Skip to contentSkip to conversation

MIPS: Sync after cacheflush on BMIPS processors

PATCH New Kevin Cernekee 1 participant 6 messages Jan 14, 2011 MIPS

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/6 MIPS: Sync after cacheflush on BMIPS processors

On BMIPS processors, cache writeback operations may complete before the data has actually been written out to memory. Subsequent uncached reads (or I/O operations) may see stale data unless a sync instruction is executed after the writeback loop.

---
 arch/mips/include/asm/hazards.h |   21 +++++++++++++++++++++
 arch/mips/mm/c-r4k.c            |    5 +++++
 2 files changed, 26 insertions(+), 0 deletions(-)
arch/mips/include/asm/hazards.h +21 -0
index 4e33216..655da05 100644
@@ -270,4 +270,25 @@ ASMMACRO(disable_fpu_hazard,
 )
 #endif
 
+/*
+ * Some processors will "pipeline" cache writeback operations, and need an
+ * extra sync instruction to ensure that they are actually flushed out to
+ * memory.  Performing an uncached read (or an I/O operation) without the
+ * flush may cause stale data to be fetched.
+ */
+
+#if defined(CONFIG_CPU_BMIPS3300) || defined(CONFIG_CPU_BMIPS4350) || \
+    defined(CONFIG_CPU_BMIPS4380) || defined(CONFIG_CPU_BMIPS5000)
+
+#define cacheflush_hazard()						\
+do {									\
+	__sync();							\
+} while (0)
+
+#else
+
+#define cacheflush_hazard() do { } while (0)
+
+#endif
+
 #endif /* _ASM_HAZARDS_H */
arch/mips/mm/c-r4k.c +5 -0
index b4923a7..6c113cd 100644
@@ -33,6 +33,7 @@
 #include <asm/mmu_context.h>
 #include <asm/war.h>
 #include <asm/cacheflush.h> /* for run_uncached() */
+#include <asm/hazards.h>
 
 
 /*
@@ -604,6 +605,7 @@ static void r4k_dma_cache_wback_inv(unsigned long addr, unsigned long size)
 			r4k_blast_scache();
 		else
 			blast_scache_range(addr, addr + size);
+		cacheflush_hazard();
 		return;
 	}
 
@@ -620,6 +622,7 @@ static void r4k_dma_cache_wback_inv(unsigned long addr, unsigned long size)
 	}
 
 	bc_wback_inv(addr, size);
+	cacheflush_hazard();
 }
 
 static void r4k_dma_cache_inv(unsigned long addr, unsigned long size)
@@ -647,6 +650,7 @@ static void r4k_dma_cache_inv(unsigned long addr, unsigned long size)
 				 (addr + size - 1) & almask);
 			blast_inv_scache_range(addr, addr + size);
 		}
+		cacheflush_hazard();
 		return;
 	}
 
@@ -663,6 +667,7 @@ static void r4k_dma_cache_inv(unsigned long addr, unsigned long size)
 	}
 
 	bc_inv(addr, size);
+	cacheflush_hazard();
 }
 #endif /* CONFIG_DMA_NONCOHERENT */
 
Signature · 1 line
1.7.0.4
#1 lore raw

PATCH v2 2/6 MIPS: pfn_valid() is broken on low memory HIGHMEM systems

pfn_valid() compares the PFN to max_mapnr:

        __pfn >= min_low_pfn && __pfn < max_mapnr;

On HIGHMEM kernels, highend_pfn is used to set the value of max_mapnr. Unfortunately, highend_pfn is left at zero if the system does not actually have enough RAM to reach into the HIGHMEM range. This causes pfn_valid() to always return false, and when debug checks are enabled the kernel will fail catastrophically:

Memory: 22432k/32768k available (2249k kernel code, 10336k reserved, 653k data, 1352k init, 0k highmem) NR_IRQS:128
kfree_debugcheck: out of range ptr 81c02900h.
Kernel bug detected[#1]:
Cpu 0

$ 0   : 00000000 10008400 00000034 00000000
$ 4   : 8003e160 802a0000 8003e160 00000000
$ 8   : 00000000 0000003e 00000747 00000747
...

On such a configuration, max_low_pfn should be used to set max_mapnr.

---
 arch/mips/mm/init.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/mips/mm/init.c +1 -1
index 2efcbd2..45447cd 100644
@@ -370,7 +370,7 @@ void __init mem_init(void)
 #ifdef CONFIG_DISCONTIGMEM
 #error "CONFIG_HIGHMEM and CONFIG_DISCONTIGMEM dont work together yet"
 #endif
-	max_mapnr = highend_pfn;
+	max_mapnr = highend_pfn ? highend_pfn : max_low_pfn;
 #else
 	max_mapnr = max_low_pfn;
 #endif
Signature · 1 line
1.7.0.4
#2 lore raw

PATCH RESEND v2 3/6 MIPS: Move FIXADDR_TOP into spaces.h

In reply to Kevin Cernekee

Memory maps and addressing quirks are normally defined in <spaces.h>. There are already three targets that need to override FIXADDR_TOP, and others exist. This will be a cleaner approach than adding lots of ifdefs in fixmap.h .

---
 arch/mips/include/asm/fixmap.h              |   10 +---------
 arch/mips/include/asm/mach-bcm63xx/spaces.h |   17 +++++++++++++++++
 arch/mips/include/asm/mach-generic/spaces.h |    4 ++++
 arch/mips/include/asm/mach-tx39xx/spaces.h  |   17 +++++++++++++++++
 arch/mips/include/asm/mach-tx49xx/spaces.h  |   17 +++++++++++++++++
 5 files changed, 56 insertions(+), 9 deletions(-)
 create mode 100644 arch/mips/include/asm/mach-bcm63xx/spaces.h
 create mode 100644 arch/mips/include/asm/mach-tx39xx/spaces.h
 create mode 100644 arch/mips/include/asm/mach-tx49xx/spaces.h
arch/mips/include/asm/fixmap.h +1 -9
index 0b89b83..98bcc98 100644
@@ -14,6 +14,7 @@
 #define _ASM_FIXMAP_H
 
 #include <asm/page.h>
+#include <spaces.h>
 #ifdef CONFIG_HIGHMEM
 #include <linux/threads.h>
 #include <asm/kmap_types.h>
@@ -67,15 +68,6 @@ enum fixed_addresses {
  * the start of the fixmap, and leave one page empty
  * at the top of mem..
  */
-#ifdef CONFIG_BCM63XX
-#define FIXADDR_TOP     ((unsigned long)(long)(int)0xff000000)
-#else
-#if defined(CONFIG_CPU_TX39XX) || defined(CONFIG_CPU_TX49XX)
-#define FIXADDR_TOP	((unsigned long)(long)(int)(0xff000000 - 0x20000))
-#else
-#define FIXADDR_TOP	((unsigned long)(long)(int)0xfffe0000)
-#endif
-#endif
 #define FIXADDR_SIZE	(__end_of_fixed_addresses << PAGE_SHIFT)
 #define FIXADDR_START	(FIXADDR_TOP - FIXADDR_SIZE)
 
arch/mips/include/asm/mach-bcm63xx/spaces.h new file +17 -0
new file mode 100644
index 0000000..61e750f
@@ -0,0 +1,17 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 1994 - 1999, 2000, 03, 04 Ralf Baechle
+ * Copyright (C) 2000, 2002  Maciej W. Rozycki
+ * Copyright (C) 1990, 1999, 2000 Silicon Graphics, Inc.
+ */
+#ifndef _ASM_BCM63XX_SPACES_H
+#define _ASM_BCM63XX_SPACES_H
+
+#define FIXADDR_TOP		((unsigned long)(long)(int)0xff000000)
+
+#include <asm/mach-generic/spaces.h>
+
+#endif /* __ASM_BCM63XX_SPACES_H */
arch/mips/include/asm/mach-generic/spaces.h +4 -0
index c9fa4b1..d7a9efd 100644
@@ -82,4 +82,8 @@
 #define PAGE_OFFSET		(CAC_BASE + PHYS_OFFSET)
 #endif
 
+#ifndef FIXADDR_TOP
+#define FIXADDR_TOP		((unsigned long)(long)(int)0xfffe0000)
+#endif
+
 #endif /* __ASM_MACH_GENERIC_SPACES_H */
arch/mips/include/asm/mach-tx39xx/spaces.h new file +17 -0
new file mode 100644
index 0000000..151fe7a
@@ -0,0 +1,17 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 1994 - 1999, 2000, 03, 04 Ralf Baechle
+ * Copyright (C) 2000, 2002  Maciej W. Rozycki
+ * Copyright (C) 1990, 1999, 2000 Silicon Graphics, Inc.
+ */
+#ifndef _ASM_TX39XX_SPACES_H
+#define _ASM_TX39XX_SPACES_H
+
+#define FIXADDR_TOP		((unsigned long)(long)(int)0xfefe0000)
+
+#include <asm/mach-generic/spaces.h>
+
+#endif /* __ASM_TX39XX_SPACES_H */
arch/mips/include/asm/mach-tx49xx/spaces.h new file +17 -0
new file mode 100644
index 0000000..0cb10a6
@@ -0,0 +1,17 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 1994 - 1999, 2000, 03, 04 Ralf Baechle
+ * Copyright (C) 2000, 2002  Maciej W. Rozycki
+ * Copyright (C) 1990, 1999, 2000 Silicon Graphics, Inc.
+ */
+#ifndef _ASM_TX49XX_SPACES_H
+#define _ASM_TX49XX_SPACES_H
+
+#define FIXADDR_TOP		((unsigned long)(long)(int)0xfefe0000)
+
+#include <asm/mach-generic/spaces.h>
+
+#endif /* __ASM_TX49XX_SPACES_H */
Signature · 1 line
1.7.0.4
#3 lore raw

PATCH RESEND v4 4/6 MIPS: HIGHMEM DMA on noncoherent MIPS32 processors

In reply to Kevin Cernekee

The MIPS DMA coherency functions do not work properly (i.e. kernel oops) when HIGHMEM pages are passed in as arguments. Use kmap_atomic() to temporarily map high pages for cache maintenance operations.

Tested on a 2.6.37-rc7 1GB HIGHMEM SMP no-alias system.

---
 arch/mips/mm/dma-default.c |  114 ++++++++++++++++++++++++++------------------
 1 files changed, 68 insertions(+), 46 deletions(-)
arch/mips/mm/dma-default.c +68 -46
index 21ea14e..4608491 100644
@@ -15,18 +15,18 @@
 #include <linux/scatterlist.h>
 #include <linux/string.h>
 #include <linux/gfp.h>
+#include <linux/highmem.h>
 
 #include <asm/cache.h>
 #include <asm/io.h>
 
 #include <dma-coherence.h>
 
-static inline unsigned long dma_addr_to_virt(struct device *dev,
+static inline struct page *dma_addr_to_page(struct device *dev,
 	dma_addr_t dma_addr)
 {
-	unsigned long addr = plat_dma_addr_to_phys(dev, dma_addr);
-
-	return (unsigned long)phys_to_virt(addr);
+	return pfn_to_page(
+		plat_dma_addr_to_phys(dev, dma_addr) >> PAGE_SHIFT);
 }
 
 /*
@@ -148,20 +148,20 @@ static void mips_dma_free_coherent(struct device *dev, size_t size, void *vaddr,
 	free_pages(addr, get_order(size));
 }
 
-static inline void __dma_sync(unsigned long addr, size_t size,
+static inline void __dma_sync_virtual(void *addr, size_t size,
 	enum dma_data_direction direction)
 {
 	switch (direction) {
 	case DMA_TO_DEVICE:
-		dma_cache_wback(addr, size);
+		dma_cache_wback((unsigned long)addr, size);
 		break;
 
 	case DMA_FROM_DEVICE:
-		dma_cache_inv(addr, size);
+		dma_cache_inv((unsigned long)addr, size);
 		break;
 
 	case DMA_BIDIRECTIONAL:
-		dma_cache_wback_inv(addr, size);
+		dma_cache_wback_inv((unsigned long)addr, size);
 		break;
 
 	default:
@@ -169,12 +169,49 @@ static inline void __dma_sync(unsigned long addr, size_t size,
 	}
 }
 
+/*
+ * A single sg entry may refer to multiple physically contiguous
+ * pages. But we still need to process highmem pages individually.
+ * If highmem is not configured then the bulk of this loop gets
+ * optimized out.
+ */
+static inline void __dma_sync(struct page *page,
+	unsigned long offset, size_t size, enum dma_data_direction direction)
+{
+	size_t left = size;
+
+	do {
+		size_t len = left;
+
+		if (PageHighMem(page)) {
+			void *addr;
+
+			if (offset + len > PAGE_SIZE) {
+				if (offset >= PAGE_SIZE) {
+					page += offset >> PAGE_SHIFT;
+					offset &= ~PAGE_MASK;
+				}
+				len = PAGE_SIZE - offset;
+			}
+
+			addr = kmap_atomic(page);
+			__dma_sync_virtual(addr + offset, len, direction);
+			kunmap_atomic(addr);
+		} else
+			__dma_sync_virtual(page_address(page) + offset,
+					   size, direction);
+		offset = 0;
+		page++;
+		left -= len;
+	} while (left);
+}
+
 static void mips_dma_unmap_page(struct device *dev, dma_addr_t dma_addr,
 	size_t size, enum dma_data_direction direction, struct dma_attrs *attrs)
 {
 	if (cpu_is_noncoherent_r10000(dev))
-		__dma_sync(dma_addr_to_virt(dev, dma_addr), size,
-		           direction);
+		__dma_sync(dma_addr_to_page(dev, dma_addr),
+			   dma_addr & ~PAGE_MASK, size, direction);
 
 	plat_unmap_dma_mem(dev, dma_addr, size, direction);
 }
@@ -185,13 +222,11 @@ static int mips_dma_map_sg(struct device *dev, struct scatterlist *sg,
 	int i;
 
 	for (i = 0; i < nents; i++, sg++) {
-		unsigned long addr;
-
-		addr = (unsigned long) sg_virt(sg);
-		if (!plat_device_is_coherent(dev) && addr)
-			__dma_sync(addr, sg->length, direction);
-		sg->dma_address = plat_map_dma_mem(dev,
-				                   (void *)addr, sg->length);
+		if (!plat_device_is_coherent(dev))
+			__dma_sync(sg_page(sg), sg->offset, sg->length,
+				   direction);
+		sg->dma_address = plat_map_dma_mem_page(dev, sg_page(sg)) +
+				  sg->offset;
 	}
 
 	return nents;
@@ -201,30 +236,23 @@ static dma_addr_t mips_dma_map_page(struct device *dev, struct page *page,
 	unsigned long offset, size_t size, enum dma_data_direction direction,
 	struct dma_attrs *attrs)
 {
-	unsigned long addr;
-
-	addr = (unsigned long) page_address(page) + offset;
-
 	if (!plat_device_is_coherent(dev))
-		__dma_sync(addr, size, direction);
+		__dma_sync(page, offset, size, direction);
 
-	return plat_map_dma_mem(dev, (void *)addr, size);
+	return plat_map_dma_mem_page(dev, page) + offset;
 }
 
 static void mips_dma_unmap_sg(struct device *dev, struct scatterlist *sg,
 	int nhwentries, enum dma_data_direction direction,
 	struct dma_attrs *attrs)
 {
-	unsigned long addr;
 	int i;
 
 	for (i = 0; i < nhwentries; i++, sg++) {
 		if (!plat_device_is_coherent(dev) &&
-		    direction != DMA_TO_DEVICE) {
-			addr = (unsigned long) sg_virt(sg);
-			if (addr)
-				__dma_sync(addr, sg->length, direction);
-		}
+		    direction != DMA_TO_DEVICE)
+			__dma_sync(sg_page(sg), sg->offset, sg->length,
+				   direction);
 		plat_unmap_dma_mem(dev, sg->dma_address, sg->length, direction);
 	}
 }
@@ -232,24 +260,18 @@ static void mips_dma_unmap_sg(struct device *dev, struct scatterlist *sg,
 static void mips_dma_sync_single_for_cpu(struct device *dev,
 	dma_addr_t dma_handle, size_t size, enum dma_data_direction direction)
 {
-	if (cpu_is_noncoherent_r10000(dev)) {
-		unsigned long addr;
-
-		addr = dma_addr_to_virt(dev, dma_handle);
-		__dma_sync(addr, size, direction);
-	}
+	if (cpu_is_noncoherent_r10000(dev))
+		__dma_sync(dma_addr_to_page(dev, dma_handle),
+			   dma_handle & ~PAGE_MASK, size, direction);
 }
 
 static void mips_dma_sync_single_for_device(struct device *dev,
 	dma_addr_t dma_handle, size_t size, enum dma_data_direction direction)
 {
 	plat_extra_sync_for_device(dev);
-	if (!plat_device_is_coherent(dev)) {
-		unsigned long addr;
-
-		addr = dma_addr_to_virt(dev, dma_handle);
-		__dma_sync(addr, size, direction);
-	}
+	if (!plat_device_is_coherent(dev))
+		__dma_sync(dma_addr_to_page(dev, dma_handle),
+			   dma_handle & ~PAGE_MASK, size, direction);
 }
 
 static void mips_dma_sync_sg_for_cpu(struct device *dev,
@@ -260,8 +282,8 @@ static void mips_dma_sync_sg_for_cpu(struct device *dev,
 	/* Make sure that gcc doesn't leave the empty loop body.  */
 	for (i = 0; i < nelems; i++, sg++) {
 		if (cpu_is_noncoherent_r10000(dev))
-			__dma_sync((unsigned long)page_address(sg_page(sg)),
-			           sg->length, direction);
+			__dma_sync(sg_page(sg), sg->offset, sg->length,
+				   direction);
 	}
 }
 
@@ -273,8 +295,8 @@ static void mips_dma_sync_sg_for_device(struct device *dev,
 	/* Make sure that gcc doesn't leave the empty loop body.  */
 	for (i = 0; i < nelems; i++, sg++) {
 		if (!plat_device_is_coherent(dev))
-			__dma_sync((unsigned long)page_address(sg_page(sg)),
-			           sg->length, direction);
+			__dma_sync(sg_page(sg), sg->offset, sg->length,
+				   direction);
 	}
 }
 
@@ -295,7 +317,7 @@ void dma_cache_sync(struct device *dev, void *vaddr, size_t size,
 
 	plat_extra_sync_for_device(dev);
 	if (!plat_device_is_coherent(dev))
-		__dma_sync((unsigned long)vaddr, size, direction);
+		__dma_sync_virtual(vaddr, size, direction);
 }
 
 EXPORT_SYMBOL(dma_cache_sync);
Signature · 1 line
1.7.0.4
#4 lore raw

PATCH RESEND 5/6 MIPS: Install handlers for BMIPS software IRQs

In reply to Kevin Cernekee

BMIPS4350/4380/5000 CMT/SMT all use SW INT0/INT1 for inter-thread signaling.

---
 arch/mips/kernel/irq_cpu.c |   14 ++++++--------
 1 files changed, 6 insertions(+), 8 deletions(-)
arch/mips/kernel/irq_cpu.c +6 -8
index 0262abe..70d4736 100644
@@ -107,14 +107,12 @@ void __init mips_cpu_irq_init(void)
 	clear_c0_status(ST0_IM);
 	clear_c0_cause(CAUSEF_IP);
 
-	/*
-	 * Only MT is using the software interrupts currently, so we just
-	 * leave them uninitialized for other processors.
-	 */
-	if (cpu_has_mipsmt)
-		for (i = irq_base; i < irq_base + 2; i++)
-			set_irq_chip_and_handler(i, &mips_mt_cpu_irq_controller,
-						 handle_percpu_irq);
+	/* Software interrupts are used for MT/CMT IPI */
+	for (i = irq_base; i < irq_base + 2; i++)
+		set_irq_chip_and_handler(i, cpu_has_mipsmt ?
+					 &mips_mt_cpu_irq_controller :
+					 &mips_cpu_irq_controller,
+					 handle_percpu_irq);
 
 	for (i = irq_base + 2; i < irq_base + 8; i++)
 		set_irq_chip_and_handler(i, &mips_cpu_irq_controller,
Signature · 1 line
1.7.0.4
#5 lore raw

PATCH RESEND 6/6 MIPS: Limit fixrange_init() to the FIXMAP region

In reply to Kevin Cernekee

fixrange_init() allocates page tables for all addresses higher than FIXADDR_TOP. On processors that override the default FIXADDR_TOP address of 0xfffe_0000, this can consume up to 4 pages (1 page per 4MB) for pgd's that are never used.

---
 arch/mips/mm/init.c       |    6 +++---
 arch/mips/mm/pgtable-32.c |    2 +-
 arch/mips/mm/pgtable-64.c |    2 +-
 3 files changed, 5 insertions(+), 5 deletions(-)
arch/mips/mm/init.c +3 -3
index 45447cd..14a89bd 100644
@@ -279,11 +279,11 @@ void __init fixrange_init(unsigned long start, unsigned long end,
 	k = __pmd_offset(vaddr);
 	pgd = pgd_base + i;
 
-	for ( ; (i < PTRS_PER_PGD) && (vaddr != end); pgd++, i++) {
+	for ( ; (i < PTRS_PER_PGD) && (vaddr < end); pgd++, i++) {
 		pud = (pud_t *)pgd;
-		for ( ; (j < PTRS_PER_PUD) && (vaddr != end); pud++, j++) {
+		for ( ; (j < PTRS_PER_PUD) && (vaddr < end); pud++, j++) {
 			pmd = (pmd_t *)pud;
-			for (; (k < PTRS_PER_PMD) && (vaddr != end); pmd++, k++) {
+			for (; (k < PTRS_PER_PMD) && (vaddr < end); pmd++, k++) {
 				if (pmd_none(*pmd)) {
 					pte = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
 					set_pmd(pmd, __pmd((unsigned long)pte));
arch/mips/mm/pgtable-32.c +1 -1
index 575e401..adc6911 100644
@@ -52,7 +52,7 @@ void __init pagetable_init(void)
 	 * Fixed mappings:
 	 */
 	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
-	fixrange_init(vaddr, 0, pgd_base);
+	fixrange_init(vaddr, vaddr + FIXADDR_SIZE, pgd_base);
 
 #ifdef CONFIG_HIGHMEM
 	/*
arch/mips/mm/pgtable-64.c +1 -1
index 78eaa4f..cda4e30 100644
@@ -76,5 +76,5 @@ void __init pagetable_init(void)
 	 * Fixed mappings:
 	 */
 	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
-	fixrange_init(vaddr, 0, pgd_base);
+	fixrange_init(vaddr, vaddr + FIXADDR_SIZE, pgd_base);
 }
Signature · 1 line
1.7.0.4