425 lines · c
1// SPDX-License-Identifier: GPL-2.0-only2/*3 * Copyright 20104 * by Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>5 *6 * This code provides a IOMMU for Xen PV guests with PCI passthrough.7 *8 * PV guests under Xen are running in an non-contiguous memory architecture.9 *10 * When PCI pass-through is utilized, this necessitates an IOMMU for11 * translating bus (DMA) to virtual and vice-versa and also providing a12 * mechanism to have contiguous pages for device drivers operations (say DMA13 * operations).14 *15 * Specifically, under Xen the Linux idea of pages is an illusion. It16 * assumes that pages start at zero and go up to the available memory. To17 * help with that, the Linux Xen MMU provides a lookup mechanism to18 * translate the page frame numbers (PFN) to machine frame numbers (MFN)19 * and vice-versa. The MFN are the "real" frame numbers. Furthermore20 * memory is not contiguous. Xen hypervisor stitches memory for guests21 * from different pools, which means there is no guarantee that PFN==MFN22 * and PFN+1==MFN+1. Lastly with Xen 4.0, pages (in debug mode) are23 * allocated in descending order (high to low), meaning the guest might24 * never get any MFN's under the 4GB mark.25 */26 27#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt28 29#include <linux/memblock.h>30#include <linux/dma-direct.h>31#include <linux/dma-map-ops.h>32#include <linux/export.h>33#include <xen/swiotlb-xen.h>34#include <xen/page.h>35#include <xen/xen-ops.h>36#include <xen/hvc-console.h>37 38#include <asm/dma-mapping.h>39 40#include <trace/events/swiotlb.h>41#define MAX_DMA_BITS 3242 43/*44 * Quick lookup value of the bus address of the IOTLB.45 */46 47static inline phys_addr_t xen_phys_to_bus(struct device *dev, phys_addr_t paddr)48{49 unsigned long bfn = pfn_to_bfn(XEN_PFN_DOWN(paddr));50 phys_addr_t baddr = (phys_addr_t)bfn << XEN_PAGE_SHIFT;51 52 baddr |= paddr & ~XEN_PAGE_MASK;53 return baddr;54}55 56static inline dma_addr_t xen_phys_to_dma(struct device *dev, phys_addr_t paddr)57{58 return phys_to_dma(dev, xen_phys_to_bus(dev, paddr));59}60 61static inline phys_addr_t xen_bus_to_phys(struct device *dev,62 phys_addr_t baddr)63{64 unsigned long xen_pfn = bfn_to_pfn(XEN_PFN_DOWN(baddr));65 phys_addr_t paddr = (xen_pfn << XEN_PAGE_SHIFT) |66 (baddr & ~XEN_PAGE_MASK);67 68 return paddr;69}70 71static inline phys_addr_t xen_dma_to_phys(struct device *dev,72 dma_addr_t dma_addr)73{74 return xen_bus_to_phys(dev, dma_to_phys(dev, dma_addr));75}76 77static inline int range_straddles_page_boundary(phys_addr_t p, size_t size)78{79 unsigned long next_bfn, xen_pfn = XEN_PFN_DOWN(p);80 unsigned int i, nr_pages = XEN_PFN_UP(xen_offset_in_page(p) + size);81 phys_addr_t algn = 1ULL << (get_order(size) + PAGE_SHIFT);82 83 next_bfn = pfn_to_bfn(xen_pfn);84 85 /* If buffer is physically aligned, ensure DMA alignment. */86 if (IS_ALIGNED(p, algn) &&87 !IS_ALIGNED((phys_addr_t)next_bfn << XEN_PAGE_SHIFT, algn))88 return 1;89 90 for (i = 1; i < nr_pages; i++)91 if (pfn_to_bfn(++xen_pfn) != ++next_bfn)92 return 1;93 94 return 0;95}96 97static struct io_tlb_pool *xen_swiotlb_find_pool(struct device *dev,98 dma_addr_t dma_addr)99{100 unsigned long bfn = XEN_PFN_DOWN(dma_to_phys(dev, dma_addr));101 unsigned long xen_pfn = bfn_to_local_pfn(bfn);102 phys_addr_t paddr = (phys_addr_t)xen_pfn << XEN_PAGE_SHIFT;103 104 /* If the address is outside our domain, it CAN105 * have the same virtual address as another address106 * in our domain. Therefore _only_ check address within our domain.107 */108 if (pfn_valid(PFN_DOWN(paddr)))109 return swiotlb_find_pool(dev, paddr);110 return NULL;111}112 113#ifdef CONFIG_X86114int xen_swiotlb_fixup(void *buf, unsigned long nslabs)115{116 int rc;117 unsigned int order = get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT);118 unsigned int i, dma_bits = order + PAGE_SHIFT;119 dma_addr_t dma_handle;120 phys_addr_t p = virt_to_phys(buf);121 122 BUILD_BUG_ON(IO_TLB_SEGSIZE & (IO_TLB_SEGSIZE - 1));123 BUG_ON(nslabs % IO_TLB_SEGSIZE);124 125 i = 0;126 do {127 do {128 rc = xen_create_contiguous_region(129 p + (i << IO_TLB_SHIFT), order,130 dma_bits, &dma_handle);131 } while (rc && dma_bits++ < MAX_DMA_BITS);132 if (rc)133 return rc;134 135 i += IO_TLB_SEGSIZE;136 } while (i < nslabs);137 return 0;138}139 140static void *141xen_swiotlb_alloc_coherent(struct device *dev, size_t size,142 dma_addr_t *dma_handle, gfp_t flags, unsigned long attrs)143{144 u64 dma_mask = dev->coherent_dma_mask;145 int order = get_order(size);146 phys_addr_t phys;147 void *ret;148 149 /* Align the allocation to the Xen page size */150 size = ALIGN(size, XEN_PAGE_SIZE);151 152 ret = (void *)__get_free_pages(flags, get_order(size));153 if (!ret)154 return ret;155 phys = virt_to_phys(ret);156 157 *dma_handle = xen_phys_to_dma(dev, phys);158 if (*dma_handle + size - 1 > dma_mask ||159 range_straddles_page_boundary(phys, size)) {160 if (xen_create_contiguous_region(phys, order, fls64(dma_mask),161 dma_handle) != 0)162 goto out_free_pages;163 SetPageXenRemapped(virt_to_page(ret));164 }165 166 memset(ret, 0, size);167 return ret;168 169out_free_pages:170 free_pages((unsigned long)ret, get_order(size));171 return NULL;172}173 174static void175xen_swiotlb_free_coherent(struct device *dev, size_t size, void *vaddr,176 dma_addr_t dma_handle, unsigned long attrs)177{178 phys_addr_t phys = virt_to_phys(vaddr);179 int order = get_order(size);180 181 /* Convert the size to actually allocated. */182 size = ALIGN(size, XEN_PAGE_SIZE);183 184 if (WARN_ON_ONCE(dma_handle + size - 1 > dev->coherent_dma_mask) ||185 WARN_ON_ONCE(range_straddles_page_boundary(phys, size)))186 return;187 188 if (TestClearPageXenRemapped(virt_to_page(vaddr)))189 xen_destroy_contiguous_region(phys, order);190 free_pages((unsigned long)vaddr, get_order(size));191}192#endif /* CONFIG_X86 */193 194/*195 * Map a single buffer of the indicated size for DMA in streaming mode. The196 * physical address to use is returned.197 *198 * Once the device is given the dma address, the device owns this memory until199 * either xen_swiotlb_unmap_page or xen_swiotlb_dma_sync_single is performed.200 */201static dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page,202 unsigned long offset, size_t size,203 enum dma_data_direction dir,204 unsigned long attrs)205{206 phys_addr_t map, phys = page_to_phys(page) + offset;207 dma_addr_t dev_addr = xen_phys_to_dma(dev, phys);208 209 BUG_ON(dir == DMA_NONE);210 /*211 * If the address happens to be in the device's DMA window,212 * we can safely return the device addr and not worry about bounce213 * buffering it.214 */215 if (dma_capable(dev, dev_addr, size, true) &&216 !range_straddles_page_boundary(phys, size) &&217 !xen_arch_need_swiotlb(dev, phys, dev_addr) &&218 !is_swiotlb_force_bounce(dev))219 goto done;220 221 /*222 * Oh well, have to allocate and map a bounce buffer.223 */224 trace_swiotlb_bounced(dev, dev_addr, size);225 226 map = swiotlb_tbl_map_single(dev, phys, size, 0, dir, attrs);227 if (map == (phys_addr_t)DMA_MAPPING_ERROR)228 return DMA_MAPPING_ERROR;229 230 phys = map;231 dev_addr = xen_phys_to_dma(dev, map);232 233 /*234 * Ensure that the address returned is DMA'ble235 */236 if (unlikely(!dma_capable(dev, dev_addr, size, true))) {237 __swiotlb_tbl_unmap_single(dev, map, size, dir,238 attrs | DMA_ATTR_SKIP_CPU_SYNC,239 swiotlb_find_pool(dev, map));240 return DMA_MAPPING_ERROR;241 }242 243done:244 if (!dev_is_dma_coherent(dev) && !(attrs & DMA_ATTR_SKIP_CPU_SYNC)) {245 if (pfn_valid(PFN_DOWN(dma_to_phys(dev, dev_addr))))246 arch_sync_dma_for_device(phys, size, dir);247 else248 xen_dma_sync_for_device(dev, dev_addr, size, dir);249 }250 return dev_addr;251}252 253/*254 * Unmap a single streaming mode DMA translation. The dma_addr and size must255 * match what was provided for in a previous xen_swiotlb_map_page call. All256 * other usages are undefined.257 *258 * After this call, reads by the cpu to the buffer are guaranteed to see259 * whatever the device wrote there.260 */261static void xen_swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr,262 size_t size, enum dma_data_direction dir, unsigned long attrs)263{264 phys_addr_t paddr = xen_dma_to_phys(hwdev, dev_addr);265 struct io_tlb_pool *pool;266 267 BUG_ON(dir == DMA_NONE);268 269 if (!dev_is_dma_coherent(hwdev) && !(attrs & DMA_ATTR_SKIP_CPU_SYNC)) {270 if (pfn_valid(PFN_DOWN(dma_to_phys(hwdev, dev_addr))))271 arch_sync_dma_for_cpu(paddr, size, dir);272 else273 xen_dma_sync_for_cpu(hwdev, dev_addr, size, dir);274 }275 276 /* NOTE: We use dev_addr here, not paddr! */277 pool = xen_swiotlb_find_pool(hwdev, dev_addr);278 if (pool)279 __swiotlb_tbl_unmap_single(hwdev, paddr, size, dir,280 attrs, pool);281}282 283static void284xen_swiotlb_sync_single_for_cpu(struct device *dev, dma_addr_t dma_addr,285 size_t size, enum dma_data_direction dir)286{287 phys_addr_t paddr = xen_dma_to_phys(dev, dma_addr);288 struct io_tlb_pool *pool;289 290 if (!dev_is_dma_coherent(dev)) {291 if (pfn_valid(PFN_DOWN(dma_to_phys(dev, dma_addr))))292 arch_sync_dma_for_cpu(paddr, size, dir);293 else294 xen_dma_sync_for_cpu(dev, dma_addr, size, dir);295 }296 297 pool = xen_swiotlb_find_pool(dev, dma_addr);298 if (pool)299 __swiotlb_sync_single_for_cpu(dev, paddr, size, dir, pool);300}301 302static void303xen_swiotlb_sync_single_for_device(struct device *dev, dma_addr_t dma_addr,304 size_t size, enum dma_data_direction dir)305{306 phys_addr_t paddr = xen_dma_to_phys(dev, dma_addr);307 struct io_tlb_pool *pool;308 309 pool = xen_swiotlb_find_pool(dev, dma_addr);310 if (pool)311 __swiotlb_sync_single_for_device(dev, paddr, size, dir, pool);312 313 if (!dev_is_dma_coherent(dev)) {314 if (pfn_valid(PFN_DOWN(dma_to_phys(dev, dma_addr))))315 arch_sync_dma_for_device(paddr, size, dir);316 else317 xen_dma_sync_for_device(dev, dma_addr, size, dir);318 }319}320 321/*322 * Unmap a set of streaming mode DMA translations. Again, cpu read rules323 * concerning calls here are the same as for swiotlb_unmap_page() above.324 */325static void326xen_swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sgl, int nelems,327 enum dma_data_direction dir, unsigned long attrs)328{329 struct scatterlist *sg;330 int i;331 332 BUG_ON(dir == DMA_NONE);333 334 for_each_sg(sgl, sg, nelems, i)335 xen_swiotlb_unmap_page(hwdev, sg->dma_address, sg_dma_len(sg),336 dir, attrs);337 338}339 340static int341xen_swiotlb_map_sg(struct device *dev, struct scatterlist *sgl, int nelems,342 enum dma_data_direction dir, unsigned long attrs)343{344 struct scatterlist *sg;345 int i;346 347 BUG_ON(dir == DMA_NONE);348 349 for_each_sg(sgl, sg, nelems, i) {350 sg->dma_address = xen_swiotlb_map_page(dev, sg_page(sg),351 sg->offset, sg->length, dir, attrs);352 if (sg->dma_address == DMA_MAPPING_ERROR)353 goto out_unmap;354 sg_dma_len(sg) = sg->length;355 }356 357 return nelems;358out_unmap:359 xen_swiotlb_unmap_sg(dev, sgl, i, dir, attrs | DMA_ATTR_SKIP_CPU_SYNC);360 sg_dma_len(sgl) = 0;361 return -EIO;362}363 364static void365xen_swiotlb_sync_sg_for_cpu(struct device *dev, struct scatterlist *sgl,366 int nelems, enum dma_data_direction dir)367{368 struct scatterlist *sg;369 int i;370 371 for_each_sg(sgl, sg, nelems, i) {372 xen_swiotlb_sync_single_for_cpu(dev, sg->dma_address,373 sg->length, dir);374 }375}376 377static void378xen_swiotlb_sync_sg_for_device(struct device *dev, struct scatterlist *sgl,379 int nelems, enum dma_data_direction dir)380{381 struct scatterlist *sg;382 int i;383 384 for_each_sg(sgl, sg, nelems, i) {385 xen_swiotlb_sync_single_for_device(dev, sg->dma_address,386 sg->length, dir);387 }388}389 390/*391 * Return whether the given device DMA address mask can be supported392 * properly. For example, if your device can only drive the low 24-bits393 * during bus mastering, then you would pass 0x00ffffff as the mask to394 * this function.395 */396static int397xen_swiotlb_dma_supported(struct device *hwdev, u64 mask)398{399 return xen_phys_to_dma(hwdev, default_swiotlb_limit()) <= mask;400}401 402const struct dma_map_ops xen_swiotlb_dma_ops = {403#ifdef CONFIG_X86404 .alloc = xen_swiotlb_alloc_coherent,405 .free = xen_swiotlb_free_coherent,406#else407 .alloc = dma_direct_alloc,408 .free = dma_direct_free,409#endif410 .sync_single_for_cpu = xen_swiotlb_sync_single_for_cpu,411 .sync_single_for_device = xen_swiotlb_sync_single_for_device,412 .sync_sg_for_cpu = xen_swiotlb_sync_sg_for_cpu,413 .sync_sg_for_device = xen_swiotlb_sync_sg_for_device,414 .map_sg = xen_swiotlb_map_sg,415 .unmap_sg = xen_swiotlb_unmap_sg,416 .map_page = xen_swiotlb_map_page,417 .unmap_page = xen_swiotlb_unmap_page,418 .dma_supported = xen_swiotlb_dma_supported,419 .mmap = dma_common_mmap,420 .get_sgtable = dma_common_get_sgtable,421 .alloc_pages_op = dma_common_alloc_pages,422 .free_pages = dma_common_free_pages,423 .max_mapping_size = swiotlb_max_mapping_size,424};425