From 70a7d3cc1308a55104fbe505d76f2aca8a4cf53e Mon Sep 17 00:00:00 2001 From: Jeremy Fitzhardinge Date: Mon, 22 Dec 2008 10:26:05 -0800 Subject: [PATCH 01/13] swiotlb: add hwdev to swiotlb_phys_to_bus() / swiotlb_sg_to_bus() Impact: extend functions with a (yet unused) parameter, update callsites Some architectures need it - in preparation for highmem swiotlb. Signed-off-by: Jeremy Fitzhardinge Signed-off-by: Ingo Molnar --- arch/x86/kernel/pci-swiotlb_64.c | 2 +- include/linux/swiotlb.h | 3 +- lib/swiotlb.c | 53 +++++++++++++------------------- 3 files changed, 25 insertions(+), 33 deletions(-) diff --git a/arch/x86/kernel/pci-swiotlb_64.c b/arch/x86/kernel/pci-swiotlb_64.c index 242c3440687..6cf8a816dc2 100644 --- a/arch/x86/kernel/pci-swiotlb_64.c +++ b/arch/x86/kernel/pci-swiotlb_64.c @@ -23,7 +23,7 @@ void *swiotlb_alloc(unsigned order, unsigned long nslabs) return (void *)__get_free_pages(GFP_DMA | __GFP_NOWARN, order); } -dma_addr_t swiotlb_phys_to_bus(phys_addr_t paddr) +dma_addr_t swiotlb_phys_to_bus(struct device *hwdev, phys_addr_t paddr) { return paddr; } diff --git a/include/linux/swiotlb.h b/include/linux/swiotlb.h index 325af1de035..dedd3c0cfe3 100644 --- a/include/linux/swiotlb.h +++ b/include/linux/swiotlb.h @@ -27,7 +27,8 @@ swiotlb_init(void); extern void *swiotlb_alloc_boot(size_t bytes, unsigned long nslabs); extern void *swiotlb_alloc(unsigned order, unsigned long nslabs); -extern dma_addr_t swiotlb_phys_to_bus(phys_addr_t address); +extern dma_addr_t swiotlb_phys_to_bus(struct device *hwdev, + phys_addr_t address); extern phys_addr_t swiotlb_bus_to_phys(dma_addr_t address); extern int swiotlb_arch_range_needs_mapping(void *ptr, size_t size); diff --git a/lib/swiotlb.c b/lib/swiotlb.c index fa2dc4e5f9b..3657da8ebbc 100644 --- a/lib/swiotlb.c +++ b/lib/swiotlb.c @@ -126,7 +126,7 @@ void * __weak swiotlb_alloc(unsigned order, unsigned long nslabs) return (void *)__get_free_pages(GFP_DMA | __GFP_NOWARN, order); } -dma_addr_t __weak swiotlb_phys_to_bus(phys_addr_t paddr) +dma_addr_t __weak swiotlb_phys_to_bus(struct device *hwdev, phys_addr_t paddr) { return paddr; } @@ -136,9 +136,10 @@ phys_addr_t __weak swiotlb_bus_to_phys(dma_addr_t baddr) return baddr; } -static dma_addr_t swiotlb_virt_to_bus(volatile void *address) +static dma_addr_t swiotlb_virt_to_bus(struct device *hwdev, + volatile void *address) { - return swiotlb_phys_to_bus(virt_to_phys(address)); + return swiotlb_phys_to_bus(hwdev, virt_to_phys(address)); } static void *swiotlb_bus_to_virt(dma_addr_t address) @@ -151,35 +152,23 @@ int __weak swiotlb_arch_range_needs_mapping(void *ptr, size_t size) return 0; } -static dma_addr_t swiotlb_sg_to_bus(struct scatterlist *sg) +static dma_addr_t swiotlb_sg_to_bus(struct device *hwdev, struct scatterlist *sg) { - return swiotlb_phys_to_bus(page_to_phys(sg_page(sg)) + sg->offset); + return swiotlb_phys_to_bus(hwdev, page_to_phys(sg_page(sg)) + sg->offset); } static void swiotlb_print_info(unsigned long bytes) { phys_addr_t pstart, pend; - dma_addr_t bstart, bend; pstart = virt_to_phys(io_tlb_start); pend = virt_to_phys(io_tlb_end); - bstart = swiotlb_phys_to_bus(pstart); - bend = swiotlb_phys_to_bus(pend); - printk(KERN_INFO "Placing %luMB software IO TLB between %p - %p\n", bytes >> 20, io_tlb_start, io_tlb_end); - if (pstart != bstart || pend != bend) - printk(KERN_INFO "software IO TLB at phys %#llx - %#llx" - " bus %#llx - %#llx\n", - (unsigned long long)pstart, - (unsigned long long)pend, - (unsigned long long)bstart, - (unsigned long long)bend); - else - printk(KERN_INFO "software IO TLB at phys %#llx - %#llx\n", - (unsigned long long)pstart, - (unsigned long long)pend); + printk(KERN_INFO "software IO TLB at phys %#llx - %#llx\n", + (unsigned long long)pstart, + (unsigned long long)pend); } /* @@ -406,7 +395,7 @@ map_single(struct device *hwdev, struct swiotlb_phys_addr buffer, size_t size, i struct swiotlb_phys_addr slot_buf; mask = dma_get_seg_boundary(hwdev); - start_dma_addr = swiotlb_virt_to_bus(io_tlb_start) & mask; + start_dma_addr = swiotlb_virt_to_bus(hwdev, io_tlb_start) & mask; offset_slots = ALIGN(start_dma_addr, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; @@ -585,7 +574,9 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size, dma_mask = hwdev->coherent_dma_mask; ret = (void *)__get_free_pages(flags, order); - if (ret && !is_buffer_dma_capable(dma_mask, swiotlb_virt_to_bus(ret), size)) { + if (ret && + !is_buffer_dma_capable(dma_mask, swiotlb_virt_to_bus(hwdev, ret), + size)) { /* * The allocated memory isn't reachable by the device. * Fall back on swiotlb_map_single(). @@ -609,7 +600,7 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size, } memset(ret, 0, size); - dev_addr = swiotlb_virt_to_bus(ret); + dev_addr = swiotlb_virt_to_bus(hwdev, ret); /* Confirm address can be DMA'd by device */ if (!is_buffer_dma_capable(dma_mask, dev_addr, size)) { @@ -669,7 +660,7 @@ dma_addr_t swiotlb_map_single_attrs(struct device *hwdev, void *ptr, size_t size, int dir, struct dma_attrs *attrs) { - dma_addr_t dev_addr = swiotlb_virt_to_bus(ptr); + dma_addr_t dev_addr = swiotlb_virt_to_bus(hwdev, ptr); void *map; struct swiotlb_phys_addr buffer; @@ -694,7 +685,7 @@ swiotlb_map_single_attrs(struct device *hwdev, void *ptr, size_t size, map = io_tlb_overflow_buffer; } - dev_addr = swiotlb_virt_to_bus(map); + dev_addr = swiotlb_virt_to_bus(hwdev, map); /* * Ensure that the address returned is DMA'ble @@ -840,7 +831,7 @@ swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, int nelems, BUG_ON(dir == DMA_NONE); for_each_sg(sgl, sg, nelems, i) { - dev_addr = swiotlb_sg_to_bus(sg); + dev_addr = swiotlb_sg_to_bus(hwdev, sg); if (range_needs_mapping(sg_virt(sg), sg->length) || address_needs_mapping(hwdev, dev_addr, sg->length)) { void *map; @@ -856,7 +847,7 @@ swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, int nelems, sgl[0].dma_length = 0; return 0; } - sg->dma_address = swiotlb_virt_to_bus(map); + sg->dma_address = swiotlb_virt_to_bus(hwdev, map); } else sg->dma_address = dev_addr; sg->dma_length = sg->length; @@ -886,7 +877,7 @@ swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl, BUG_ON(dir == DMA_NONE); for_each_sg(sgl, sg, nelems, i) { - if (sg->dma_address != swiotlb_sg_to_bus(sg)) + if (sg->dma_address != swiotlb_sg_to_bus(hwdev, sg)) unmap_single(hwdev, swiotlb_bus_to_virt(sg->dma_address), sg->dma_length, dir); else if (dir == DMA_FROM_DEVICE) @@ -919,7 +910,7 @@ swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl, BUG_ON(dir == DMA_NONE); for_each_sg(sgl, sg, nelems, i) { - if (sg->dma_address != swiotlb_sg_to_bus(sg)) + if (sg->dma_address != swiotlb_sg_to_bus(hwdev, sg)) sync_single(hwdev, swiotlb_bus_to_virt(sg->dma_address), sg->dma_length, dir, target); else if (dir == DMA_FROM_DEVICE) @@ -944,7 +935,7 @@ swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg, int swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr) { - return (dma_addr == swiotlb_virt_to_bus(io_tlb_overflow_buffer)); + return (dma_addr == swiotlb_virt_to_bus(hwdev, io_tlb_overflow_buffer)); } /* @@ -956,7 +947,7 @@ swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr) int swiotlb_dma_supported(struct device *hwdev, u64 mask) { - return swiotlb_virt_to_bus(io_tlb_end - 1) <= mask; + return swiotlb_virt_to_bus(hwdev, io_tlb_end - 1) <= mask; } EXPORT_SYMBOL(swiotlb_map_single); From bc40ac66988a7721f2a244b6df65f8c13d16479c Mon Sep 17 00:00:00 2001 From: Becky Bruce Date: Mon, 22 Dec 2008 10:26:08 -0800 Subject: [PATCH 02/13] swiotlb: store phys address in io_tlb_orig_addr array Impact: refactor code, cleanup When we enable swiotlb for platforms that support HIGHMEM, we can no longer store the virtual address of the original dma buffer, because that buffer might not have a permament mapping. Change the swiotlb code to instead store the physical address of the original buffer. Signed-off-by: Becky Bruce Signed-off-by: Ingo Molnar --- lib/swiotlb.c | 120 +++++++++++++------------------------------------- 1 file changed, 30 insertions(+), 90 deletions(-) diff --git a/lib/swiotlb.c b/lib/swiotlb.c index 3657da8ebbc..98a7a4450e0 100644 --- a/lib/swiotlb.c +++ b/lib/swiotlb.c @@ -89,10 +89,7 @@ static unsigned int io_tlb_index; * We need to save away the original address corresponding to a mapped entry * for the sync operations. */ -static struct swiotlb_phys_addr { - struct page *page; - unsigned int offset; -} *io_tlb_orig_addr; +static phys_addr_t *io_tlb_orig_addr; /* * Protect the above data structures in the map and unmap calls @@ -204,7 +201,7 @@ swiotlb_init_with_default_size(size_t default_size) for (i = 0; i < io_tlb_nslabs; i++) io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE); io_tlb_index = 0; - io_tlb_orig_addr = alloc_bootmem(io_tlb_nslabs * sizeof(struct swiotlb_phys_addr)); + io_tlb_orig_addr = alloc_bootmem(io_tlb_nslabs * sizeof(phys_addr_t)); /* * Get the overflow emergency buffer @@ -278,12 +275,14 @@ swiotlb_late_init_with_default_size(size_t default_size) io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE); io_tlb_index = 0; - io_tlb_orig_addr = (struct swiotlb_phys_addr *)__get_free_pages(GFP_KERNEL, - get_order(io_tlb_nslabs * sizeof(struct swiotlb_phys_addr))); + io_tlb_orig_addr = (phys_addr_t *) + __get_free_pages(GFP_KERNEL, + get_order(io_tlb_nslabs * + sizeof(phys_addr_t))); if (!io_tlb_orig_addr) goto cleanup3; - memset(io_tlb_orig_addr, 0, io_tlb_nslabs * sizeof(struct swiotlb_phys_addr)); + memset(io_tlb_orig_addr, 0, io_tlb_nslabs * sizeof(phys_addr_t)); /* * Get the overflow emergency buffer @@ -298,8 +297,8 @@ swiotlb_late_init_with_default_size(size_t default_size) return 0; cleanup4: - free_pages((unsigned long)io_tlb_orig_addr, get_order(io_tlb_nslabs * - sizeof(char *))); + free_pages((unsigned long)io_tlb_orig_addr, + get_order(io_tlb_nslabs * sizeof(phys_addr_t))); io_tlb_orig_addr = NULL; cleanup3: free_pages((unsigned long)io_tlb_list, get_order(io_tlb_nslabs * @@ -330,59 +329,11 @@ static int is_swiotlb_buffer(char *addr) return addr >= io_tlb_start && addr < io_tlb_end; } -static struct swiotlb_phys_addr swiotlb_bus_to_phys_addr(char *dma_addr) -{ - int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT; - struct swiotlb_phys_addr buffer = io_tlb_orig_addr[index]; - buffer.offset += (long)dma_addr & ((1 << IO_TLB_SHIFT) - 1); - buffer.page += buffer.offset >> PAGE_SHIFT; - buffer.offset &= PAGE_SIZE - 1; - return buffer; -} - -static void -__sync_single(struct swiotlb_phys_addr buffer, char *dma_addr, size_t size, int dir) -{ - if (PageHighMem(buffer.page)) { - size_t len, bytes; - char *dev, *host, *kmp; - - len = size; - while (len != 0) { - unsigned long flags; - - bytes = len; - if ((bytes + buffer.offset) > PAGE_SIZE) - bytes = PAGE_SIZE - buffer.offset; - local_irq_save(flags); /* protects KM_BOUNCE_READ */ - kmp = kmap_atomic(buffer.page, KM_BOUNCE_READ); - dev = dma_addr + size - len; - host = kmp + buffer.offset; - if (dir == DMA_FROM_DEVICE) - memcpy(host, dev, bytes); - else - memcpy(dev, host, bytes); - kunmap_atomic(kmp, KM_BOUNCE_READ); - local_irq_restore(flags); - len -= bytes; - buffer.page++; - buffer.offset = 0; - } - } else { - void *v = page_address(buffer.page) + buffer.offset; - - if (dir == DMA_TO_DEVICE) - memcpy(dma_addr, v, size); - else - memcpy(v, dma_addr, size); - } -} - /* * Allocates bounce buffer and returns its kernel virtual address. */ static void * -map_single(struct device *hwdev, struct swiotlb_phys_addr buffer, size_t size, int dir) +map_single(struct device *hwdev, phys_addr_t phys, size_t size, int dir) { unsigned long flags; char *dma_addr; @@ -392,7 +343,6 @@ map_single(struct device *hwdev, struct swiotlb_phys_addr buffer, size_t size, i unsigned long mask; unsigned long offset_slots; unsigned long max_slots; - struct swiotlb_phys_addr slot_buf; mask = dma_get_seg_boundary(hwdev); start_dma_addr = swiotlb_virt_to_bus(hwdev, io_tlb_start) & mask; @@ -477,15 +427,10 @@ found: * This is needed when we sync the memory. Then we sync the buffer if * needed. */ - slot_buf = buffer; - for (i = 0; i < nslots; i++) { - slot_buf.page += slot_buf.offset >> PAGE_SHIFT; - slot_buf.offset &= PAGE_SIZE - 1; - io_tlb_orig_addr[index+i] = slot_buf; - slot_buf.offset += 1 << IO_TLB_SHIFT; - } + for (i = 0; i < nslots; i++) + io_tlb_orig_addr[index+i] = phys + (i << IO_TLB_SHIFT); if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) - __sync_single(buffer, dma_addr, size, DMA_TO_DEVICE); + memcpy(dma_addr, phys_to_virt(phys), size); return dma_addr; } @@ -499,17 +444,17 @@ unmap_single(struct device *hwdev, char *dma_addr, size_t size, int dir) unsigned long flags; int i, count, nslots = ALIGN(size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT; - struct swiotlb_phys_addr buffer = swiotlb_bus_to_phys_addr(dma_addr); + phys_addr_t phys = io_tlb_orig_addr[index]; /* * First, sync the memory before unmapping the entry */ - if ((dir == DMA_FROM_DEVICE) || (dir == DMA_BIDIRECTIONAL)) + if (phys && ((dir == DMA_FROM_DEVICE) || (dir == DMA_BIDIRECTIONAL))) /* * bounce... copy the data back into the original buffer * and * delete the bounce buffer. */ - __sync_single(buffer, dma_addr, size, DMA_FROM_DEVICE); + memcpy(phys_to_virt(phys), dma_addr, size); /* * Return the buffer to the free list by setting the corresponding @@ -541,18 +486,21 @@ static void sync_single(struct device *hwdev, char *dma_addr, size_t size, int dir, int target) { - struct swiotlb_phys_addr buffer = swiotlb_bus_to_phys_addr(dma_addr); + int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT; + phys_addr_t phys = io_tlb_orig_addr[index]; + + phys += ((unsigned long)dma_addr & ((1 << IO_TLB_SHIFT) - 1)); switch (target) { case SYNC_FOR_CPU: if (likely(dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)) - __sync_single(buffer, dma_addr, size, DMA_FROM_DEVICE); + memcpy(phys_to_virt(phys), dma_addr, size); else BUG_ON(dir != DMA_TO_DEVICE); break; case SYNC_FOR_DEVICE: if (likely(dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL)) - __sync_single(buffer, dma_addr, size, DMA_TO_DEVICE); + memcpy(dma_addr, phys_to_virt(phys), size); else BUG_ON(dir != DMA_FROM_DEVICE); break; @@ -591,10 +539,7 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size, * swiotlb_map_single(), which will grab memory from * the lowest available address range. */ - struct swiotlb_phys_addr buffer; - buffer.page = virt_to_page(NULL); - buffer.offset = 0; - ret = map_single(hwdev, buffer, size, DMA_FROM_DEVICE); + ret = map_single(hwdev, 0, size, DMA_FROM_DEVICE); if (!ret) return NULL; } @@ -662,7 +607,6 @@ swiotlb_map_single_attrs(struct device *hwdev, void *ptr, size_t size, { dma_addr_t dev_addr = swiotlb_virt_to_bus(hwdev, ptr); void *map; - struct swiotlb_phys_addr buffer; BUG_ON(dir == DMA_NONE); /* @@ -677,9 +621,7 @@ swiotlb_map_single_attrs(struct device *hwdev, void *ptr, size_t size, /* * Oh well, have to allocate and map a bounce buffer. */ - buffer.page = virt_to_page(ptr); - buffer.offset = (unsigned long)ptr & ~PAGE_MASK; - map = map_single(hwdev, buffer, size, dir); + map = map_single(hwdev, virt_to_phys(ptr), size, dir); if (!map) { swiotlb_full(hwdev, size, dir, 1); map = io_tlb_overflow_buffer; @@ -824,20 +766,18 @@ swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, int nelems, int dir, struct dma_attrs *attrs) { struct scatterlist *sg; - struct swiotlb_phys_addr buffer; - dma_addr_t dev_addr; int i; BUG_ON(dir == DMA_NONE); for_each_sg(sgl, sg, nelems, i) { - dev_addr = swiotlb_sg_to_bus(hwdev, sg); - if (range_needs_mapping(sg_virt(sg), sg->length) || + void *addr = sg_virt(sg); + dma_addr_t dev_addr = swiotlb_virt_to_bus(hwdev, addr); + + if (range_needs_mapping(addr, sg->length) || address_needs_mapping(hwdev, dev_addr, sg->length)) { - void *map; - buffer.page = sg_page(sg); - buffer.offset = sg->offset; - map = map_single(hwdev, buffer, sg->length, dir); + void *map = map_single(hwdev, sg_phys(sg), + sg->length, dir); if (!map) { /* Don't panic here, we expect map_sg users to do proper error handling. */ From fb05a37929e0cd99016b4f5e5a5ef077fb10a947 Mon Sep 17 00:00:00 2001 From: Becky Bruce Date: Mon, 22 Dec 2008 10:26:09 -0800 Subject: [PATCH 03/13] swiotlb: add support for systems with highmem Impact: extend code for highmem - existing users unaffected On highmem systems, the original dma buffer might not have a virtual mapping - we need to kmap it in to perform the bounce. Extract the code that does the actual copy into a function that does the kmap if highmem is enabled, and default to the normal swiotlb memcpy if not. [ ported by Jeremy Fitzhardinge ] Signed-off-by: Becky Bruce Signed-off-by: Jeremy Fitzhardinge Signed-off-by: Ingo Molnar --- lib/swiotlb.c | 68 ++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 51 insertions(+), 17 deletions(-) diff --git a/lib/swiotlb.c b/lib/swiotlb.c index 98a7a4450e0..785046e4c3a 100644 --- a/lib/swiotlb.c +++ b/lib/swiotlb.c @@ -14,6 +14,7 @@ * 04/07/.. ak Better overflow handling. Assorted fixes. * 05/09/10 linville Add support for syncing ranges, support syncing for * DMA_BIDIRECTIONAL mappings, miscellaneous cleanup. + * 08/12/11 beckyb Add highmem support */ #include @@ -24,6 +25,7 @@ #include #include #include +#include #include #include #include @@ -149,11 +151,6 @@ int __weak swiotlb_arch_range_needs_mapping(void *ptr, size_t size) return 0; } -static dma_addr_t swiotlb_sg_to_bus(struct device *hwdev, struct scatterlist *sg) -{ - return swiotlb_phys_to_bus(hwdev, page_to_phys(sg_page(sg)) + sg->offset); -} - static void swiotlb_print_info(unsigned long bytes) { phys_addr_t pstart, pend; @@ -329,6 +326,47 @@ static int is_swiotlb_buffer(char *addr) return addr >= io_tlb_start && addr < io_tlb_end; } +/* + * Bounce: copy the swiotlb buffer back to the original dma location + */ +static void swiotlb_bounce(phys_addr_t phys, char *dma_addr, size_t size, + enum dma_data_direction dir) +{ + unsigned long pfn = PFN_DOWN(phys); + + if (PageHighMem(pfn_to_page(pfn))) { + /* The buffer does not have a mapping. Map it in and copy */ + unsigned int offset = phys & ~PAGE_MASK; + char *buffer; + unsigned int sz = 0; + unsigned long flags; + + while (size) { + sz = min(PAGE_SIZE - offset, size); + + local_irq_save(flags); + buffer = kmap_atomic(pfn_to_page(pfn), + KM_BOUNCE_READ); + if (dir == DMA_TO_DEVICE) + memcpy(dma_addr, buffer + offset, sz); + else + memcpy(buffer + offset, dma_addr, sz); + kunmap_atomic(buffer, KM_BOUNCE_READ); + local_irq_restore(flags); + + size -= sz; + pfn++; + dma_addr += sz; + offset = 0; + } + } else { + if (dir == DMA_TO_DEVICE) + memcpy(dma_addr, phys_to_virt(phys), size); + else + memcpy(phys_to_virt(phys), dma_addr, size); + } +} + /* * Allocates bounce buffer and returns its kernel virtual address. */ @@ -430,7 +468,7 @@ found: for (i = 0; i < nslots; i++) io_tlb_orig_addr[index+i] = phys + (i << IO_TLB_SHIFT); if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) - memcpy(dma_addr, phys_to_virt(phys), size); + swiotlb_bounce(phys, dma_addr, size, DMA_TO_DEVICE); return dma_addr; } @@ -450,11 +488,7 @@ unmap_single(struct device *hwdev, char *dma_addr, size_t size, int dir) * First, sync the memory before unmapping the entry */ if (phys && ((dir == DMA_FROM_DEVICE) || (dir == DMA_BIDIRECTIONAL))) - /* - * bounce... copy the data back into the original buffer * and - * delete the bounce buffer. - */ - memcpy(phys_to_virt(phys), dma_addr, size); + swiotlb_bounce(phys, dma_addr, size, DMA_FROM_DEVICE); /* * Return the buffer to the free list by setting the corresponding @@ -494,13 +528,13 @@ sync_single(struct device *hwdev, char *dma_addr, size_t size, switch (target) { case SYNC_FOR_CPU: if (likely(dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)) - memcpy(phys_to_virt(phys), dma_addr, size); + swiotlb_bounce(phys, dma_addr, size, DMA_FROM_DEVICE); else BUG_ON(dir != DMA_TO_DEVICE); break; case SYNC_FOR_DEVICE: if (likely(dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL)) - memcpy(dma_addr, phys_to_virt(phys), size); + swiotlb_bounce(phys, dma_addr, size, DMA_TO_DEVICE); else BUG_ON(dir != DMA_FROM_DEVICE); break; @@ -817,11 +851,11 @@ swiotlb_unmap_sg_attrs(struct device *hwdev, struct scatterlist *sgl, BUG_ON(dir == DMA_NONE); for_each_sg(sgl, sg, nelems, i) { - if (sg->dma_address != swiotlb_sg_to_bus(hwdev, sg)) + if (sg->dma_address != swiotlb_virt_to_bus(hwdev, sg_virt(sg))) unmap_single(hwdev, swiotlb_bus_to_virt(sg->dma_address), sg->dma_length, dir); else if (dir == DMA_FROM_DEVICE) - dma_mark_clean(swiotlb_bus_to_virt(sg->dma_address), sg->dma_length); + dma_mark_clean(sg_virt(sg), sg->dma_length); } } EXPORT_SYMBOL(swiotlb_unmap_sg_attrs); @@ -850,11 +884,11 @@ swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl, BUG_ON(dir == DMA_NONE); for_each_sg(sgl, sg, nelems, i) { - if (sg->dma_address != swiotlb_sg_to_bus(hwdev, sg)) + if (sg->dma_address != swiotlb_virt_to_bus(hwdev, sg_virt(sg))) sync_single(hwdev, swiotlb_bus_to_virt(sg->dma_address), sg->dma_length, dir, target); else if (dir == DMA_FROM_DEVICE) - dma_mark_clean(swiotlb_bus_to_virt(sg->dma_address), sg->dma_length); + dma_mark_clean(sg_virt(sg), sg->dma_length); } } From 1da4f9894c243a9c58c505fd8f3e6cc0709a8825 Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Sun, 28 Dec 2008 15:02:05 +0900 Subject: [PATCH 04/13] swiotlb: replace architecture-specific swiotlb.h with linux/swiotlb.h Impact: cleanup This replaces architecture-specific swiotlb.h (X86 and IA64) with linux/swiotlb.h. Signed-off-by: FUJITA Tomonori Signed-off-by: Ingo Molnar --- arch/ia64/include/asm/swiotlb.h | 39 +-------------------------------- arch/x86/include/asm/swiotlb.h | 38 +------------------------------- 2 files changed, 2 insertions(+), 75 deletions(-) diff --git a/arch/ia64/include/asm/swiotlb.h b/arch/ia64/include/asm/swiotlb.h index fb79423834d..dcbaea7ce12 100644 --- a/arch/ia64/include/asm/swiotlb.h +++ b/arch/ia64/include/asm/swiotlb.h @@ -2,44 +2,7 @@ #define ASM_IA64__SWIOTLB_H #include - -/* SWIOTLB interface */ - -extern dma_addr_t swiotlb_map_single(struct device *hwdev, void *ptr, - size_t size, int dir); -extern void *swiotlb_alloc_coherent(struct device *hwdev, size_t size, - dma_addr_t *dma_handle, gfp_t flags); -extern void swiotlb_unmap_single(struct device *hwdev, dma_addr_t dev_addr, - size_t size, int dir); -extern void swiotlb_sync_single_for_cpu(struct device *hwdev, - dma_addr_t dev_addr, - size_t size, int dir); -extern void swiotlb_sync_single_for_device(struct device *hwdev, - dma_addr_t dev_addr, - size_t size, int dir); -extern void swiotlb_sync_single_range_for_cpu(struct device *hwdev, - dma_addr_t dev_addr, - unsigned long offset, - size_t size, int dir); -extern void swiotlb_sync_single_range_for_device(struct device *hwdev, - dma_addr_t dev_addr, - unsigned long offset, - size_t size, int dir); -extern void swiotlb_sync_sg_for_cpu(struct device *hwdev, - struct scatterlist *sg, int nelems, - int dir); -extern void swiotlb_sync_sg_for_device(struct device *hwdev, - struct scatterlist *sg, int nelems, - int dir); -extern int swiotlb_map_sg(struct device *hwdev, struct scatterlist *sg, - int nents, int direction); -extern void swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sg, - int nents, int direction); -extern int swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr); -extern void swiotlb_free_coherent(struct device *hwdev, size_t size, - void *vaddr, dma_addr_t dma_handle); -extern int swiotlb_dma_supported(struct device *hwdev, u64 mask); -extern void swiotlb_init(void); +#include extern int swiotlb_force; diff --git a/arch/x86/include/asm/swiotlb.h b/arch/x86/include/asm/swiotlb.h index 51fb2c76ad7..b9e4e20174f 100644 --- a/arch/x86/include/asm/swiotlb.h +++ b/arch/x86/include/asm/swiotlb.h @@ -1,46 +1,10 @@ #ifndef _ASM_X86_SWIOTLB_H #define _ASM_X86_SWIOTLB_H -#include +#include /* SWIOTLB interface */ -extern dma_addr_t swiotlb_map_single(struct device *hwdev, void *ptr, - size_t size, int dir); -extern void *swiotlb_alloc_coherent(struct device *hwdev, size_t size, - dma_addr_t *dma_handle, gfp_t flags); -extern void swiotlb_unmap_single(struct device *hwdev, dma_addr_t dev_addr, - size_t size, int dir); -extern void swiotlb_sync_single_for_cpu(struct device *hwdev, - dma_addr_t dev_addr, - size_t size, int dir); -extern void swiotlb_sync_single_for_device(struct device *hwdev, - dma_addr_t dev_addr, - size_t size, int dir); -extern void swiotlb_sync_single_range_for_cpu(struct device *hwdev, - dma_addr_t dev_addr, - unsigned long offset, - size_t size, int dir); -extern void swiotlb_sync_single_range_for_device(struct device *hwdev, - dma_addr_t dev_addr, - unsigned long offset, - size_t size, int dir); -extern void swiotlb_sync_sg_for_cpu(struct device *hwdev, - struct scatterlist *sg, int nelems, - int dir); -extern void swiotlb_sync_sg_for_device(struct device *hwdev, - struct scatterlist *sg, int nelems, - int dir); -extern int swiotlb_map_sg(struct device *hwdev, struct scatterlist *sg, - int nents, int direction); -extern void swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sg, - int nents, int direction); -extern int swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr); -extern void swiotlb_free_coherent(struct device *hwdev, size_t size, - void *vaddr, dma_addr_t dma_handle); -extern int swiotlb_dma_supported(struct device *hwdev, u64 mask); -extern void swiotlb_init(void); - extern int swiotlb_force; #ifdef CONFIG_SWIOTLB From ac86ccc637458dea53077823f98bc3e1b874d478 Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Sun, 28 Dec 2008 15:02:06 +0900 Subject: [PATCH 05/13] swiotlb: remove unnecessary declaration Impact: cleanup Signed-off-by: FUJITA Tomonori Signed-off-by: Ingo Molnar --- lib/swiotlb.c | 3 --- 1 file changed, 3 deletions(-) diff --git a/lib/swiotlb.c b/lib/swiotlb.c index 785046e4c3a..5a7d0a4a9da 100644 --- a/lib/swiotlb.c +++ b/lib/swiotlb.c @@ -22,7 +22,6 @@ #include #include #include -#include #include #include #include @@ -777,8 +776,6 @@ swiotlb_sync_single_range_for_device(struct device *hwdev, dma_addr_t dev_addr, SYNC_FOR_DEVICE); } -void swiotlb_unmap_sg_attrs(struct device *, struct scatterlist *, int, int, - struct dma_attrs *); /* * Map a set of buffers described by scatterlist in streaming mode for DMA. * This is the scatter-gather version of the above swiotlb_map_single From 874d6a9555516cbac828131d573ce3d16ce47164 Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Sun, 28 Dec 2008 15:02:07 +0900 Subject: [PATCH 06/13] swiotlb: clean up EXPORT_SYMBOL usage Impact: cleanup swiotlb uses EXPORT_SYMBOL in an inconsistent way. Some functions use EXPORT_SYMBOL at the end of functions. Some use it at the end of swiotlb.c. This cleans up swiotlb to use EXPORT_SYMBOL in a consistent way (at the end of functions). Signed-off-by: FUJITA Tomonori Signed-off-by: Ingo Molnar --- lib/swiotlb.c | 28 ++++++++++++++-------------- 1 file changed, 14 insertions(+), 14 deletions(-) diff --git a/lib/swiotlb.c b/lib/swiotlb.c index 5a7d0a4a9da..ef3cfe58a15 100644 --- a/lib/swiotlb.c +++ b/lib/swiotlb.c @@ -593,6 +593,7 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size, *dma_handle = dev_addr; return ret; } +EXPORT_SYMBOL(swiotlb_alloc_coherent); void swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr, @@ -605,6 +606,7 @@ swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr, /* DMA_TO_DEVICE to avoid memcpy in unmap_single */ unmap_single(hwdev, vaddr, size, DMA_TO_DEVICE); } +EXPORT_SYMBOL(swiotlb_free_coherent); static void swiotlb_full(struct device *dev, size_t size, int dir, int do_panic) @@ -677,6 +679,7 @@ swiotlb_map_single(struct device *hwdev, void *ptr, size_t size, int dir) { return swiotlb_map_single_attrs(hwdev, ptr, size, dir, NULL); } +EXPORT_SYMBOL(swiotlb_map_single); /* * Unmap a single streaming mode DMA translation. The dma_addr and size must @@ -706,6 +709,8 @@ swiotlb_unmap_single(struct device *hwdev, dma_addr_t dev_addr, size_t size, { return swiotlb_unmap_single_attrs(hwdev, dev_addr, size, dir, NULL); } +EXPORT_SYMBOL(swiotlb_unmap_single); + /* * Make physical memory consistent for a single streaming mode DMA translation * after a transfer. @@ -735,6 +740,7 @@ swiotlb_sync_single_for_cpu(struct device *hwdev, dma_addr_t dev_addr, { swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_CPU); } +EXPORT_SYMBOL(swiotlb_sync_single_for_cpu); void swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr, @@ -742,6 +748,7 @@ swiotlb_sync_single_for_device(struct device *hwdev, dma_addr_t dev_addr, { swiotlb_sync_single(hwdev, dev_addr, size, dir, SYNC_FOR_DEVICE); } +EXPORT_SYMBOL(swiotlb_sync_single_for_device); /* * Same as above, but for a sub-range of the mapping. @@ -767,6 +774,7 @@ swiotlb_sync_single_range_for_cpu(struct device *hwdev, dma_addr_t dev_addr, swiotlb_sync_single_range(hwdev, dev_addr, offset, size, dir, SYNC_FOR_CPU); } +EXPORT_SYMBOL_GPL(swiotlb_sync_single_range_for_cpu); void swiotlb_sync_single_range_for_device(struct device *hwdev, dma_addr_t dev_addr, @@ -775,6 +783,7 @@ swiotlb_sync_single_range_for_device(struct device *hwdev, dma_addr_t dev_addr, swiotlb_sync_single_range(hwdev, dev_addr, offset, size, dir, SYNC_FOR_DEVICE); } +EXPORT_SYMBOL_GPL(swiotlb_sync_single_range_for_device); /* * Map a set of buffers described by scatterlist in streaming mode for DMA. @@ -833,6 +842,7 @@ swiotlb_map_sg(struct device *hwdev, struct scatterlist *sgl, int nelems, { return swiotlb_map_sg_attrs(hwdev, sgl, nelems, dir, NULL); } +EXPORT_SYMBOL(swiotlb_map_sg); /* * Unmap a set of streaming mode DMA translations. Again, cpu read rules @@ -863,6 +873,7 @@ swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sgl, int nelems, { return swiotlb_unmap_sg_attrs(hwdev, sgl, nelems, dir, NULL); } +EXPORT_SYMBOL(swiotlb_unmap_sg); /* * Make physical memory consistent for a set of streaming mode DMA translations @@ -895,6 +906,7 @@ swiotlb_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg, { swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_CPU); } +EXPORT_SYMBOL(swiotlb_sync_sg_for_cpu); void swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg, @@ -902,12 +914,14 @@ swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg, { swiotlb_sync_sg(hwdev, sg, nelems, dir, SYNC_FOR_DEVICE); } +EXPORT_SYMBOL(swiotlb_sync_sg_for_device); int swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr) { return (dma_addr == swiotlb_virt_to_bus(hwdev, io_tlb_overflow_buffer)); } +EXPORT_SYMBOL(swiotlb_dma_mapping_error); /* * Return whether the given device DMA address mask can be supported @@ -920,18 +934,4 @@ swiotlb_dma_supported(struct device *hwdev, u64 mask) { return swiotlb_virt_to_bus(hwdev, io_tlb_end - 1) <= mask; } - -EXPORT_SYMBOL(swiotlb_map_single); -EXPORT_SYMBOL(swiotlb_unmap_single); -EXPORT_SYMBOL(swiotlb_map_sg); -EXPORT_SYMBOL(swiotlb_unmap_sg); -EXPORT_SYMBOL(swiotlb_sync_single_for_cpu); -EXPORT_SYMBOL(swiotlb_sync_single_for_device); -EXPORT_SYMBOL_GPL(swiotlb_sync_single_range_for_cpu); -EXPORT_SYMBOL_GPL(swiotlb_sync_single_range_for_device); -EXPORT_SYMBOL(swiotlb_sync_sg_for_cpu); -EXPORT_SYMBOL(swiotlb_sync_sg_for_device); -EXPORT_SYMBOL(swiotlb_dma_mapping_error); -EXPORT_SYMBOL(swiotlb_alloc_coherent); -EXPORT_SYMBOL(swiotlb_free_coherent); EXPORT_SYMBOL(swiotlb_dma_supported); From ea319518ba3de282c13ae1cf4bf2215c5e03e67e Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 26 Dec 2008 15:08:55 +0100 Subject: [PATCH 07/13] locking, percpu counters: introduce separate lock classes Impact: fix lockdep false positives Classify percpu_counter instances similar to regular lock objects -- that is, per instantiation site. The networking code has increased its use of percpu_counters, which leads to false positives if they are treated as a single class. Signed-off-by: Peter Zijlstra Signed-off-by: Ingo Molnar --- include/linux/percpu_counter.h | 14 ++++++++++---- lib/percpu_counter.c | 18 ++++-------------- lib/proportions.c | 6 +++--- mm/backing-dev.c | 2 +- 4 files changed, 18 insertions(+), 22 deletions(-) diff --git a/include/linux/percpu_counter.h b/include/linux/percpu_counter.h index 9007ccdfc11..96bdde36599 100644 --- a/include/linux/percpu_counter.h +++ b/include/linux/percpu_counter.h @@ -30,8 +30,16 @@ struct percpu_counter { #define FBC_BATCH (NR_CPUS*4) #endif -int percpu_counter_init(struct percpu_counter *fbc, s64 amount); -int percpu_counter_init_irq(struct percpu_counter *fbc, s64 amount); +int __percpu_counter_init(struct percpu_counter *fbc, s64 amount, + struct lock_class_key *key); + +#define percpu_counter_init(fbc, value) \ + ({ \ + static struct lock_class_key __key; \ + \ + __percpu_counter_init(fbc, value, &__key); \ + }) + void percpu_counter_destroy(struct percpu_counter *fbc); void percpu_counter_set(struct percpu_counter *fbc, s64 amount); void __percpu_counter_add(struct percpu_counter *fbc, s64 amount, s32 batch); @@ -85,8 +93,6 @@ static inline int percpu_counter_init(struct percpu_counter *fbc, s64 amount) return 0; } -#define percpu_counter_init_irq percpu_counter_init - static inline void percpu_counter_destroy(struct percpu_counter *fbc) { } diff --git a/lib/percpu_counter.c b/lib/percpu_counter.c index a8663890a88..c7fe2e4e8ed 100644 --- a/lib/percpu_counter.c +++ b/lib/percpu_counter.c @@ -71,11 +71,11 @@ s64 __percpu_counter_sum(struct percpu_counter *fbc) } EXPORT_SYMBOL(__percpu_counter_sum); -static struct lock_class_key percpu_counter_irqsafe; - -int percpu_counter_init(struct percpu_counter *fbc, s64 amount) +int __percpu_counter_init(struct percpu_counter *fbc, s64 amount, + struct lock_class_key *key) { spin_lock_init(&fbc->lock); + lockdep_set_class(&fbc->lock, key); fbc->count = amount; fbc->counters = alloc_percpu(s32); if (!fbc->counters) @@ -87,17 +87,7 @@ int percpu_counter_init(struct percpu_counter *fbc, s64 amount) #endif return 0; } -EXPORT_SYMBOL(percpu_counter_init); - -int percpu_counter_init_irq(struct percpu_counter *fbc, s64 amount) -{ - int err; - - err = percpu_counter_init(fbc, amount); - if (!err) - lockdep_set_class(&fbc->lock, &percpu_counter_irqsafe); - return err; -} +EXPORT_SYMBOL(__percpu_counter_init); void percpu_counter_destroy(struct percpu_counter *fbc) { diff --git a/lib/proportions.c b/lib/proportions.c index 4f387a643d7..7367f2b727d 100644 --- a/lib/proportions.c +++ b/lib/proportions.c @@ -83,11 +83,11 @@ int prop_descriptor_init(struct prop_descriptor *pd, int shift) pd->index = 0; pd->pg[0].shift = shift; mutex_init(&pd->mutex); - err = percpu_counter_init_irq(&pd->pg[0].events, 0); + err = percpu_counter_init(&pd->pg[0].events, 0); if (err) goto out; - err = percpu_counter_init_irq(&pd->pg[1].events, 0); + err = percpu_counter_init(&pd->pg[1].events, 0); if (err) percpu_counter_destroy(&pd->pg[0].events); @@ -191,7 +191,7 @@ int prop_local_init_percpu(struct prop_local_percpu *pl) spin_lock_init(&pl->lock); pl->shift = 0; pl->period = 0; - return percpu_counter_init_irq(&pl->events, 0); + return percpu_counter_init(&pl->events, 0); } void prop_local_destroy_percpu(struct prop_local_percpu *pl) diff --git a/mm/backing-dev.c b/mm/backing-dev.c index f2e574dbc30..f3b12585782 100644 --- a/mm/backing-dev.c +++ b/mm/backing-dev.c @@ -220,7 +220,7 @@ int bdi_init(struct backing_dev_info *bdi) bdi->max_prop_frac = PROP_FRAC_BASE; for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { - err = percpu_counter_init_irq(&bdi->bdi_stat[i], 0); + err = percpu_counter_init(&bdi->bdi_stat[i], 0); if (err) goto err; } From 42d35d48ce7cefb9429880af19d1c329d1554e7a Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Mon, 29 Dec 2008 15:49:53 -0800 Subject: [PATCH 08/13] futex: make futex_(get|put)_key() calls symmetric Impact: cleanup This patch makes the calls to futex_get_key_refs() and futex_drop_key_refs() explicitly symmetric by only "putting" keys we successfully "got". Also cleanup a couple return points that didn't "put" after a successful "get". Build and boot tested on an x86_64 system. Signed-off-by: Darren Hart Cc: Signed-off-by: Ingo Molnar --- kernel/futex.c | 67 +++++++++++++++++++++++++++----------------------- 1 file changed, 36 insertions(+), 31 deletions(-) diff --git a/kernel/futex.c b/kernel/futex.c index b4f87bac91c..c5ac55cc0c1 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -723,8 +723,8 @@ static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset) } spin_unlock(&hb->lock); -out: put_futex_key(fshared, &key); +out: return ret; } @@ -748,7 +748,7 @@ retryfull: goto out; ret = get_futex_key(uaddr2, fshared, &key2); if (unlikely(ret != 0)) - goto out; + goto out_put_key1; hb1 = hash_futex(&key1); hb2 = hash_futex(&key2); @@ -770,12 +770,12 @@ retry: * but we might get them from range checking */ ret = op_ret; - goto out; + goto out_put_keys; #endif if (unlikely(op_ret != -EFAULT)) { ret = op_ret; - goto out; + goto out_put_keys; } /* @@ -789,7 +789,7 @@ retry: ret = futex_handle_fault((unsigned long)uaddr2, attempt); if (ret) - goto out; + goto out_put_keys; goto retry; } @@ -827,10 +827,11 @@ retry: spin_unlock(&hb1->lock); if (hb1 != hb2) spin_unlock(&hb2->lock); -out: +out_put_keys: put_futex_key(fshared, &key2); +out_put_key1: put_futex_key(fshared, &key1); - +out: return ret; } @@ -847,13 +848,13 @@ static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, struct futex_q *this, *next; int ret, drop_count = 0; - retry: +retry: ret = get_futex_key(uaddr1, fshared, &key1); if (unlikely(ret != 0)) goto out; ret = get_futex_key(uaddr2, fshared, &key2); if (unlikely(ret != 0)) - goto out; + goto out_put_key1; hb1 = hash_futex(&key1); hb2 = hash_futex(&key2); @@ -875,7 +876,7 @@ static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, if (!ret) goto retry; - return ret; + goto out_put_keys; } if (curval != *cmpval) { ret = -EAGAIN; @@ -920,9 +921,11 @@ out_unlock: while (--drop_count >= 0) drop_futex_key_refs(&key1); -out: +out_put_keys: put_futex_key(fshared, &key2); +out_put_key1: put_futex_key(fshared, &key1); +out: return ret; } @@ -983,7 +986,7 @@ static int unqueue_me(struct futex_q *q) int ret = 0; /* In the common case we don't take the spinlock, which is nice. */ - retry: +retry: lock_ptr = q->lock_ptr; barrier(); if (lock_ptr != NULL) { @@ -1165,11 +1168,11 @@ static int futex_wait(u32 __user *uaddr, int fshared, q.pi_state = NULL; q.bitset = bitset; - retry: +retry: q.key = FUTEX_KEY_INIT; ret = get_futex_key(uaddr, fshared, &q.key); if (unlikely(ret != 0)) - goto out_release_sem; + goto out; hb = queue_lock(&q); @@ -1197,6 +1200,7 @@ static int futex_wait(u32 __user *uaddr, int fshared, if (unlikely(ret)) { queue_unlock(&q, hb); + put_futex_key(fshared, &q.key); ret = get_user(uval, uaddr); @@ -1206,7 +1210,7 @@ static int futex_wait(u32 __user *uaddr, int fshared, } ret = -EWOULDBLOCK; if (uval != val) - goto out_unlock_release_sem; + goto out_unlock_put_key; /* Only actually queue if *uaddr contained val. */ queue_me(&q, hb); @@ -1298,11 +1302,11 @@ static int futex_wait(u32 __user *uaddr, int fshared, return -ERESTART_RESTARTBLOCK; } - out_unlock_release_sem: +out_unlock_put_key: queue_unlock(&q, hb); - - out_release_sem: put_futex_key(fshared, &q.key); + +out: return ret; } @@ -1351,16 +1355,16 @@ static int futex_lock_pi(u32 __user *uaddr, int fshared, } q.pi_state = NULL; - retry: +retry: q.key = FUTEX_KEY_INIT; ret = get_futex_key(uaddr, fshared, &q.key); if (unlikely(ret != 0)) - goto out_release_sem; + goto out; - retry_unlocked: +retry_unlocked: hb = queue_lock(&q); - retry_locked: +retry_locked: ret = lock_taken = 0; /* @@ -1381,14 +1385,14 @@ static int futex_lock_pi(u32 __user *uaddr, int fshared, */ if (unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(current))) { ret = -EDEADLK; - goto out_unlock_release_sem; + goto out_unlock_put_key; } /* * Surprise - we got the lock. Just return to userspace: */ if (unlikely(!curval)) - goto out_unlock_release_sem; + goto out_unlock_put_key; uval = curval; @@ -1424,7 +1428,7 @@ static int futex_lock_pi(u32 __user *uaddr, int fshared, * We took the lock due to owner died take over. */ if (unlikely(lock_taken)) - goto out_unlock_release_sem; + goto out_unlock_put_key; /* * We dont have the lock. Look up the PI state (or create it if @@ -1463,7 +1467,7 @@ static int futex_lock_pi(u32 __user *uaddr, int fshared, goto retry_locked; } default: - goto out_unlock_release_sem; + goto out_unlock_put_key; } } @@ -1554,16 +1558,17 @@ static int futex_lock_pi(u32 __user *uaddr, int fshared, destroy_hrtimer_on_stack(&to->timer); return ret != -EINTR ? ret : -ERESTARTNOINTR; - out_unlock_release_sem: +out_unlock_put_key: queue_unlock(&q, hb); - out_release_sem: +out_put_key: put_futex_key(fshared, &q.key); +out: if (to) destroy_hrtimer_on_stack(&to->timer); return ret; - uaddr_faulted: +uaddr_faulted: /* * We have to r/w *(int __user *)uaddr, and we have to modify it * atomically. Therefore, if we continue to fault after get_user() @@ -1576,7 +1581,7 @@ static int futex_lock_pi(u32 __user *uaddr, int fshared, if (attempt++) { ret = futex_handle_fault((unsigned long)uaddr, attempt); if (ret) - goto out_release_sem; + goto out_put_key; goto retry_unlocked; } @@ -1668,9 +1673,9 @@ retry_unlocked: out_unlock: spin_unlock(&hb->lock); -out: put_futex_key(fshared, &key); +out: return ret; pi_faulted: From 90621c40cc4ab7b0a414311ce37e7cc7173403b6 Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Mon, 29 Dec 2008 19:43:21 -0800 Subject: [PATCH 09/13] futex: catch certain assymetric (get|put)_futex_key calls Impact: add debug check Following up on my previous key reference accounting patches, this patch will catch puts on keys that haven't been "got". This won't catch nested get/put mismatches though. Build and boot tested, with minimal desktop activity and a run of the open_posix_testsuite in LTP for testing. No warnings logged. Signed-off-by: Darren Hart Cc: Signed-off-by: Ingo Molnar --- kernel/futex.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/kernel/futex.c b/kernel/futex.c index c5ac55cc0c1..206d4c90688 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -170,8 +170,11 @@ static void get_futex_key_refs(union futex_key *key) */ static void drop_futex_key_refs(union futex_key *key) { - if (!key->both.ptr) + if (!key->both.ptr) { + /* If we're here then we tried to put a key we failed to get */ + WARN_ON_ONCE(1); return; + } switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) { case FUT_OFF_INODE: From c12172c0251761c54260376eb29a5f6547495580 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Sun, 4 Jan 2009 20:30:06 -0800 Subject: [PATCH 10/13] rcu: fix rcutree grace-period-latency bug on small systems Impact: fix delays during bootup Kudos to Andi Kleen for finding a grace-period-latency problem! The problem was that the special-case code for small machines never updated the ->signaled field to indicate that grace-period initialization had completed, which prevented force_quiescent_state() from ever expediting grace periods. This problem resulted in grace periods extending for more than 20 seconds. Not subtle. I introduced this bug during my inspection process when I fixed a race between grace-period initialization and force_quiescent_state() execution. The following patch properly updates the ->signaled field for the "small"-system case (no more than 32 CPUs for 32-bit kernels and no more than 64 CPUs for 64-bit kernels). Reported-by: Andi Kleen Tested-by: Andi Kleen Signed-off-by: Paul E. McKenney Signed-off-by: Ingo Molnar --- kernel/rcutree.c | 1 + 1 file changed, 1 insertion(+) diff --git a/kernel/rcutree.c b/kernel/rcutree.c index a342b032112..88d921c5c44 100644 --- a/kernel/rcutree.c +++ b/kernel/rcutree.c @@ -572,6 +572,7 @@ rcu_start_gp(struct rcu_state *rsp, unsigned long flags) /* Special-case the common single-level case. */ if (NUM_RCU_NODES == 1) { rnp->qsmask = rnp->qsmaskinit; + rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ spin_unlock_irqrestore(&rnp->lock, flags); return; } From 90a4d2c0106bb690f0b6af3d506febc35c658aa7 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Sun, 4 Jan 2009 11:41:11 -0800 Subject: [PATCH 11/13] rcu: make treercu safe for suspend and resume Impact: fix kernel warnings [and potential crash] during suspend+resume Kudos to both Dhaval Giani and Jens Axboe for finding a bug in treercu that causes warnings after suspend-resume cycles in Dhaval's case and during stress tests in Jens's case. It would also probably cause failures if heavily stressed. The solution, ironically enough, is to revert to rcupreempt's code for initializing the dynticks state. And the patch even results in smaller code -- so what was I thinking??? This is 2.6.29 material, given that people really do suspend and resume Linux these days. ;-) Reported-by: Dhaval Giani Reported-by: Jens Axboe Tested-by: Dhaval Giani Tested-by: Jens Axboe Signed-off-by: Paul E. McKenney Signed-off-by: Ingo Molnar --- kernel/rcutree.c | 12 ++++-------- 1 file changed, 4 insertions(+), 8 deletions(-) diff --git a/kernel/rcutree.c b/kernel/rcutree.c index 88d921c5c44..f2d8638e6c6 100644 --- a/kernel/rcutree.c +++ b/kernel/rcutree.c @@ -79,7 +79,10 @@ struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); #ifdef CONFIG_NO_HZ -DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks); +DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { + .dynticks_nesting = 1, + .dynticks = 1, +}; #endif /* #ifdef CONFIG_NO_HZ */ static int blimit = 10; /* Maximum callbacks per softirq. */ @@ -1380,13 +1383,6 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp) static void __cpuinit rcu_online_cpu(int cpu) { -#ifdef CONFIG_NO_HZ - struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); - - rdtp->dynticks_nesting = 1; - rdtp->dynticks |= 1; /* need consecutive #s even for hotplug. */ - rdtp->dynticks_nmi = (rdtp->dynticks_nmi + 1) & ~0x1; -#endif /* #ifdef CONFIG_NO_HZ */ rcu_init_percpu_data(cpu, &rcu_state); rcu_init_percpu_data(cpu, &rcu_bh_state); open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); From ea7d3fef4222cd98556a0b386598268d4dbf6670 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Sun, 4 Jan 2009 13:03:02 -0800 Subject: [PATCH 12/13] rcu: eliminate synchronize_rcu_xxx macro Impact: cleanup Expand macro into two files. The synchronize_rcu_xxx macro is quite ugly and it's only used by two callers, so expand it instead. This makes this code easier to change. Signed-off-by: Andi Kleen Signed-off-by: Paul E. McKenney Signed-off-by: Ingo Molnar --- include/linux/rcupdate.h | 12 ------------ kernel/rcupdate.c | 11 +++++++++-- kernel/rcupreempt.c | 11 ++++++++++- 3 files changed, 19 insertions(+), 15 deletions(-) diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h index 1168fbcea8d..921340a7b71 100644 --- a/include/linux/rcupdate.h +++ b/include/linux/rcupdate.h @@ -204,18 +204,6 @@ struct rcu_synchronize { extern void wakeme_after_rcu(struct rcu_head *head); -#define synchronize_rcu_xxx(name, func) \ -void name(void) \ -{ \ - struct rcu_synchronize rcu; \ - \ - init_completion(&rcu.completion); \ - /* Will wake me after RCU finished. */ \ - func(&rcu.head, wakeme_after_rcu); \ - /* Wait for it. */ \ - wait_for_completion(&rcu.completion); \ -} - /** * synchronize_sched - block until all CPUs have exited any non-preemptive * kernel code sequences. diff --git a/kernel/rcupdate.c b/kernel/rcupdate.c index ad63af8b252..d92a76a881a 100644 --- a/kernel/rcupdate.c +++ b/kernel/rcupdate.c @@ -77,8 +77,15 @@ void wakeme_after_rcu(struct rcu_head *head) * sections are delimited by rcu_read_lock() and rcu_read_unlock(), * and may be nested. */ -void synchronize_rcu(void); /* Makes kernel-doc tools happy */ -synchronize_rcu_xxx(synchronize_rcu, call_rcu) +void synchronize_rcu(void) +{ + struct rcu_synchronize rcu; + init_completion(&rcu.completion); + /* Will wake me after RCU finished. */ + call_rcu(&rcu.head, wakeme_after_rcu); + /* Wait for it. */ + wait_for_completion(&rcu.completion); +} EXPORT_SYMBOL_GPL(synchronize_rcu); static void rcu_barrier_callback(struct rcu_head *notused) diff --git a/kernel/rcupreempt.c b/kernel/rcupreempt.c index f9dc8f3720f..33cfc50781f 100644 --- a/kernel/rcupreempt.c +++ b/kernel/rcupreempt.c @@ -1177,7 +1177,16 @@ EXPORT_SYMBOL_GPL(call_rcu_sched); * in -rt this does -not- necessarily result in all currently executing * interrupt -handlers- having completed. */ -synchronize_rcu_xxx(__synchronize_sched, call_rcu_sched) +void __synchronize_sched(void) +{ + struct rcu_synchronize rcu; + + init_completion(&rcu.completion); + /* Will wake me after RCU finished. */ + call_rcu_sched(&rcu.head, wakeme_after_rcu); + /* Wait for it. */ + wait_for_completion(&rcu.completion); +} EXPORT_SYMBOL_GPL(__synchronize_sched); /* From c59ab97e9ecdee9084d2da09e5a8ceea9a396508 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Sun, 4 Jan 2009 18:28:27 -0800 Subject: [PATCH 13/13] rcu: fix rcutorture bug Fix an rcutorture bug that prevents the shutdown notifier from ever actually having any effect, due to the fact that kthreads ignore all signals. Signed-off-by: Paul E. McKenney Signed-off-by: Ingo Molnar --- kernel/rcutorture.c | 18 ++++++++---------- 1 file changed, 8 insertions(+), 10 deletions(-) diff --git a/kernel/rcutorture.c b/kernel/rcutorture.c index 3245b40952c..1cff28db56b 100644 --- a/kernel/rcutorture.c +++ b/kernel/rcutorture.c @@ -136,7 +136,7 @@ static int stutter_pause_test = 0; #endif int rcutorture_runnable = RCUTORTURE_RUNNABLE_INIT; -#define FULLSTOP_SIGNALED 1 /* Bail due to signal. */ +#define FULLSTOP_SHUTDOWN 1 /* Bail due to system shutdown/panic. */ #define FULLSTOP_CLEANUP 2 /* Orderly shutdown. */ static int fullstop; /* stop generating callbacks at test end. */ DEFINE_MUTEX(fullstop_mutex); /* protect fullstop transitions and */ @@ -151,12 +151,10 @@ rcutorture_shutdown_notify(struct notifier_block *unused1, { if (fullstop) return NOTIFY_DONE; - if (signal_pending(current)) { - mutex_lock(&fullstop_mutex); - if (!ACCESS_ONCE(fullstop)) - fullstop = FULLSTOP_SIGNALED; - mutex_unlock(&fullstop_mutex); - } + mutex_lock(&fullstop_mutex); + if (!fullstop) + fullstop = FULLSTOP_SHUTDOWN; + mutex_unlock(&fullstop_mutex); return NOTIFY_DONE; } @@ -624,7 +622,7 @@ rcu_torture_writer(void *arg) rcu_stutter_wait(); } while (!kthread_should_stop() && !fullstop); VERBOSE_PRINTK_STRING("rcu_torture_writer task stopping"); - while (!kthread_should_stop() && fullstop != FULLSTOP_SIGNALED) + while (!kthread_should_stop() && fullstop != FULLSTOP_SHUTDOWN) schedule_timeout_uninterruptible(1); return 0; } @@ -649,7 +647,7 @@ rcu_torture_fakewriter(void *arg) } while (!kthread_should_stop() && !fullstop); VERBOSE_PRINTK_STRING("rcu_torture_fakewriter task stopping"); - while (!kthread_should_stop() && fullstop != FULLSTOP_SIGNALED) + while (!kthread_should_stop() && fullstop != FULLSTOP_SHUTDOWN) schedule_timeout_uninterruptible(1); return 0; } @@ -759,7 +757,7 @@ rcu_torture_reader(void *arg) VERBOSE_PRINTK_STRING("rcu_torture_reader task stopping"); if (irqreader && cur_ops->irqcapable) del_timer_sync(&t); - while (!kthread_should_stop() && fullstop != FULLSTOP_SIGNALED) + while (!kthread_should_stop() && fullstop != FULLSTOP_SHUTDOWN) schedule_timeout_uninterruptible(1); return 0; }