nv-tegra-mirror/linux-nvgpu
1
0
Fork 0
mirror of git://nv-tegra.nvidia.com/linux-nvgpu.git synced 2025-12-22 17:36:20 +03:00
Code Issues Packages Projects Releases Wiki Activity

gpu: nvgpu: Use our own vmap() for coherent DMA buffers

Browse Source 
For some reason the GPU does not like the mappings created by the
DMA API for coherent sysmem buffers. But a plain vmap() does seem
to work. To work around this, when we are using coherent sysmem,
force the NO_KERNEL_MAPPING flag to on and then make a vmap() in
the nvgpu DMA API wrapper. The rest of the driver will be none the
wiser but will work as expected.

This problem is not understood yet but it is being tracked in bug
2040115. Once this bug is understood this WAR should either be
determined as necessary or reverted with an appropriate fix.

Bug 2040115
JIRA EVLR-2333

Change-Id: Idae7a0c92441f0309df572ac18697af49bb6ff2b
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1657568
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
This commit is contained in:
Alex Waterman
2018-02-14 10:58:24 -08:00
committed by mobile promotions
parent 1170687c33
commit 3fdd8e38b2
4 changed files with 63 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
drivers/gpu/nvgpu/common/linux/dma.c
View File

@@ -221,6 +221,16 @@ int nvgpu_dma_alloc_flags_sys(struct gk20a *g, unsigned long flags,
NVGPU_DEFINE_DMA_ATTRS(dma_attrs);
void *alloc_ret;
/*
* WAR for IO coherent chips: the DMA API does not seem to generate
* mappings that work correctly. Unclear why - Bug ID: 2040115.
*
* Basically we just tell the DMA API not to map with NO_KERNEL_MAPPING
* and then make a vmap() ourselves.
*/
if (nvgpu_is_enabled(g, NVGPU_USE_COHERENT_SYSMEM))
flags |= NVGPU_DMA_NO_KERNEL_MAPPING;
/*
* Before the debug print so we see this in the total. But during
* cleanup in the fail path this has to be subtracted.
@@ -255,7 +265,17 @@ int nvgpu_dma_alloc_flags_sys(struct gk20a *g, unsigned long flags,
iova, size, flags);
}
if (err)
goto fail_free;
goto fail_free_dma;
if (nvgpu_is_enabled(g, NVGPU_USE_COHERENT_SYSMEM)) {
mem->cpu_va = vmap(mem->priv.pages,
size >> PAGE_SHIFT,
0, PAGE_KERNEL);
if (!mem->cpu_va) {
err = -ENOMEM;
goto fail_free_sgt;
}
}
mem->aligned_size = size;
mem->aperture = APERTURE_SYSMEM;
@@ -265,12 +285,14 @@ int nvgpu_dma_alloc_flags_sys(struct gk20a *g, unsigned long flags,
return 0;
fail_free:
g->dma_memory_used -= mem->aligned_size;
fail_free_sgt:
nvgpu_free_sgtable(g, &mem->priv.sgt);
fail_free_dma:
dma_free_attrs(d, size, alloc_ret, iova, NVGPU_DMA_ATTR(dma_attrs));
mem->cpu_va = NULL;
mem->priv.sgt = NULL;
mem->size = 0;
g->dma_memory_used -= mem->aligned_size;
return err;
}
@@ -466,6 +488,12 @@ static void nvgpu_dma_free_sys(struct gk20a *g, struct nvgpu_mem *mem)
if (!(mem->mem_flags & NVGPU_MEM_FLAG_SHADOW_COPY) &&
!(mem->mem_flags & __NVGPU_MEM_FLAG_NO_DMA) &&
(mem->cpu_va || mem->priv.pages)) {
/*
* Free side of WAR for bug 2040115.
*/
if (nvgpu_is_enabled(g, NVGPU_USE_COHERENT_SYSMEM))
vunmap(mem->cpu_va);
if (mem->priv.flags) {
NVGPU_DEFINE_DMA_ATTRS(dma_attrs);