History

Merge tag 'dma-mapping-6.8-2024-01-08' of git://git.infradead.org/users/hch/dma-mapping

Pull dma-mapping updates from Christoph Hellwig: - reduce area lock contention for non-primary IO TLB pools (Petr Tesarik) - don't store redundant offsets in the dma_ranges stuctures (Robin Murphy) - clear dev->dma_mem when freeing per-device pools (Joakim Zhang) * tag 'dma-mapping-6.8-2024-01-08' of git://git.infradead.org/users/hch/dma-mapping: dma-mapping: clear dev->dma_mem to NULL after freeing it swiotlb: reduce area lock contention for non-primary IO TLB pools dma-mapping: don't store redundant offsets

Linus Torvalds committed 2 years ago

Commit 893e2f92 parents 457e4f9 b07bc236 files · +70 −46Browse files

Unified · Split

6 files
  1. drivers/acpi/scan.c +0 −1
  2. drivers/of/address.c +0 −1
  3. include/linux/dma-direct.h +12 −7
  4. kernel/dma/coherent.c +3 −1
  5. kernel/dma/direct.c +0 −1
  6. kernel/dma/swiotlb.c +55 −35
drivers/acpi/scan.c+0−1
@@ -1532,7 +1532,6 @@ int acpi_dma_get_range(struct device *dev, const struct bus_dma_region **map) · 1531 unmodified lines
15321532 r->cpu_start = rentry->res->start;
15331533 r->dma_start = rentry->res->start - rentry->offset;
15341534 r->size = resource_size(rentry->res);
1535− r->offset = rentry->offset;
15361535 r++;
15371536 }
15381537 }
drivers/of/address.c+0−1
@@ -955,7 +955,6 @@ int of_dma_get_range(struct device_node *np, const struct bus_dma_region **map) · 954 unmodified lines
955955 r->cpu_start = range.cpu_addr;
956956 r->dma_start = range.bus_addr;
957957 r->size = range.size;
958− r->offset = range.cpu_addr - range.bus_addr;
959958 r++;
960959 }
961960 out:
include/linux/dma-direct.h+12−7
@@ -21,7 +21,6 @@ struct bus_dma_region { · 20 unmodified lines
2121 phys_addr_t cpu_start;
2222 dma_addr_t dma_start;
2323 u64 size;
24− u64 offset;
2524 };
2625
2726 static inline dma_addr_t translate_phys_to_dma(struct device *dev,
@@ -29,9 +28,12 @@ static inline dma_addr_t translate_phys_to_dma(struct device *dev, · 1 unmodified line
2928 {
3029 const struct bus_dma_region *m;
3130
32− for (m = dev->dma_range_map; m->size; m++)
33− if (paddr >= m->cpu_start && paddr - m->cpu_start < m->size)
34− return (dma_addr_t)paddr - m->offset;
31+ for (m = dev->dma_range_map; m->size; m++) {
32+ u64 offset = paddr - m->cpu_start;
33+
34+ if (paddr >= m->cpu_start && offset < m->size)
35+ return m->dma_start + offset;
36+ }
3537
3638 /* make sure dma_capable fails when no translation is available */
3739 return DMA_MAPPING_ERROR;
@@ -42,9 +44,12 @@ static inline phys_addr_t translate_dma_to_phys(struct device *dev, · 4 unmodified lines
4244 {
4345 const struct bus_dma_region *m;
4446
45− for (m = dev->dma_range_map; m->size; m++)
46− if (dma_addr >= m->dma_start && dma_addr - m->dma_start < m->size)
47− return (phys_addr_t)dma_addr + m->offset;
47+ for (m = dev->dma_range_map; m->size; m++) {
48+ u64 offset = dma_addr - m->dma_start;
49+
50+ if (dma_addr >= m->dma_start && offset < m->size)
51+ return m->cpu_start + offset;
52+ }
4853
4954 return (phys_addr_t)-1;
5055 }
kernel/dma/coherent.c+3−1
@@ -132,8 +132,10 @@ int dma_declare_coherent_memory(struct device *dev, phys_addr_t phys_addr, · 131 unmodified lines
132132
133133 void dma_release_coherent_memory(struct device *dev)
134134 {
135− if (dev)
135+ if (dev) {
136136 _dma_release_coherent_memory(dev->dma_mem);
137+ dev->dma_mem = NULL;
138+ }
137139 }
138140
139141 static void *__dma_alloc_from_coherent(struct device *dev,
kernel/dma/direct.c+0−1
@@ -677,7 +677,6 @@ int dma_direct_set_offset(struct device *dev, phys_addr_t cpu_start, · 676 unmodified lines
677677 return -ENOMEM;
678678 map[0].cpu_start = cpu_start;
679679 map[0].dma_start = dma_start;
680− map[0].offset = offset;
681680 map[0].size = size;
682681 dev->dma_range_map = map;
683682 return 0;
kernel/dma/swiotlb.c+55−35
@@ -957,7 +957,7 @@ static void dec_used(struct io_tlb_mem *mem, unsigned int nslots) · 956 unmodified lines
957957 #endif /* CONFIG_DEBUG_FS */
958958
959959 /**
960− * swiotlb_area_find_slots() - search for slots in one IO TLB memory area
960+ * swiotlb_search_pool_area() - search one memory area in one pool
961961 * @dev: Device which maps the buffer.
962962 * @pool: Memory pool to be searched.
963963 * @area_index: Index of the IO TLB memory area to be searched.
@@ -972,7 +972,7 @@ static void dec_used(struct io_tlb_mem *mem, unsigned int nslots) · 8 unmodified lines
972972 *
973973 * Return: Index of the first allocated slot, or -1 on error.
974974 */
975−static int swiotlb_area_find_slots(struct device *dev, struct io_tlb_pool *pool,
975+static int swiotlb_search_pool_area(struct device *dev, struct io_tlb_pool *pool,
976976 int area_index, phys_addr_t orig_addr, size_t alloc_size,
977977 unsigned int alloc_align_mask)
978978 {
@@ -1066,41 +1066,50 @@ found: · 87 unmodified lines
10661066 return slot_index;
10671067 }
10681068
1069+#ifdef CONFIG_SWIOTLB_DYNAMIC
1070+
10691071 /**
1070− * swiotlb_pool_find_slots() - search for slots in one memory pool
1072+ * swiotlb_search_area() - search one memory area in all pools
10711073 * @dev: Device which maps the buffer.
1072− * @pool: Memory pool to be searched.
1074+ * @start_cpu: Start CPU number.
1075+ * @cpu_offset: Offset from @start_cpu.
10731076 * @orig_addr: Original (non-bounced) IO buffer address.
10741077 * @alloc_size: Total requested size of the bounce buffer,
10751078 * including initial alignment padding.
10761079 * @alloc_align_mask: Required alignment of the allocated buffer.
1080+ * @retpool: Used memory pool, updated on return.
10771081 *
1078− * Search through one memory pool to find a sequence of slots that match the
1082+ * Search one memory area in all pools for a sequence of slots that match the
10791083 * allocation constraints.
10801084 *
10811085 * Return: Index of the first allocated slot, or -1 on error.
10821086 */
1083−static int swiotlb_pool_find_slots(struct device *dev, struct io_tlb_pool *pool,
1084− phys_addr_t orig_addr, size_t alloc_size,
1085− unsigned int alloc_align_mask)
1087+static int swiotlb_search_area(struct device *dev, int start_cpu,
1088+ int cpu_offset, phys_addr_t orig_addr, size_t alloc_size,
1089+ unsigned int alloc_align_mask, struct io_tlb_pool **retpool)
10861090 {
1087− int start = raw_smp_processor_id() & (pool->nareas - 1);
1088− int i = start, index;
1089−
1090− do {
1091− index = swiotlb_area_find_slots(dev, pool, i, orig_addr,
1092− alloc_size, alloc_align_mask);
1093− if (index >= 0)
1094− return index;
1095− if (++i >= pool->nareas)
1096− i = 0;
1097− } while (i != start);
1091+ struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
1092+ struct io_tlb_pool *pool;
1093+ int area_index;
1094+ int index = -1;
10981095
1099− return -1;
1096+ rcu_read_lock();
1097+ list_for_each_entry_rcu(pool, &mem->pools, node) {
1098+ if (cpu_offset >= pool->nareas)
1099+ continue;
1100+ area_index = (start_cpu + cpu_offset) & (pool->nareas - 1);
1101+ index = swiotlb_search_pool_area(dev, pool, area_index,
1102+ orig_addr, alloc_size,
1103+ alloc_align_mask);
1104+ if (index >= 0) {
1105+ *retpool = pool;
1106+ break;
1107+ }
1108+ }
1109+ rcu_read_unlock();
1110+ return index;
11001111 }
11011112
1102−#ifdef CONFIG_SWIOTLB_DYNAMIC
1103−
11041113 /**
11051114 * swiotlb_find_slots() - search for slots in the whole swiotlb
11061115 * @dev: Device which maps the buffer.
@@ -1124,18 +1133,17 @@ static int swiotlb_find_slots(struct device *dev, phys_addr_t orig_addr, · 17 unmodified lines
11241133 unsigned long nslabs;
11251134 unsigned long flags;
11261135 u64 phys_limit;
1136+ int cpu, i;
11271137 int index;
11281138
1129− rcu_read_lock();
1130− list_for_each_entry_rcu(pool, &mem->pools, node) {
1131− index = swiotlb_pool_find_slots(dev, pool, orig_addr,
1132− alloc_size, alloc_align_mask);
1133− if (index >= 0) {
1134− rcu_read_unlock();
1139+ cpu = raw_smp_processor_id();
1140+ for (i = 0; i < default_nareas; ++i) {
1141+ index = swiotlb_search_area(dev, cpu, i, orig_addr, alloc_size,
1142+ alloc_align_mask, &pool);
1143+ if (index >= 0)
11351144 goto found;
1136− }
11371145 }
1138− rcu_read_unlock();
1146+
11391147 if (!mem->can_grow)
11401148 return -1;
11411149
@@ -1148,8 +1156,8 @@ static int swiotlb_find_slots(struct device *dev, phys_addr_t orig_addr, · 6 unmodified lines
11481156 if (!pool)
11491157 return -1;
11501158
1151− index = swiotlb_pool_find_slots(dev, pool, orig_addr,
1152− alloc_size, alloc_align_mask);
1159+ index = swiotlb_search_pool_area(dev, pool, 0, orig_addr,
1160+ alloc_size, alloc_align_mask);
11531161 if (index < 0) {
11541162 swiotlb_dyn_free(&pool->rcu);
11551163 return -1;
@@ -1192,9 +1200,21 @@ static int swiotlb_find_slots(struct device *dev, phys_addr_t orig_addr, · 36 unmodified lines
11921200 size_t alloc_size, unsigned int alloc_align_mask,
11931201 struct io_tlb_pool **retpool)
11941202 {
1195− *retpool = &dev->dma_io_tlb_mem->defpool;
1196− return swiotlb_pool_find_slots(dev, *retpool,
1197− orig_addr, alloc_size, alloc_align_mask);
1203+ struct io_tlb_pool *pool;
1204+ int start, i;
1205+ int index;
1206+
1207+ *retpool = pool = &dev->dma_io_tlb_mem->defpool;
1208+ i = start = raw_smp_processor_id() & (pool->nareas - 1);
1209+ do {
1210+ index = swiotlb_search_pool_area(dev, pool, i, orig_addr,
1211+ alloc_size, alloc_align_mask);
1212+ if (index >= 0)
1213+ return index;
1214+ if (++i >= pool->nareas)
1215+ i = 0;
1216+ } while (i != start);
1217+ return -1;
11981218 }
11991219
12001220 #endif /* CONFIG_SWIOTLB_DYNAMIC */