/* $NetBSD: amdgpu_gart.c,v 1.10.4.2 2024/10/04 11:40:50 martin Exp $ */ /* * Copyright 2008 Advanced Micro Devices, Inc. * Copyright 2008 Red Hat Inc. * Copyright 2009 Jerome Glisse. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), * to deal in the Software without restriction, including without limitation * the rights to use, copy, modify, merge, publish, distribute, sublicense, * and/or sell copies of the Software, and to permit persons to whom the * Software is furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice shall be included in * all copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR * OTHER DEALINGS IN THE SOFTWARE. * * Authors: Dave Airlie * Alex Deucher * Jerome Glisse */ #include __KERNEL_RCSID(0, "$NetBSD: amdgpu_gart.c,v 1.10.4.2 2024/10/04 11:40:50 martin Exp $"); #include #include #include #ifdef CONFIG_X86 #include #endif #include "amdgpu.h" /* * GART * The GART (Graphics Aperture Remapping Table) is an aperture * in the GPU's address space. System pages can be mapped into * the aperture and look like contiguous pages from the GPU's * perspective. A page table maps the pages in the aperture * to the actual backing pages in system memory. * * Radeon GPUs support both an internal GART, as described above, * and AGP. AGP works similarly, but the GART table is configured * and maintained by the northbridge rather than the driver. * Radeon hw has a separate AGP aperture that is programmed to * point to the AGP aperture provided by the northbridge and the * requests are passed through to the northbridge aperture. * Both AGP and internal GART can be used at the same time, however * that is not currently supported by the driver. * * This file handles the common internal GART management. */ /* * Common GART table functions. */ /** * amdgpu_dummy_page_init - init dummy page used by the driver * * @adev: amdgpu_device pointer * * Allocate the dummy page used by the driver (all asics). * This dummy page is used by the driver as a filler for gart entries * when pages are taken out of the GART * Returns 0 on sucess, -ENOMEM on failure. */ static int amdgpu_gart_dummy_page_init(struct amdgpu_device *adev) { #ifdef __NetBSD__ int rsegs; void *p; int error; /* XXX Can this be called more than once?? */ if (adev->dummy_page_map != NULL) return 0; error = bus_dmamem_alloc(adev->ddev->dmat, PAGE_SIZE, PAGE_SIZE, 0, &adev->dummy_page_seg, 1, &rsegs, BUS_DMA_WAITOK); if (error) goto fail0; KASSERT(rsegs == 1); error = bus_dmamem_map(adev->ddev->dmat, &adev->dummy_page_seg, 1, PAGE_SIZE, &p, BUS_DMA_WAITOK); if (error) goto fail1; memset(p, 0, PAGE_SIZE); bus_dmamem_unmap(adev->ddev->dmat, p, PAGE_SIZE); error = bus_dmamap_create(adev->ddev->dmat, PAGE_SIZE, 1, PAGE_SIZE, 0, BUS_DMA_WAITOK, &adev->dummy_page_map); if (error) goto fail1; error = bus_dmamap_load_raw(adev->ddev->dmat, adev->dummy_page_map, &adev->dummy_page_seg, 1, PAGE_SIZE, BUS_DMA_WAITOK); if (error) goto fail2; bus_dmamap_sync(adev->ddev->dmat, adev->dummy_page_map, 0, PAGE_SIZE, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); /* Success! */ adev->dummy_page_addr = adev->dummy_page_map->dm_segs[0].ds_addr; return 0; fail3: __unused bus_dmamap_unload(adev->ddev->dmat, adev->dummy_page_map); fail2: bus_dmamap_destroy(adev->ddev->dmat, adev->dummy_page_map); fail1: bus_dmamem_free(adev->ddev->dmat, &adev->dummy_page_seg, 1); fail0: KASSERT(error); adev->dummy_page_map = NULL; adev->dummy_page_addr = 0; /* paranoia */ /* XXX errno NetBSD->Linux */ return -error; #else /* __NetBSD__ */ struct page *dummy_page = ttm_bo_glob.dummy_read_page; if (adev->dummy_page_addr) return 0; adev->dummy_page_addr = pci_map_page(adev->pdev, dummy_page, 0, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); if (pci_dma_mapping_error(adev->pdev, adev->dummy_page_addr)) { dev_err(&adev->pdev->dev, "Failed to DMA MAP the dummy page\n"); adev->dummy_page_addr = 0; return -ENOMEM; } return 0; #endif /* __NetBSD__ */ } /** * amdgpu_dummy_page_fini - free dummy page used by the driver * * @adev: amdgpu_device pointer * * Frees the dummy page used by the driver (all asics). */ static void amdgpu_gart_dummy_page_fini(struct amdgpu_device *adev) { if (!adev->dummy_page_addr) return; #ifdef __NetBSD__ bus_dmamap_unload(adev->ddev->dmat, adev->dummy_page_map); bus_dmamap_destroy(adev->ddev->dmat, adev->dummy_page_map); bus_dmamem_free(adev->ddev->dmat, &adev->dummy_page_seg, 1); adev->dummy_page_map = NULL; #else pci_unmap_page(adev->pdev, adev->dummy_page_addr, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); #endif adev->dummy_page_addr = 0; } /** * amdgpu_gart_table_vram_alloc - allocate vram for gart page table * * @adev: amdgpu_device pointer * * Allocate video memory for GART page table * (pcie r4xx, r5xx+). These asics require the * gart table to be in video memory. * Returns 0 for success, error for failure. */ int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev) { int r; if (adev->gart.bo == NULL) { struct amdgpu_bo_param bp; memset(&bp, 0, sizeof(bp)); bp.size = adev->gart.table_size; bp.byte_align = PAGE_SIZE; bp.domain = AMDGPU_GEM_DOMAIN_VRAM; bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED | AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS; bp.type = ttm_bo_type_kernel; bp.resv = NULL; r = amdgpu_bo_create(adev, &bp, &adev->gart.bo); if (r) { return r; } } return 0; } /** * amdgpu_gart_table_vram_pin - pin gart page table in vram * * @adev: amdgpu_device pointer * * Pin the GART page table in vram so it will not be moved * by the memory manager (pcie r4xx, r5xx+). These asics require the * gart table to be in video memory. * Returns 0 for success, error for failure. */ int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev) { int r; r = amdgpu_bo_reserve(adev->gart.bo, false); if (unlikely(r != 0)) return r; r = amdgpu_bo_pin(adev->gart.bo, AMDGPU_GEM_DOMAIN_VRAM); if (r) { amdgpu_bo_unreserve(adev->gart.bo); return r; } r = amdgpu_bo_kmap(adev->gart.bo, &adev->gart.ptr); if (r) amdgpu_bo_unpin(adev->gart.bo); amdgpu_bo_unreserve(adev->gart.bo); return r; } /** * amdgpu_gart_table_vram_unpin - unpin gart page table in vram * * @adev: amdgpu_device pointer * * Unpin the GART page table in vram (pcie r4xx, r5xx+). * These asics require the gart table to be in video memory. */ void amdgpu_gart_table_vram_unpin(struct amdgpu_device *adev) { int r; if (adev->gart.bo == NULL) { return; } r = amdgpu_bo_reserve(adev->gart.bo, true); if (likely(r == 0)) { amdgpu_bo_kunmap(adev->gart.bo); amdgpu_bo_unpin(adev->gart.bo); amdgpu_bo_unreserve(adev->gart.bo); adev->gart.ptr = NULL; } } /** * amdgpu_gart_table_vram_free - free gart page table vram * * @adev: amdgpu_device pointer * * Free the video memory used for the GART page table * (pcie r4xx, r5xx+). These asics require the gart table to * be in video memory. */ void amdgpu_gart_table_vram_free(struct amdgpu_device *adev) { if (adev->gart.bo == NULL) { return; } amdgpu_bo_unref(&adev->gart.bo); } #ifdef __NetBSD__ static void amdgpu_gart_pre_update(struct amdgpu_device *adev, unsigned gpu_pgstart, unsigned gpu_npages) { if (adev->gart.ag_table_map != NULL) { const unsigned entsize = adev->gart.table_size / adev->gart.num_gpu_pages; bus_dmamap_sync(adev->ddev->dmat, adev->gart.ag_table_map, gpu_pgstart*entsize, gpu_npages*entsize, BUS_DMASYNC_POSTWRITE); } } static void amdgpu_gart_post_update(struct amdgpu_device *adev, unsigned gpu_pgstart, unsigned gpu_npages) { unsigned i; if (adev->gart.ag_table_map != NULL) { const unsigned entsize = adev->gart.table_size / adev->gart.num_gpu_pages; bus_dmamap_sync(adev->ddev->dmat, adev->gart.ag_table_map, gpu_pgstart*entsize, gpu_npages*entsize, BUS_DMASYNC_PREWRITE); } mb(); /* XXX why is bus_dmamap_sync not enough? */ amdgpu_asic_flush_hdp(adev, NULL); for (i = 0; i < adev->num_vmhubs; i++) amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0); } #endif /* * Common gart functions. */ #ifdef __NetBSD__ int amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t gpu_start, unsigned npages) { const unsigned gpu_per_cpu = AMDGPU_GPU_PAGES_IN_CPU_PAGE; const unsigned gpu_npages = (npages * gpu_per_cpu); const uint64_t gpu_pgstart = (gpu_start / AMDGPU_GPU_PAGE_SIZE); const uint64_t pgstart __diagused = (gpu_pgstart / gpu_per_cpu); uint64_t pgno, gpu_pgno; uint32_t flags = AMDGPU_PTE_SYSTEM; KASSERT(pgstart == (gpu_start / PAGE_SIZE)); KASSERT(npages <= adev->gart.num_cpu_pages); KASSERT(gpu_npages <= adev->gart.num_cpu_pages); if (!adev->gart.ready) { WARN(1, "trying to bind memory to uninitialized GART !\n"); return -EINVAL; } amdgpu_gart_pre_update(adev, gpu_pgstart, gpu_npages); for (pgno = 0; pgno < npages; pgno++) { #ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS adev->gart.pages[pgstart + pgno] = NULL; #endif if (adev->gart.ptr == NULL) continue; for (gpu_pgno = 0; gpu_pgno < gpu_per_cpu; gpu_pgno++) { amdgpu_gmc_set_pte_pde(adev, adev->gart.ptr, gpu_pgstart + gpu_per_cpu*pgno + gpu_pgno, adev->dummy_page_addr, flags); } } amdgpu_gart_post_update(adev, gpu_pgstart, gpu_npages); return 0; } #else /* __NetBSD__ */ /** * amdgpu_gart_unbind - unbind pages from the gart page table * * @adev: amdgpu_device pointer * @offset: offset into the GPU's gart aperture * @pages: number of pages to unbind * * Unbinds the requested pages from the gart page table and * replaces them with the dummy page (all asics). * Returns 0 for success, -EINVAL for failure. */ int amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, int pages) { unsigned t; unsigned p; int i, j; u64 page_base; /* Starting from VEGA10, system bit must be 0 to mean invalid. */ uint64_t flags = 0; if (!adev->gart.ready) { WARN(1, "trying to unbind memory from uninitialized GART !\n"); return -EINVAL; } t = offset / AMDGPU_GPU_PAGE_SIZE; p = t / AMDGPU_GPU_PAGES_IN_CPU_PAGE; for (i = 0; i < pages; i++, p++) { #ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS adev->gart.pages[p] = NULL; #endif page_base = adev->dummy_page_addr; if (!adev->gart.ptr) continue; for (j = 0; j < AMDGPU_GPU_PAGES_IN_CPU_PAGE; j++, t++) { amdgpu_gmc_set_pte_pde(adev, adev->gart.ptr, t, page_base, flags); page_base += AMDGPU_GPU_PAGE_SIZE; } } mb(); amdgpu_asic_flush_hdp(adev, NULL); for (i = 0; i < adev->num_vmhubs; i++) amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0); return 0; } #endif /* __NetBSD__ */ /** * amdgpu_gart_map - map dma_addresses into GART entries * * @adev: amdgpu_device pointer * @offset: offset into the GPU's gart aperture * @pages: number of pages to bind * @dma_addr: DMA addresses of pages * @flags: page table entry flags * @dst: CPU address of the gart table * * Map the dma_addresses into GART entries (all asics). * Returns 0 for success, -EINVAL for failure. */ #ifdef __NetBSD__ int amdgpu_gart_map(struct amdgpu_device *adev, uint64_t gpu_start, unsigned npages, bus_size_t map_start, bus_dmamap_t dmamap, uint32_t flags, void *dst) { bus_size_t seg_off = 0; unsigned i, j, t; CTASSERT(AMDGPU_GPU_PAGE_SIZE <= PAGE_SIZE); CTASSERT((PAGE_SIZE % AMDGPU_GPU_PAGE_SIZE) == 0); KASSERT((gpu_start & (PAGE_SIZE - 1)) == 0); if (!adev->gart.ready) { WARN(1, "trying to bind memory to uninitialized GART !\n"); return -EINVAL; } for (i = 0; i < dmamap->dm_nsegs; i++) { KASSERT((dmamap->dm_segs[i].ds_len & (PAGE_SIZE - 1)) == 0); if (map_start == 0) break; if (map_start < dmamap->dm_segs[i].ds_len) { seg_off = map_start; break; } map_start -= dmamap->dm_segs[i].ds_len; } KASSERT(i < dmamap->dm_nsegs); t = gpu_start / AMDGPU_GPU_PAGE_SIZE; for (i = 0; npages --> 0;) { for (j = 0; j < AMDGPU_GPU_PAGES_IN_CPU_PAGE; j++, t++) { KASSERT(i < dmamap->dm_nsegs); KASSERT(seg_off < dmamap->dm_segs[i].ds_len); amdgpu_gmc_set_pte_pde(adev, dst, t, dmamap->dm_segs[i].ds_addr + seg_off, flags); seg_off += AMDGPU_GPU_PAGE_SIZE; if (seg_off == dmamap->dm_segs[i].ds_len) { i++; seg_off = 0; } } } return 0; } #else int amdgpu_gart_map(struct amdgpu_device *adev, uint64_t offset, int pages, dma_addr_t *dma_addr, uint64_t flags, void *dst) { uint64_t page_base; unsigned i, j, t; if (!adev->gart.ready) { WARN(1, "trying to bind memory to uninitialized GART !\n"); return -EINVAL; } t = offset / AMDGPU_GPU_PAGE_SIZE; for (i = 0; i < pages; i++) { page_base = dma_addr[i]; for (j = 0; j < AMDGPU_GPU_PAGES_IN_CPU_PAGE; j++, t++) { amdgpu_gmc_set_pte_pde(adev, dst, t, page_base, flags); page_base += AMDGPU_GPU_PAGE_SIZE; } } return 0; } #endif /* __NetBSD__ */ #ifdef __NetBSD__ int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t gpu_start, unsigned npages, struct page **pages, bus_dmamap_t dmamap, uint32_t flags) { const unsigned gpu_per_cpu = AMDGPU_GPU_PAGES_IN_CPU_PAGE; const unsigned gpu_npages = (npages * gpu_per_cpu); const uint64_t gpu_pgstart = (gpu_start / AMDGPU_GPU_PAGE_SIZE); const uint64_t pgstart __diagused = (gpu_pgstart / gpu_per_cpu); uint64_t pgno, gpu_pgno; KASSERT(pgstart == (gpu_start / PAGE_SIZE)); KASSERT(npages == dmamap->dm_nsegs); KASSERT(npages <= adev->gart.num_cpu_pages); KASSERT(gpu_npages <= adev->gart.num_cpu_pages); if (!adev->gart.ready) { WARN(1, "trying to bind memory to uninitialized GART !\n"); return -EINVAL; } amdgpu_gart_pre_update(adev, gpu_pgstart, gpu_npages); for (pgno = 0; pgno < npages; pgno++) { const bus_addr_t addr = dmamap->dm_segs[pgno].ds_addr; KASSERT(dmamap->dm_segs[pgno].ds_len == PAGE_SIZE); #ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS adev->gart.pages[pgstart + pgno] = NULL; #endif if (adev->gart.ptr == NULL) continue; for (gpu_pgno = 0; gpu_pgno < gpu_per_cpu; gpu_pgno++) { amdgpu_gmc_set_pte_pde(adev, adev->gart.ptr, gpu_pgstart + gpu_per_cpu*pgno + gpu_pgno, addr + gpu_pgno*AMDGPU_GPU_PAGE_SIZE, flags); } } amdgpu_gart_post_update(adev, gpu_pgstart, gpu_npages); return 0; } #else /* __NetBSD__ */ /** * amdgpu_gart_bind - bind pages into the gart page table * * @adev: amdgpu_device pointer * @offset: offset into the GPU's gart aperture * @pages: number of pages to bind * @pagelist: pages to bind * @dma_addr: DMA addresses of pages * @flags: page table entry flags * * Binds the requested pages to the gart page table * (all asics). * Returns 0 for success, -EINVAL for failure. */ int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset, int pages, struct page **pagelist, dma_addr_t *dma_addr, uint64_t flags) { #ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS unsigned t,p; #endif int r, i; if (!adev->gart.ready) { WARN(1, "trying to bind memory to uninitialized GART !\n"); return -EINVAL; } #ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS t = offset / AMDGPU_GPU_PAGE_SIZE; p = t / AMDGPU_GPU_PAGES_IN_CPU_PAGE; for (i = 0; i < pages; i++, p++) adev->gart.pages[p] = pagelist ? pagelist[i] : NULL; #endif if (!adev->gart.ptr) return 0; r = amdgpu_gart_map(adev, offset, pages, dma_addr, flags, adev->gart.ptr); if (r) return r; mb(); amdgpu_asic_flush_hdp(adev, NULL); for (i = 0; i < adev->num_vmhubs; i++) amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0); return 0; } #endif /** * amdgpu_gart_init - init the driver info for managing the gart * * @adev: amdgpu_device pointer * * Allocate the dummy page and init the gart driver info (all asics). * Returns 0 for success, error for failure. */ int amdgpu_gart_init(struct amdgpu_device *adev) { int r; if (adev->dummy_page_addr) return 0; /* We need PAGE_SIZE >= AMDGPU_GPU_PAGE_SIZE */ if (PAGE_SIZE < AMDGPU_GPU_PAGE_SIZE) { DRM_ERROR("Page size is smaller than GPU page size!\n"); return -EINVAL; } r = amdgpu_gart_dummy_page_init(adev); if (r) return r; /* Compute table size */ adev->gart.num_cpu_pages = adev->gmc.gart_size / PAGE_SIZE; adev->gart.num_gpu_pages = adev->gmc.gart_size / AMDGPU_GPU_PAGE_SIZE; DRM_INFO("GART: num cpu pages %u, num gpu pages %u\n", adev->gart.num_cpu_pages, adev->gart.num_gpu_pages); #ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS /* Allocate pages table */ adev->gart.pages = vzalloc(array_size(sizeof(void *), adev->gart.num_cpu_pages)); if (adev->gart.pages == NULL) return -ENOMEM; #endif return 0; } /** * amdgpu_gart_fini - tear down the driver info for managing the gart * * @adev: amdgpu_device pointer * * Tear down the gart driver info and free the dummy page (all asics). */ void amdgpu_gart_fini(struct amdgpu_device *adev) { #ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS vfree(adev->gart.pages); adev->gart.pages = NULL; #endif amdgpu_gart_dummy_page_fini(adev); }