/* * 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
*/
/* * 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_gart_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.
*/ staticint amdgpu_gart_dummy_page_init(struct amdgpu_device *adev)
{ struct page *dummy_page = ttm_glob.dummy_read_page;
if (adev->dummy_page_addr) return 0;
adev->dummy_page_addr = dma_map_page_attrs(&adev->pdev->dev, dummy_page, 0,
PAGE_SIZE, DMA_BIDIRECTIONAL,
DMA_ATTR_SKIP_CPU_SYNC); if (dma_mapping_error(&adev->pdev->dev, 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;
}
/** * amdgpu_gart_dummy_page_fini - free dummy page used by the driver * * @adev: amdgpu_device pointer * * Frees the dummy page used by the driver (all asics).
*/ void amdgpu_gart_dummy_page_fini(struct amdgpu_device *adev)
{ if (!adev->dummy_page_addr) return;
dma_unmap_page_attrs(&adev->pdev->dev, adev->dummy_page_addr, PAGE_SIZE,
DMA_BIDIRECTIONAL,
DMA_ATTR_SKIP_CPU_SYNC);
adev->dummy_page_addr = 0;
}
/** * amdgpu_gart_table_ram_alloc - allocate system ram for gart page table * * @adev: amdgpu_device pointer * * Allocate system memory for GART page table for ASICs that don't have * dedicated VRAM. * Returns 0 for success, error for failure.
*/ int amdgpu_gart_table_ram_alloc(struct amdgpu_device *adev)
{ unsignedint order = get_order(adev->gart.table_size);
gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO; struct amdgpu_bo *bo = NULL; struct sg_table *sg = NULL; struct amdgpu_bo_param bp;
dma_addr_t dma_addr; struct page *p; unsignedlong x; int ret;
if (adev->gart.bo != NULL) return 0;
p = alloc_pages(gfp_flags, order); if (!p) return -ENOMEM;
/* assign pages to this device */ for (x = 0; x < (1UL << order); x++)
p[x].mapping = adev->mman.bdev.dev_mapping;
/* If the hardware does not support UTCL2 snooping of the CPU caches * then set_memory_wc() could be used as a workaround to mark the pages * as write combine memory.
*/
dma_addr = dma_map_page(&adev->pdev->dev, p, 0, adev->gart.table_size,
DMA_BIDIRECTIONAL); if (dma_mapping_error(&adev->pdev->dev, dma_addr)) {
dev_err(&adev->pdev->dev, "Failed to DMA MAP the GART BO page\n");
__free_pages(p, order);
p = NULL; return -EFAULT;
}
dev_info(adev->dev, "%s dma_addr:%pad\n", __func__, &dma_addr); /* Create SG table */
sg = kmalloc(sizeof(*sg), GFP_KERNEL); if (!sg) {
ret = -ENOMEM; goto error;
}
ret = sg_alloc_table(sg, 1, GFP_KERNEL); if (ret) goto error;
/** * amdgpu_gart_table_ram_free - free gart page table system ram * * @adev: amdgpu_device pointer * * Free the system memory used for the GART page tableon ASICs that don't * have dedicated VRAM.
*/ void amdgpu_gart_table_ram_free(struct amdgpu_device *adev)
{ unsignedint order = get_order(adev->gart.table_size); struct sg_table *sg = adev->gart.bo->tbo.sg; struct page *p; unsignedlong x; int ret;
ret = amdgpu_bo_reserve(adev->gart.bo, false); if (!ret) {
amdgpu_bo_unpin(adev->gart.bo);
amdgpu_bo_unreserve(adev->gart.bo);
}
amdgpu_bo_unref(&adev->gart.bo);
sg_free_table(sg);
kfree(sg);
p = virt_to_page(adev->gart.ptr); for (x = 0; x < (1UL << order); x++)
p[x].mapping = NULL;
__free_pages(p, order);
adev->gart.ptr = NULL;
}
/** * 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)
{ if (adev->gart.bo != NULL) return 0;
/** * 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)
{
amdgpu_bo_free_kernel(&adev->gart.bo, NULL, (void *)&adev->gart.ptr);
}
/* * Common gart functions.
*/ /** * 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.
*/ void 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; int idx;
if (!adev->gart.ptr) return;
if (!drm_dev_enter(adev_to_drm(adev), &idx)) return;
t = offset / AMDGPU_GPU_PAGE_SIZE;
p = t / AMDGPU_GPU_PAGES_IN_CPU_PAGE; for (i = 0; i < pages; i++, p++) {
page_base = adev->dummy_page_addr; if (!adev->gart.ptr) continue;
/** * amdgpu_gart_invalidate_tlb - invalidate gart TLB * * @adev: amdgpu device driver pointer * * Invalidate gart TLB which can be use as a way to flush gart changes *
*/ void amdgpu_gart_invalidate_tlb(struct amdgpu_device *adev)
{ int i;
if (!adev->gart.ptr) return;
mb(); if (down_read_trylock(&adev->reset_domain->sem)) {
amdgpu_device_flush_hdp(adev, NULL);
up_read(&adev->reset_domain->sem);
}
for_each_set_bit(i, adev->vmhubs_mask, AMDGPU_MAX_VMHUBS)
amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0);
}
/** * 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);
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.