Linux MIPS Architecture development
 help / color / mirror / Atom feed
* [PATCH 2.6.32 33/62] MIPS: Octeon: Remove udelay() causing huge IRQ latency
       [not found] <08d3b586eb2e764308c3de9ee398a17c@local>
@ 2015-09-12 22:56 ` Willy Tarreau
  2015-09-12 22:56 ` [PATCH 2.6.32 34/62] MIPS: Fix race condition in lazy cache flushing Willy Tarreau
  1 sibling, 0 replies; 2+ messages in thread
From: Willy Tarreau @ 2015-09-12 22:56 UTC (permalink / raw)
  To: linux-kernel, stable
  Cc: Alexander Sverdlin, linux-mips, David Daney, Rob Herring,
	Jiri Kosina, Randy Dunlap, Masanari Iida, Bjorn Helgaas, Mathias,
	Ralf Baechle, Ben Hutchings, Willy Tarreau

2.6.32-longterm review patch.  If anyone has any objections, please let me know.

------------------

From: Alexander Sverdlin <alexander.sverdlin@nokia.com>

commit 73bf3c2a500b2db8ac966469591196bf55afb409 upstream.

udelay() in PCI/PCIe read/write callbacks cause 30ms IRQ latency on Octeon
platforms because these operations are called from PCI_OP_READ() and
PCI_OP_WRITE() under raw_spin_lock_irqsave().

Signed-off-by: Alexander Sverdlin <alexander.sverdlin@nokia.com>
Cc: linux-mips@linux-mips.org
Cc: David Daney <ddaney@cavium.com>
Cc: Rob Herring <robh@kernel.org>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Masanari Iida <standby24x7@gmail.com>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: Mathias <mathias.rulf@nokia.com>
Patchwork: https://patchwork.linux-mips.org/patch/9576/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
(cherry picked from commit 53493d44a771a3155ee12b6ac668fb2543d21a7a)

Signed-off-by: Willy Tarreau <w@1wt.eu>
---
 arch/mips/include/asm/octeon/pci-octeon.h | 3 ---
 arch/mips/pci/pci-octeon.c                | 6 ------
 arch/mips/pci/pcie-octeon.c               | 3 ---
 3 files changed, 12 deletions(-)

diff --git a/arch/mips/include/asm/octeon/pci-octeon.h b/arch/mips/include/asm/octeon/pci-octeon.h
index 6ac5d3e..5eda9f0 100644
--- a/arch/mips/include/asm/octeon/pci-octeon.h
+++ b/arch/mips/include/asm/octeon/pci-octeon.h
@@ -11,9 +11,6 @@
 
 #include <linux/pci.h>
 
-/* Some PCI cards require delays when accessing config space. */
-#define PCI_CONFIG_SPACE_DELAY 10000
-
 /*
  * pcibios_map_irq() is defined inside pci-octeon.c. All it does is
  * call the Octeon specific version pointed to by this variable. This
diff --git a/arch/mips/pci/pci-octeon.c b/arch/mips/pci/pci-octeon.c
index 9cb0c80..dae7ff7 100644
--- a/arch/mips/pci/pci-octeon.c
+++ b/arch/mips/pci/pci-octeon.c
@@ -274,9 +274,6 @@ static int octeon_read_config(struct pci_bus *bus, unsigned int devfn,
 	pci_addr.s.func = devfn & 0x7;
 	pci_addr.s.reg = reg;
 
-#if PCI_CONFIG_SPACE_DELAY
-	udelay(PCI_CONFIG_SPACE_DELAY);
-#endif
 	switch (size) {
 	case 4:
 		*val = le32_to_cpu(cvmx_read64_uint32(pci_addr.u64));
@@ -311,9 +308,6 @@ static int octeon_write_config(struct pci_bus *bus, unsigned int devfn,
 	pci_addr.s.func = devfn & 0x7;
 	pci_addr.s.reg = reg;
 
-#if PCI_CONFIG_SPACE_DELAY
-	udelay(PCI_CONFIG_SPACE_DELAY);
-#endif
 	switch (size) {
 	case 4:
 		cvmx_write64_uint32(pci_addr.u64, cpu_to_le32(val));
diff --git a/arch/mips/pci/pcie-octeon.c b/arch/mips/pci/pcie-octeon.c
index 6aa5c54..97813f3 100644
--- a/arch/mips/pci/pcie-octeon.c
+++ b/arch/mips/pci/pcie-octeon.c
@@ -1192,9 +1192,6 @@ static inline int octeon_pcie_write_config(int pcie_port, struct pci_bus *bus,
 					devfn & 0x7, reg, val);
 		return PCIBIOS_SUCCESSFUL;
 	}
-#if PCI_CONFIG_SPACE_DELAY
-	udelay(PCI_CONFIG_SPACE_DELAY);
-#endif
 	return PCIBIOS_FUNC_NOT_SUPPORTED;
 }
 
-- 
1.7.12.2.21.g234cd45.dirty

^ permalink raw reply related	[flat|nested] 2+ messages in thread

* [PATCH 2.6.32 34/62] MIPS: Fix race condition in lazy cache flushing.
       [not found] <08d3b586eb2e764308c3de9ee398a17c@local>
  2015-09-12 22:56 ` [PATCH 2.6.32 33/62] MIPS: Octeon: Remove udelay() causing huge IRQ latency Willy Tarreau
@ 2015-09-12 22:56 ` Willy Tarreau
  1 sibling, 0 replies; 2+ messages in thread
From: Willy Tarreau @ 2015-09-12 22:56 UTC (permalink / raw)
  To: linux-kernel, stable
  Cc: Lars Persson, linux-mips, paul.burton, Ralf Baechle,
	Ben Hutchings, Willy Tarreau

2.6.32-longterm review patch.  If anyone has any objections, please let me know.

------------------

From: Lars Persson <lars.persson@axis.com>

commit 4d46a67a3eb827ccf1125959936fd51ba318dabc upstream.

The lazy cache flushing implemented in the MIPS kernel suffers from a
race condition that is exposed by do_set_pte() in mm/memory.c.

A pre-condition is a file-system that writes to the page from the CPU
in its readpage method and then calls flush_dcache_page(). One example
is ubifs. Another pre-condition is that the dcache flush is postponed
in __flush_dcache_page().

Upon a page fault for an executable mapping not existing in the
page-cache, the following will happen:
1. Write to the page
2. flush_dcache_page
3. flush_icache_page
4. set_pte_at
5. update_mmu_cache (commits the flush of a dcache-dirty page)

Between steps 4 and 5 another thread can hit the same page and it will
encounter a valid pte. Because the data still is in the L1 dcache the CPU
will fetch stale data from L2 into the icache and execute garbage.

This fix moves the commit of the cache flush to step 3 to close the
race window. It also reduces the amount of flushes on non-executable
mappings because we never enter __flush_dcache_page() for non-aliasing
CPUs.

Regressions can occur in drivers that mistakenly relies on the
flush_dcache_page() in get_user_pages() for DMA operations.

[ralf@linux-mips.org: Folded in patch 9346 to fix highmem issue.]

Signed-off-by: Lars Persson <larper@axis.com>
Cc: linux-mips@linux-mips.org
Cc: paul.burton@imgtec.com
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/9346/
Patchwork: https://patchwork.linux-mips.org/patch/9738/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
(cherry picked from commit 6bde6a3df0b4c8680d51c987d446b0ff2d6df0a6)

Signed-off-by: Willy Tarreau <w@1wt.eu>
---
 arch/mips/include/asm/cacheflush.h | 38 +++++++++++++++++++++++---------------
 arch/mips/mm/cache.c               | 12 ++++++++++++
 2 files changed, 35 insertions(+), 15 deletions(-)

diff --git a/arch/mips/include/asm/cacheflush.h b/arch/mips/include/asm/cacheflush.h
index 03b1d69..2211f24 100644
--- a/arch/mips/include/asm/cacheflush.h
+++ b/arch/mips/include/asm/cacheflush.h
@@ -29,6 +29,20 @@
  *  - flush_icache_all() flush the entire instruction cache
  *  - flush_data_cache_page() flushes a page from the data cache
  */
+
+ /*
+ * This flag is used to indicate that the page pointed to by a pte
+ * is dirty and requires cleaning before returning it to the user.
+ */
+#define PG_dcache_dirty			PG_arch_1
+
+#define Page_dcache_dirty(page)		\
+	test_bit(PG_dcache_dirty, &(page)->flags)
+#define SetPageDcacheDirty(page)	\
+	set_bit(PG_dcache_dirty, &(page)->flags)
+#define ClearPageDcacheDirty(page)	\
+	clear_bit(PG_dcache_dirty, &(page)->flags)
+
 extern void (*flush_cache_all)(void);
 extern void (*__flush_cache_all)(void);
 extern void (*flush_cache_mm)(struct mm_struct *mm);
@@ -37,12 +51,14 @@ extern void (*flush_cache_range)(struct vm_area_struct *vma,
 	unsigned long start, unsigned long end);
 extern void (*flush_cache_page)(struct vm_area_struct *vma, unsigned long page, unsigned long pfn);
 extern void __flush_dcache_page(struct page *page);
+extern void __flush_icache_page(struct vm_area_struct *vma, struct page *page);
 
 static inline void flush_dcache_page(struct page *page)
 {
-	if (cpu_has_dc_aliases || !cpu_has_ic_fills_f_dc)
+	if (cpu_has_dc_aliases)
 		__flush_dcache_page(page);
-
+	else if (!cpu_has_ic_fills_f_dc)
+		SetPageDcacheDirty(page);
 }
 
 #define flush_dcache_mmap_lock(mapping)		do { } while (0)
@@ -60,6 +76,11 @@ static inline void flush_anon_page(struct vm_area_struct *vma,
 static inline void flush_icache_page(struct vm_area_struct *vma,
 	struct page *page)
 {
+	if (!cpu_has_ic_fills_f_dc && (vma->vm_flags & VM_EXEC) &&
+	    Page_dcache_dirty(page)) {
+		__flush_icache_page(vma, page);
+		ClearPageDcacheDirty(page);
+	}
 }
 
 extern void (*flush_icache_range)(unsigned long start, unsigned long end);
@@ -94,19 +115,6 @@ extern void (*flush_icache_all)(void);
 extern void (*local_flush_data_cache_page)(void * addr);
 extern void (*flush_data_cache_page)(unsigned long addr);
 
-/*
- * This flag is used to indicate that the page pointed to by a pte
- * is dirty and requires cleaning before returning it to the user.
- */
-#define PG_dcache_dirty			PG_arch_1
-
-#define Page_dcache_dirty(page)		\
-	test_bit(PG_dcache_dirty, &(page)->flags)
-#define SetPageDcacheDirty(page)	\
-	set_bit(PG_dcache_dirty, &(page)->flags)
-#define ClearPageDcacheDirty(page)	\
-	clear_bit(PG_dcache_dirty, &(page)->flags)
-
 /* Run kernel code uncached, useful for cache probing functions. */
 unsigned long run_uncached(void *func);
 
diff --git a/arch/mips/mm/cache.c b/arch/mips/mm/cache.c
index 694d51f..37603a4 100644
--- a/arch/mips/mm/cache.c
+++ b/arch/mips/mm/cache.c
@@ -113,6 +113,18 @@ void __flush_anon_page(struct page *page, unsigned long vmaddr)
 
 EXPORT_SYMBOL(__flush_anon_page);
 
+void __flush_icache_page(struct vm_area_struct *vma, struct page *page)
+{
+	unsigned long addr;
+
+	if (PageHighMem(page))
+		return;
+
+	addr = (unsigned long) page_address(page);
+	flush_data_cache_page(addr);
+}
+EXPORT_SYMBOL_GPL(__flush_icache_page);
+
 void __update_cache(struct vm_area_struct *vma, unsigned long address,
 	pte_t pte)
 {
-- 
1.7.12.2.21.g234cd45.dirty

^ permalink raw reply related	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2015-09-12 23:13 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
     [not found] <08d3b586eb2e764308c3de9ee398a17c@local>
2015-09-12 22:56 ` [PATCH 2.6.32 33/62] MIPS: Octeon: Remove udelay() causing huge IRQ latency Willy Tarreau
2015-09-12 22:56 ` [PATCH 2.6.32 34/62] MIPS: Fix race condition in lazy cache flushing Willy Tarreau

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox