_range_chunked(): Write any modified data cache blocks out to
> + * memory and invalidate them, in chunks of up to FLUSH_CHUNK_SIZE
> + * Does not invalidate the corresponding instruction cache blocks.
> + *
> + * @start: the start address
> + * @stop: the stop address (exclusive)
> + * @chunk: the max size of the chunks
> + */
> +static void flush_dcache_range_chunked(unsigned long start, unsigned long stop,
> + unsigned long chunk)
> +{
> + unsigned long i;
> +
> + for (i = start; i < stop; i += FLUSH_CHUNK_SIZE) {
Here you ignore the function parameter "chunk" and use the define
FLUSH_CHUNK_SIZE.
You should do one or the other; use the parameter or remove it.
> + flush_dcache_range(i, min(stop, start + FLUSH_CHUNK_SIZE));
> + cond_resched();
> + }
> +}
> +
> int __ref arch_add_memory(int nid, u64 start, u64 size,
> struct mhp_restrictions *restrictions)
> {
> @@ -120,7 +141,8 @@ int __ref arch_add_memory(int nid, u64 start, u64 size,
> start, start + size, rc);
> return -EFAULT;
> }
> - flush_dcache_range(start, start + size);
> +
> + flush_dcache_range_chunked(start, start + size, FLUSH_CHUNK_SIZE);
>
> return __add_pages(nid, start_pfn, nr_pages, restrictions);
> }
> @@ -137,7 +159,8 @@ void __ref arch_remove_memory(int nid, u64 start, u64 size,
>
> /* Remove htab bolted mappings for this section of memory */
> start = (unsigned long)__va(start);
> - flush_dcache_range(start, start + size);
> + flush_dcache_range_chunked(start, start + size, FLUSH_CHUNK_SIZE);
> +
> ret = remove_section_mapping(start, start + size);
> WARN_ON_ONCE(ret);
>
> --
> 2.21.0
>
^ permalink raw reply [flat|nested] 9+ messages in thread* RE: [PATCH v4 5/6] powerpc: Chunk calls to flush_dcache_range in arch_*_memory
2019-09-27 6:37 ` Mark Marshall
@ 2019-09-27 6:46 ` Alastair D'Silva
0 siblings, 0 replies; 9+ messages in thread
From: Alastair D'Silva @ 2019-09-27 6:46 UTC (permalink / raw)
To: Mark Marshall
Cc: Michal Hocko, David Hildenbrand, linux-kernel, Nicholas Piggin,
Mike Rapoport, Qian Cai, Greg Kroah-Hartman, Paul Mackerras,
Thomas Gleixner, linuxppc-dev, Andrew Morton, Allison Randal
On Fri, 2019-09-27 at 08:37 +0200, Mark Marshall wrote:
> Comment below...
>
> On Thu, 26 Sep 2019 at 12:18, Alastair D'Silva <alastair@au1.ibm.com>
> wrote:
> > From: Alastair D'Silva <alastair@d-silva.org>
> >
> > When presented with large amounts of memory being hotplugged
> > (in my test case, ~890GB), the call to flush_dcache_range takes
> > a while (~50 seconds), triggering RCU stalls.
> >
> > This patch breaks up the call into 1GB chunks, calling
> > cond_resched() inbetween to allow the scheduler to run.
> >
> > Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
> > ---
> > arch/powerpc/mm/mem.c | 27 +++++++++++++++++++++++++--
> > 1 file changed, 25 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/powerpc/mm/mem.c b/arch/powerpc/mm/mem.c
> > index cff947cb2a84..a2758e473d58 100644
> > --- a/arch/powerpc/mm/mem.c
> > +++ b/arch/powerpc/mm/mem.c
> > @@ -104,6 +104,27 @@ int __weak remove_section_mapping(unsigned
> > long start, unsigned long end)
> > return -ENODEV;
> > }
> >
> > +#define FLUSH_CHUNK_SIZE SZ_1G
> > +/**
> > + * flush_dcache_range_chunked(): Write any modified data cache
> > blocks out to
> > + * memory and invalidate them, in chunks of up to FLUSH_CHUNK_SIZE
> > + * Does not invalidate the corresponding instruction cache blocks.
> > + *
> > + * @start: the start address
> > + * @stop: the stop address (exclusive)
> > + * @chunk: the max size of the chunks
> > + */
> > +static void flush_dcache_range_chunked(unsigned long start,
> > unsigned long stop,
> > + unsigned long chunk)
> > +{
> > + unsigned long i;
> > +
> > + for (i = start; i < stop; i += FLUSH_CHUNK_SIZE) {
> Here you ignore the function parameter "chunk" and use the define
> FLUSH_CHUNK_SIZE.
> You should do one or the other; use the parameter or remove it.
Good catch, thankyou :)
> > + flush_dcache_range(i, min(stop, start +
> > FLUSH_CHUNK_SIZE));
> > + cond_resched();
> > + }
> > +}
> > +
> > int __ref arch_add_memory(int nid, u64 start, u64 size,
> > struct mhp_restrictions *restrictions)
> > {
> > @@ -120,7 +141,8 @@ int __ref arch_add_memory(int nid, u64 start,
> > u64 size,
> > start, start + size, rc);
> > return -EFAULT;
> > }
> > - flush_dcache_range(start, start + size);
> > +
> > + flush_dcache_range_chunked(start, start + size,
> > FLUSH_CHUNK_SIZE);
> >
> > return __add_pages(nid, start_pfn, nr_pages, restrictions);
> > }
> > @@ -137,7 +159,8 @@ void __ref arch_remove_memory(int nid, u64
> > start, u64 size,
> >
> > /* Remove htab bolted mappings for this section of memory
> > */
> > start = (unsigned long)__va(start);
> > - flush_dcache_range(start, start + size);
> > + flush_dcache_range_chunked(start, start + size,
> > FLUSH_CHUNK_SIZE);
> > +
> > ret = remove_section_mapping(start, start + size);
> > WARN_ON_ONCE(ret);
> >
> > --
> > 2.21.0
> >
--
Alastair D'Silva
Open Source Developer
Linux Technology Centre, IBM Australia
mob: 0423 762 819
^ permalink raw reply [flat|nested] 9+ messages in thread
* [PATCH v4 6/6] powerpc: Don't flush caches when adding memory
2019-09-26 4:54 [PATCH v4 0/6] powerpc: convert cache asm to C Alastair D'Silva
` (4 preceding siblings ...)
2019-09-26 4:54 ` [PATCH v4 5/6] powerpc: Chunk calls to flush_dcache_range in arch_*_memory Alastair D'Silva
@ 2019-09-26 4:54 ` Alastair D'Silva
5 siblings, 0 replies; 9+ messages in thread
From: Alastair D'Silva @ 2019-09-26 4:54 UTC (permalink / raw)
To: alastair
Cc: Michal Hocko, Greg Kroah-Hartman, David Hildenbrand, linux-kernel,
Nicholas Piggin, Mike Rapoport, Qian Cai, Paul Mackerras,
Thomas Gleixner, linuxppc-dev, Andrew Morton, Allison Randal
From: Alastair D'Silva <alastair@d-silva.org>
This operation takes a significant amount of time when hotplugging
large amounts of memory (~50 seconds with 890GB of persistent memory).
This was orignally in commit fb5924fddf9e
("powerpc/mm: Flush cache on memory hot(un)plug") to support memtrace,
but the flush on add is not needed as it is flushed on remove.
Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
---
arch/powerpc/mm/mem.c | 2 --
1 file changed, 2 deletions(-)
diff --git a/arch/powerpc/mm/mem.c b/arch/powerpc/mm/mem.c
index a2758e473d58..9afb77a8c8d5 100644
--- a/arch/powerpc/mm/mem.c
+++ b/arch/powerpc/mm/mem.c
@@ -142,8 +142,6 @@ int __ref arch_add_memory(int nid, u64 start, u64 size,
return -EFAULT;
}
- flush_dcache_range_chunked(start, start + size, FLUSH_CHUNK_SIZE);
-
return __add_pages(nid, start_pfn, nr_pages, restrictions);
}
--
2.21.0
^ permalink raw reply related [flat|nested] 9+ messages in thread