diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_migrate.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 446 |
1 files changed, 237 insertions, 209 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index 22b077ac9a19..af53e796ea1b 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -21,14 +21,12 @@ * OTHER DEALINGS IN THE SOFTWARE. */ #include <linux/types.h> -#include <linux/hmm.h> #include <linux/dma-direction.h> #include <linux/dma-mapping.h> #include <linux/migrate.h> #include "amdgpu_sync.h" #include "amdgpu_object.h" #include "amdgpu_vm.h" -#include "amdgpu_mn.h" #include "amdgpu_res_cursor.h" #include "kfd_priv.h" #include "kfd_svm.h" @@ -40,22 +38,22 @@ #endif #define dev_fmt(fmt) "kfd_migrate: " fmt -static uint64_t -svm_migrate_direct_mapping_addr(struct amdgpu_device *adev, uint64_t addr) +static u64 +svm_migrate_direct_mapping_addr(struct amdgpu_device *adev, u64 addr) { return addr + amdgpu_ttm_domain_start(adev, TTM_PL_VRAM); } static int -svm_migrate_gart_map(struct amdgpu_ring *ring, uint64_t npages, - dma_addr_t *addr, uint64_t *gart_addr, uint64_t flags) +svm_migrate_gart_map(struct amdgpu_ring *ring, u64 npages, + dma_addr_t *addr, u64 *gart_addr, u64 flags) { struct amdgpu_device *adev = ring->adev; struct amdgpu_job *job; unsigned int num_dw, num_bytes; struct dma_fence *fence; - uint64_t src_addr, dst_addr; - uint64_t pte_flags; + u64 src_addr, dst_addr; + u64 pte_flags; void *cpu_addr; int r; @@ -65,8 +63,12 @@ svm_migrate_gart_map(struct amdgpu_ring *ring, uint64_t npages, num_dw = ALIGN(adev->mman.buffer_funcs->copy_num_dw, 8); num_bytes = npages * 8; - r = amdgpu_job_alloc_with_ib(adev, num_dw * 4 + num_bytes, - AMDGPU_IB_POOL_DELAYED, &job); + r = amdgpu_job_alloc_with_ib(adev, &adev->mman.high_pr, + AMDGPU_FENCE_OWNER_UNDEFINED, + num_dw * 4 + num_bytes, + AMDGPU_IB_POOL_DELAYED, + &job, + AMDGPU_KERNEL_JOB_ID_KFD_GART_MAP); if (r) return r; @@ -75,7 +77,7 @@ svm_migrate_gart_map(struct amdgpu_ring *ring, uint64_t npages, dst_addr = amdgpu_bo_gpu_offset(adev->gart.bo); amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_addr, - dst_addr, num_bytes, false); + dst_addr, num_bytes, 0); amdgpu_ring_pad_ib(ring, &job->ibs[0]); WARN_ON(job->ibs[0].length_dw > num_dw); @@ -89,18 +91,10 @@ svm_migrate_gart_map(struct amdgpu_ring *ring, uint64_t npages, cpu_addr = &job->ibs[0].ptr[num_dw]; amdgpu_gart_map(adev, 0, npages, addr, pte_flags, cpu_addr); - r = amdgpu_job_submit(job, &adev->mman.entity, - AMDGPU_FENCE_OWNER_UNDEFINED, &fence); - if (r) - goto error_free; - + fence = amdgpu_job_submit(job); dma_fence_put(fence); return r; - -error_free: - amdgpu_job_free(job); - return r; } /** @@ -128,15 +122,15 @@ error_free: static int svm_migrate_copy_memory_gart(struct amdgpu_device *adev, dma_addr_t *sys, - uint64_t *vram, uint64_t npages, + u64 *vram, u64 npages, enum MIGRATION_COPY_DIR direction, struct dma_fence **mfence) { - const uint64_t GTT_MAX_PAGES = AMDGPU_GTT_MAX_TRANSFER_SIZE; + const u64 GTT_MAX_PAGES = AMDGPU_GTT_MAX_TRANSFER_SIZE; struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; - uint64_t gart_s, gart_d; + u64 gart_s, gart_d; struct dma_fence *next; - uint64_t size; + u64 size; int r; mutex_lock(&adev->mman.gtt_window_lock); @@ -159,7 +153,7 @@ svm_migrate_copy_memory_gart(struct amdgpu_device *adev, dma_addr_t *sys, } r = amdgpu_copy_buffer(ring, gart_s, gart_d, size * PAGE_SIZE, - NULL, &next, false, true, false); + NULL, &next, false, true, 0); if (r) { dev_err(adev->dev, "fail %d to copy memory\n", r); goto out_unlock; @@ -212,7 +206,7 @@ svm_migrate_copy_done(struct amdgpu_device *adev, struct dma_fence *mfence) unsigned long svm_migrate_addr_to_pfn(struct amdgpu_device *adev, unsigned long addr) { - return (addr + adev->kfd.dev->pgmap.range.start) >> PAGE_SHIFT; + return (addr + adev->kfd.pgmap.range.start) >> PAGE_SHIFT; } static void @@ -223,7 +217,7 @@ svm_migrate_get_vram_page(struct svm_range *prange, unsigned long pfn) page = pfn_to_page(pfn); svm_range_bo_ref(prange->svm_bo); page->zone_device_data = prange->svm_bo; - zone_device_page_init(page); + zone_device_page_init(page, 0); } static void @@ -242,7 +236,7 @@ svm_migrate_addr(struct amdgpu_device *adev, struct page *page) unsigned long addr; addr = page_to_pfn(page) << PAGE_SHIFT; - return (addr - adev->kfd.dev->pgmap.range.start); + return (addr - adev->kfd.pgmap.range.start); } static struct page * @@ -268,72 +262,57 @@ static void svm_migrate_put_sys_page(unsigned long addr) static unsigned long svm_migrate_successful_pages(struct migrate_vma *migrate) { - unsigned long cpages = 0; + unsigned long mpages = 0; unsigned long i; for (i = 0; i < migrate->npages; i++) { - if (migrate->src[i] & MIGRATE_PFN_VALID && + if (migrate->dst[i] & MIGRATE_PFN_VALID && migrate->src[i] & MIGRATE_PFN_MIGRATE) - cpages++; - } - return cpages; -} - -static unsigned long svm_migrate_unsuccessful_pages(struct migrate_vma *migrate) -{ - unsigned long upages = 0; - unsigned long i; - - for (i = 0; i < migrate->npages; i++) { - if (migrate->src[i] & MIGRATE_PFN_VALID && - !(migrate->src[i] & MIGRATE_PFN_MIGRATE)) - upages++; + mpages++; } - return upages; + return mpages; } static int -svm_migrate_copy_to_vram(struct amdgpu_device *adev, struct svm_range *prange, +svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, struct migrate_vma *migrate, struct dma_fence **mfence, - dma_addr_t *scratch) + dma_addr_t *scratch, u64 ttm_res_offset) { - uint64_t npages = migrate->npages; + u64 npages = migrate->npages; + struct amdgpu_device *adev = node->adev; struct device *dev = adev->dev; struct amdgpu_res_cursor cursor; + u64 mpages = 0; dma_addr_t *src; - uint64_t *dst; - uint64_t i, j; + u64 *dst; + u64 i, j; int r; - pr_debug("svms 0x%p [0x%lx 0x%lx]\n", prange->svms, prange->start, - prange->last); + pr_debug("svms 0x%p [0x%lx 0x%lx 0x%llx]\n", prange->svms, prange->start, + prange->last, ttm_res_offset); src = scratch; - dst = (uint64_t *)(scratch + npages); + dst = (u64 *)(scratch + npages); - r = svm_range_vram_node_new(adev, prange, true); - if (r) { - dev_dbg(adev->dev, "fail %d to alloc vram\n", r); - goto out; - } - - amdgpu_res_first(prange->ttm_res, prange->offset << PAGE_SHIFT, + amdgpu_res_first(prange->ttm_res, ttm_res_offset, npages << PAGE_SHIFT, &cursor); - for (i = j = 0; i < npages; i++) { + for (i = j = 0; (i < npages) && (mpages < migrate->cpages); i++) { struct page *spage; - dst[i] = cursor.start + (j << PAGE_SHIFT); - migrate->dst[i] = svm_migrate_addr_to_pfn(adev, dst[i]); - svm_migrate_get_vram_page(prange, migrate->dst[i]); - migrate->dst[i] = migrate_pfn(migrate->dst[i]); - + if (migrate->src[i] & MIGRATE_PFN_MIGRATE) { + dst[i] = cursor.start + (j << PAGE_SHIFT); + migrate->dst[i] = svm_migrate_addr_to_pfn(adev, dst[i]); + svm_migrate_get_vram_page(prange, migrate->dst[i]); + migrate->dst[i] = migrate_pfn(migrate->dst[i]); + mpages++; + } spage = migrate_pfn_to_page(migrate->src[i]); if (spage && !is_zone_device_page(spage)) { src[i] = dma_map_page(dev, spage, 0, PAGE_SIZE, - DMA_TO_DEVICE); + DMA_BIDIRECTIONAL); r = dma_mapping_error(dev, src[i]); if (r) { - dev_err(adev->dev, "%s: fail %d dma_map_page\n", + dev_err(dev, "%s: fail %d dma_map_page\n", __func__, r); goto out_free_vram_pages; } @@ -377,9 +356,12 @@ svm_migrate_copy_to_vram(struct amdgpu_device *adev, struct svm_range *prange, out_free_vram_pages: if (r) { pr_debug("failed %d to copy memory to vram\n", r); - while (i--) { + for (i = 0; i < npages && mpages; i++) { + if (!dst[i]) + continue; svm_migrate_put_vram_page(adev, dst[i]); migrate->dst[i] = 0; + mpages--; } } @@ -397,21 +379,23 @@ out_free_vram_pages: migrate->dst[i + 3] = 0; } #endif -out: + return r; } static long -svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange, - struct vm_area_struct *vma, uint64_t start, - uint64_t end, uint32_t trigger) +svm_migrate_vma_to_vram(struct kfd_node *node, struct svm_range *prange, + struct vm_area_struct *vma, u64 start, + u64 end, uint32_t trigger, u64 ttm_res_offset) { struct kfd_process *p = container_of(prange->svms, struct kfd_process, svms); - uint64_t npages = (end - start) >> PAGE_SHIFT; + u64 npages = (end - start) >> PAGE_SHIFT; + struct amdgpu_device *adev = node->adev; struct kfd_process_device *pdd; struct dma_fence *mfence = NULL; struct migrate_vma migrate = { 0 }; unsigned long cpages = 0; + unsigned long mpages = 0; dma_addr_t *scratch; void *buf; int r = -ENOMEM; @@ -424,7 +408,7 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange, migrate.pgmap_owner = SVM_ADEV_PGMAP_OWNER(adev); buf = kvcalloc(npages, - 2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t), + 2 * sizeof(*migrate.src) + sizeof(u64) + sizeof(dma_addr_t), GFP_KERNEL); if (!buf) goto out; @@ -433,9 +417,9 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange, migrate.dst = migrate.src + npages; scratch = (dma_addr_t *)(migrate.dst + npages); - kfd_smi_event_migration_start(adev->kfd.dev, p->lead_thread->pid, + kfd_smi_event_migration_start(node, p->lead_thread->pid, start >> PAGE_SHIFT, end >> PAGE_SHIFT, - 0, adev->kfd.dev->id, prange->prefetch_loc, + 0, node->id, prange->prefetch_loc, prange->preferred_loc, trigger); r = migrate_vma_setup(&migrate); @@ -452,36 +436,35 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange, goto out_free; } if (cpages != npages) - pr_debug("partial migration, 0x%lx/0x%llx pages migrated\n", + pr_debug("partial migration, 0x%lx/0x%llx pages collected\n", cpages, npages); else - pr_debug("0x%lx pages migrated\n", cpages); + pr_debug("0x%lx pages collected\n", cpages); - r = svm_migrate_copy_to_vram(adev, prange, &migrate, &mfence, scratch); + r = svm_migrate_copy_to_vram(node, prange, &migrate, &mfence, scratch, ttm_res_offset); migrate_vma_pages(&migrate); - pr_debug("successful/cpages/npages 0x%lx/0x%lx/0x%lx\n", - svm_migrate_successful_pages(&migrate), cpages, migrate.npages); - svm_migrate_copy_done(adev, mfence); migrate_vma_finalize(&migrate); - kfd_smi_event_migration_end(adev->kfd.dev, p->lead_thread->pid, - start >> PAGE_SHIFT, end >> PAGE_SHIFT, - 0, adev->kfd.dev->id, trigger); + mpages = svm_migrate_successful_pages(&migrate); + pr_debug("migrated/collected/requested 0x%lx/0x%lx/0x%lx\n", + mpages, cpages, migrate.npages); - svm_range_dma_unmap(adev->dev, scratch, 0, npages); - svm_range_free_dma_mappings(prange); + svm_range_dma_unmap_dev(adev->dev, scratch, 0, npages); out_free: kvfree(buf); + kfd_smi_event_migration_end(node, p->lead_thread->pid, + start >> PAGE_SHIFT, end >> PAGE_SHIFT, + 0, node->id, trigger, r); out: - if (!r && cpages) { - pdd = svm_range_get_pdd_by_adev(prange, adev); + if (!r && mpages) { + pdd = svm_range_get_pdd_by_node(prange, node); if (pdd) - WRITE_ONCE(pdd->page_in, pdd->page_in + cpages); + WRITE_ONCE(pdd->page_in, pdd->page_in + mpages); - return cpages; + return mpages; } return r; } @@ -490,6 +473,8 @@ out: * svm_migrate_ram_to_vram - migrate svm range from system to device * @prange: range structure * @best_loc: the device to migrate to + * @start_mgr: start page to migrate + * @last_mgr: last page to migrate * @mm: the process mm structure * @trigger: reason of migration * @@ -500,58 +485,91 @@ out: */ static int svm_migrate_ram_to_vram(struct svm_range *prange, uint32_t best_loc, + unsigned long start_mgr, unsigned long last_mgr, struct mm_struct *mm, uint32_t trigger) { unsigned long addr, start, end; struct vm_area_struct *vma; - struct amdgpu_device *adev; - unsigned long cpages = 0; + u64 ttm_res_offset; + struct kfd_node *node; + unsigned long mpages = 0; long r = 0; - if (prange->actual_loc == best_loc) { - pr_debug("svms 0x%p [0x%lx 0x%lx] already on best_loc 0x%x\n", - prange->svms, prange->start, prange->last, best_loc); - return 0; + if (start_mgr < prange->start || last_mgr > prange->last) { + pr_debug("range [0x%lx 0x%lx] out prange [0x%lx 0x%lx]\n", + start_mgr, last_mgr, prange->start, prange->last); + return -EFAULT; } - adev = svm_range_get_adev_by_id(prange, best_loc); - if (!adev) { - pr_debug("failed to get device by id 0x%x\n", best_loc); + node = svm_range_get_node_by_id(prange, best_loc); + if (!node) { + pr_debug("failed to get kfd node by id 0x%x\n", best_loc); return -ENODEV; } - pr_debug("svms 0x%p [0x%lx 0x%lx] to gpu 0x%x\n", prange->svms, - prange->start, prange->last, best_loc); + pr_debug("svms 0x%p [0x%lx 0x%lx] in [0x%lx 0x%lx] to gpu 0x%x\n", + prange->svms, start_mgr, last_mgr, prange->start, prange->last, + best_loc); - start = prange->start << PAGE_SHIFT; - end = (prange->last + 1) << PAGE_SHIFT; + start = start_mgr << PAGE_SHIFT; + end = (last_mgr + 1) << PAGE_SHIFT; + + r = amdgpu_amdkfd_reserve_mem_limit(node->adev, + prange->npages * PAGE_SIZE, + KFD_IOC_ALLOC_MEM_FLAGS_VRAM, + node->xcp ? node->xcp->id : 0); + if (r) { + dev_dbg(node->adev->dev, "failed to reserve VRAM, r: %ld\n", r); + return -ENOSPC; + } + + r = svm_range_vram_node_new(node, prange, true); + if (r) { + dev_dbg(node->adev->dev, "fail %ld to alloc vram\n", r); + goto out; + } + ttm_res_offset = (start_mgr - prange->start + prange->offset) << PAGE_SHIFT; for (addr = start; addr < end;) { unsigned long next; - vma = find_vma(mm, addr); - if (!vma || addr < vma->vm_start) + vma = vma_lookup(mm, addr); + if (!vma) break; next = min(vma->vm_end, end); - r = svm_migrate_vma_to_vram(adev, prange, vma, addr, next, trigger); + r = svm_migrate_vma_to_vram(node, prange, vma, addr, next, trigger, ttm_res_offset); if (r < 0) { pr_debug("failed %ld to migrate\n", r); break; } else { - cpages += r; + mpages += r; } + ttm_res_offset += next - addr; addr = next; } - if (cpages) + if (mpages) { prange->actual_loc = best_loc; + prange->vram_pages += mpages; + } else if (!prange->actual_loc) { + /* if no page migrated and all pages from prange are at + * sys ram drop svm_bo got from svm_range_vram_node_new + */ + svm_range_vram_node_free(prange); + } +out: + amdgpu_amdkfd_unreserve_mem_limit(node->adev, + prange->npages * PAGE_SIZE, + KFD_IOC_ALLOC_MEM_FLAGS_VRAM, + node->xcp ? node->xcp->id : 0); return r < 0 ? r : 0; } -static void svm_migrate_page_free(struct page *page) +static void svm_migrate_folio_free(struct folio *folio) { + struct page *page = &folio->page; struct svm_range_bo *svm_bo = page->zone_device_data; if (svm_bo) { @@ -563,22 +581,22 @@ static void svm_migrate_page_free(struct page *page) static int svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange, struct migrate_vma *migrate, struct dma_fence **mfence, - dma_addr_t *scratch, uint64_t npages) + dma_addr_t *scratch, u64 npages) { struct device *dev = adev->dev; - uint64_t *src; + u64 *src; dma_addr_t *dst; struct page *dpage; - uint64_t i = 0, j; - uint64_t addr; + u64 i = 0, j; + u64 addr; int r = 0; pr_debug("svms 0x%p [0x%lx 0x%lx]\n", prange->svms, prange->start, prange->last); - addr = prange->start << PAGE_SHIFT; + addr = migrate->start; - src = (uint64_t *)(scratch + npages); + src = (u64 *)(scratch + npages); dst = scratch; for (i = 0, j = 0; i < npages; i++, addr += PAGE_SIZE) { @@ -618,7 +636,7 @@ svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange, goto out_oom; } - dst[i] = dma_map_page(dev, dpage, 0, PAGE_SIZE, DMA_FROM_DEVICE); + dst[i] = dma_map_page(dev, dpage, 0, PAGE_SIZE, DMA_BIDIRECTIONAL); r = dma_mapping_error(dev, dst[i]); if (r) { dev_err(adev->dev, "%s: fail %d dma_map_page\n", __func__, r); @@ -650,28 +668,30 @@ out_oom: /** * svm_migrate_vma_to_ram - migrate range inside one vma from device to system * - * @adev: amdgpu device to migrate from * @prange: svm range structure * @vma: vm_area_struct that range [start, end] belongs to * @start: range start virtual address in pages * @end: range end virtual address in pages + * @node: kfd node device to migrate from + * @trigger: reason of migration + * @fault_page: is from vmf->page, svm_migrate_to_ram(), this is CPU page fault callback * * Context: Process context, caller hold mmap read lock, prange->migrate_mutex * * Return: - * 0 - success with all pages migrated * negative values - indicate error - * positive values - partial migration, number of pages not migrated + * positive values or zero - number of pages got migrated */ static long -svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange, - struct vm_area_struct *vma, uint64_t start, uint64_t end, +svm_migrate_vma_to_ram(struct kfd_node *node, struct svm_range *prange, + struct vm_area_struct *vma, u64 start, u64 end, uint32_t trigger, struct page *fault_page) { struct kfd_process *p = container_of(prange->svms, struct kfd_process, svms); - uint64_t npages = (end - start) >> PAGE_SHIFT; - unsigned long upages = npages; + u64 npages = (end - start) >> PAGE_SHIFT; unsigned long cpages = 0; + unsigned long mpages = 0; + struct amdgpu_device *adev = node->adev; struct kfd_process_device *pdd; struct dma_fence *mfence = NULL; struct migrate_vma migrate = { 0 }; @@ -690,7 +710,7 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange, migrate.flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE; buf = kvcalloc(npages, - 2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t), + 2 * sizeof(*migrate.src) + sizeof(u64) + sizeof(dma_addr_t), GFP_KERNEL); if (!buf) goto out; @@ -700,9 +720,9 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange, migrate.fault_page = fault_page; scratch = (dma_addr_t *)(migrate.dst + npages); - kfd_smi_event_migration_start(adev->kfd.dev, p->lead_thread->pid, + kfd_smi_event_migration_start(node, p->lead_thread->pid, start >> PAGE_SHIFT, end >> PAGE_SHIFT, - adev->kfd.dev->id, 0, prange->prefetch_loc, + node->id, 0, prange->prefetch_loc, prange->preferred_loc, trigger); r = migrate_vma_setup(&migrate); @@ -716,48 +736,50 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange, if (!cpages) { pr_debug("failed collect migrate device pages [0x%lx 0x%lx]\n", prange->start, prange->last); - upages = svm_migrate_unsuccessful_pages(&migrate); goto out_free; } if (cpages != npages) - pr_debug("partial migration, 0x%lx/0x%llx pages migrated\n", + pr_debug("partial migration, 0x%lx/0x%llx pages collected\n", cpages, npages); else - pr_debug("0x%lx pages migrated\n", cpages); + pr_debug("0x%lx pages collected\n", cpages); r = svm_migrate_copy_to_ram(adev, prange, &migrate, &mfence, scratch, npages); migrate_vma_pages(&migrate); - upages = svm_migrate_unsuccessful_pages(&migrate); - pr_debug("unsuccessful/cpages/npages 0x%lx/0x%lx/0x%lx\n", - upages, cpages, migrate.npages); + mpages = svm_migrate_successful_pages(&migrate); + pr_debug("migrated/collected/requested 0x%lx/0x%lx/0x%lx\n", + mpages, cpages, migrate.npages); svm_migrate_copy_done(adev, mfence); migrate_vma_finalize(&migrate); - kfd_smi_event_migration_end(adev->kfd.dev, p->lead_thread->pid, - start >> PAGE_SHIFT, end >> PAGE_SHIFT, - adev->kfd.dev->id, 0, trigger); - - svm_range_dma_unmap(adev->dev, scratch, 0, npages); + svm_range_dma_unmap_dev(adev->dev, scratch, 0, npages); out_free: kvfree(buf); + kfd_smi_event_migration_end(node, p->lead_thread->pid, + start >> PAGE_SHIFT, end >> PAGE_SHIFT, + node->id, 0, trigger, r); out: - if (!r && cpages) { - pdd = svm_range_get_pdd_by_adev(prange, adev); + if (!r && mpages) { + pdd = svm_range_get_pdd_by_node(prange, node); if (pdd) - WRITE_ONCE(pdd->page_out, pdd->page_out + cpages); + WRITE_ONCE(pdd->page_out, pdd->page_out + mpages); } - return r ? r : upages; + + return r ? r : mpages; } /** * svm_migrate_vram_to_ram - migrate svm range from device to system * @prange: range structure * @mm: process mm, use current->mm if NULL + * @start_mgr: start page need be migrated to sys ram + * @last_mgr: last page need be migrated to sys ram * @trigger: reason of migration + * @fault_page: is from vmf->page, svm_migrate_to_ram(), this is CPU page fault callback * * Context: Process context, caller hold mmap read lock, prange->migrate_mutex * @@ -765,61 +787,77 @@ out: * 0 - OK, otherwise error code */ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm, + unsigned long start_mgr, unsigned long last_mgr, uint32_t trigger, struct page *fault_page) { - struct amdgpu_device *adev; + struct kfd_node *node; struct vm_area_struct *vma; unsigned long addr; unsigned long start; unsigned long end; - unsigned long upages = 0; + unsigned long mpages = 0; long r = 0; + /* this pragne has no any vram page to migrate to sys ram */ if (!prange->actual_loc) { pr_debug("[0x%lx 0x%lx] already migrated to ram\n", prange->start, prange->last); return 0; } - adev = svm_range_get_adev_by_id(prange, prange->actual_loc); - if (!adev) { - pr_debug("failed to get device by id 0x%x\n", - prange->actual_loc); - return -ENODEV; + if (start_mgr < prange->start || last_mgr > prange->last) { + pr_debug("range [0x%lx 0x%lx] out prange [0x%lx 0x%lx]\n", + start_mgr, last_mgr, prange->start, prange->last); + return -EFAULT; } + node = svm_range_get_node_by_id(prange, prange->actual_loc); + if (!node) { + pr_debug("failed to get kfd node by id 0x%x\n", prange->actual_loc); + return -ENODEV; + } pr_debug("svms 0x%p prange 0x%p [0x%lx 0x%lx] from gpu 0x%x to ram\n", - prange->svms, prange, prange->start, prange->last, + prange->svms, prange, start_mgr, last_mgr, prange->actual_loc); - start = prange->start << PAGE_SHIFT; - end = (prange->last + 1) << PAGE_SHIFT; + start = start_mgr << PAGE_SHIFT; + end = (last_mgr + 1) << PAGE_SHIFT; for (addr = start; addr < end;) { unsigned long next; - vma = find_vma(mm, addr); - if (!vma || addr < vma->vm_start) { + vma = vma_lookup(mm, addr); + if (!vma) { pr_debug("failed to find vma for prange %p\n", prange); r = -EFAULT; break; } next = min(vma->vm_end, end); - r = svm_migrate_vma_to_ram(adev, prange, vma, addr, next, trigger, + r = svm_migrate_vma_to_ram(node, prange, vma, addr, next, trigger, fault_page); if (r < 0) { pr_debug("failed %ld to migrate prange %p\n", r, prange); break; } else { - upages += r; + mpages += r; } addr = next; } - if (r >= 0 && !upages) { - svm_range_vram_node_free(prange); - prange->actual_loc = 0; + if (r >= 0) { + WARN_ONCE(prange->vram_pages < mpages, + "Recorded vram pages(0x%llx) should not be less than migration pages(0x%lx).", + prange->vram_pages, mpages); + prange->vram_pages -= mpages; + + /* prange does not have vram page set its actual_loc to system + * and drop its svm_bo ref + */ + if (prange->vram_pages == 0 && prange->ttm_res) { + prange->actual_loc = 0; + svm_range_vram_node_free(prange); + } } return r < 0 ? r : 0; @@ -829,17 +867,23 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm, * svm_migrate_vram_to_vram - migrate svm range from device to device * @prange: range structure * @best_loc: the device to migrate to + * @start: start page need be migrated to sys ram + * @last: last page need be migrated to sys ram * @mm: process mm, use current->mm if NULL * @trigger: reason of migration * * Context: Process context, caller hold mmap read lock, svms lock, prange lock * + * migrate all vram pages in prange to sys ram, then migrate + * [start, last] pages from sys ram to gpu node best_loc. + * * Return: * 0 - OK, otherwise error code */ static int svm_migrate_vram_to_vram(struct svm_range *prange, uint32_t best_loc, - struct mm_struct *mm, uint32_t trigger) + unsigned long start, unsigned long last, + struct mm_struct *mm, uint32_t trigger) { int r, retries = 3; @@ -851,7 +895,8 @@ svm_migrate_vram_to_vram(struct svm_range *prange, uint32_t best_loc, pr_debug("from gpu 0x%x to gpu 0x%x\n", prange->actual_loc, best_loc); do { - r = svm_migrate_vram_to_ram(prange, mm, trigger, NULL); + r = svm_migrate_vram_to_ram(prange, mm, prange->start, prange->last, + trigger, NULL); if (r) return r; } while (prange->actual_loc && --retries); @@ -859,17 +904,21 @@ svm_migrate_vram_to_vram(struct svm_range *prange, uint32_t best_loc, if (prange->actual_loc) return -EDEADLK; - return svm_migrate_ram_to_vram(prange, best_loc, mm, trigger); + return svm_migrate_ram_to_vram(prange, best_loc, start, last, mm, trigger); } int svm_migrate_to_vram(struct svm_range *prange, uint32_t best_loc, + unsigned long start, unsigned long last, struct mm_struct *mm, uint32_t trigger) { - if (!prange->actual_loc) - return svm_migrate_ram_to_vram(prange, best_loc, mm, trigger); + if (!prange->actual_loc || prange->actual_loc == best_loc) + return svm_migrate_ram_to_vram(prange, best_loc, start, last, + mm, trigger); + else - return svm_migrate_vram_to_vram(prange, best_loc, mm, trigger); + return svm_migrate_vram_to_vram(prange, best_loc, start, last, + mm, trigger); } @@ -885,10 +934,9 @@ svm_migrate_to_vram(struct svm_range *prange, uint32_t best_loc, */ static vm_fault_t svm_migrate_to_ram(struct vm_fault *vmf) { + unsigned long start, last, size; unsigned long addr = vmf->address; struct svm_range_bo *svm_bo; - enum svm_work_list_ops op; - struct svm_range *parent; struct svm_range *prange; struct kfd_process *p; struct mm_struct *mm; @@ -925,51 +973,31 @@ static vm_fault_t svm_migrate_to_ram(struct vm_fault *vmf) mutex_lock(&p->svms.lock); - prange = svm_range_from_addr(&p->svms, addr, &parent); + prange = svm_range_from_addr(&p->svms, addr, NULL); if (!prange) { pr_debug("failed get range svms 0x%p addr 0x%lx\n", &p->svms, addr); r = -EFAULT; goto out_unlock_svms; } - mutex_lock(&parent->migrate_mutex); - if (prange != parent) - mutex_lock_nested(&prange->migrate_mutex, 1); + mutex_lock(&prange->migrate_mutex); if (!prange->actual_loc) goto out_unlock_prange; - svm_range_lock(parent); - if (prange != parent) - mutex_lock_nested(&prange->lock, 1); - r = svm_range_split_by_granularity(p, mm, addr, parent, prange); - if (prange != parent) - mutex_unlock(&prange->lock); - svm_range_unlock(parent); - if (r) { - pr_debug("failed %d to split range by granularity\n", r); - goto out_unlock_prange; - } + /* Align migration range start and size to granularity size */ + size = 1UL << prange->granularity; + start = max(ALIGN_DOWN(addr, size), prange->start); + last = min(ALIGN(addr + 1, size) - 1, prange->last); - r = svm_migrate_vram_to_ram(prange, vmf->vma->vm_mm, - KFD_MIGRATE_TRIGGER_PAGEFAULT_CPU, - vmf->page); + r = svm_migrate_vram_to_ram(prange, vmf->vma->vm_mm, start, last, + KFD_MIGRATE_TRIGGER_PAGEFAULT_CPU, vmf->page); if (r) pr_debug("failed %d migrate svms 0x%p range 0x%p [0x%lx 0x%lx]\n", - r, prange->svms, prange, prange->start, prange->last); - - /* xnack on, update mapping on GPUs with ACCESS_IN_PLACE */ - if (p->xnack_enabled && parent == prange) - op = SVM_OP_UPDATE_RANGE_NOTIFIER_AND_MAP; - else - op = SVM_OP_UPDATE_RANGE_NOTIFIER; - svm_range_add_list_work(&p->svms, parent, mm, op); - schedule_deferred_list_work(&p->svms); + r, prange->svms, prange, start, last); out_unlock_prange: - if (prange != parent) - mutex_unlock(&prange->migrate_mutex); - mutex_unlock(&parent->migrate_mutex); + mutex_unlock(&prange->migrate_mutex); out_unlock_svms: mutex_unlock(&p->svms.lock); out_unref_process: @@ -981,25 +1009,28 @@ out_mmput: } static const struct dev_pagemap_ops svm_migrate_pgmap_ops = { - .page_free = svm_migrate_page_free, + .folio_free = svm_migrate_folio_free, .migrate_to_ram = svm_migrate_to_ram, }; /* Each VRAM page uses sizeof(struct page) on system memory */ #define SVM_HMM_PAGE_STRUCT_SIZE(size) ((size)/PAGE_SIZE * sizeof(struct page)) -int svm_migrate_init(struct amdgpu_device *adev) +int kgd2kfd_init_zone_device(struct amdgpu_device *adev) { - struct kfd_dev *kfddev = adev->kfd.dev; + struct amdgpu_kfd_dev *kfddev = &adev->kfd; struct dev_pagemap *pgmap; struct resource *res = NULL; unsigned long size; void *r; - /* Page migration works on Vega10 or newer */ - if (!KFD_IS_SOC15(kfddev)) + /* Page migration works on gfx9 or newer */ + if (amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(9, 0, 1)) return -EINVAL; + if (adev->apu_prefer_gtt) + return 0; + pgmap = &kfddev->pgmap; memset(pgmap, 0, sizeof(*pgmap)); @@ -1014,7 +1045,7 @@ int svm_migrate_init(struct amdgpu_device *adev) } else { res = devm_request_free_mem_region(adev->dev, &iomem_resource, size); if (IS_ERR(res)) - return -ENOMEM; + return PTR_ERR(res); pgmap->range.start = res->start; pgmap->range.end = res->end; pgmap->type = MEMORY_DEVICE_PRIVATE; @@ -1030,11 +1061,10 @@ int svm_migrate_init(struct amdgpu_device *adev) r = devm_memremap_pages(adev->dev, pgmap); if (IS_ERR(r)) { pr_err("failed to register HMM device memory\n"); + if (pgmap->type == MEMORY_DEVICE_PRIVATE) + devm_release_mem_region(adev->dev, res->start, resource_size(res)); /* Disable SVM support capability */ pgmap->type = 0; - if (pgmap->type == MEMORY_DEVICE_PRIVATE) - devm_release_mem_region(adev->dev, res->start, - res->end - res->start + 1); return PTR_ERR(r); } @@ -1043,8 +1073,6 @@ int svm_migrate_init(struct amdgpu_device *adev) amdgpu_amdkfd_reserve_system_mem(SVM_HMM_PAGE_STRUCT_SIZE(size)); - svm_range_set_max_pages(adev); - pr_info("HMM registered %ldMB device memory\n", size >> 20); return 0; |