nv-tegra-mirror/linux-nvgpu
1
0
Fork 0
mirror of git://nv-tegra.nvidia.com/linux-nvgpu.git synced 2025-12-24 02:22:34 +03:00
Code Issues Packages Projects Releases Wiki Activity

gpu: nvgpu: Use only contig CBCs

Browse Source 
Modify the LTC code to only use a contiguous CompBit Cache (CBC). The
original code had two allocation schemes: "physical" and "virtual" -
what they meant was virtually contiguous or physically contiguous. The
CBC must appear contiguous to the GPU be it either from the IOMMU or
from physical pages allocated contiguously.

This change makes the CBC get allocated with the FORCE_CONTIGUOUS flag
if the GPU is not IOMMU'able. If we can get contiguous mem with the
IOMMU then no need to force the underlying pages to be contiguous.
However, not all GPUs may be IOMMU'able so we do need to handle that
case.

Also delete the gk20a/ltc_gk20a.[ch] code. All that remained in these
files was the CBC alloc functions which were completely chip agnostic.
As a result these functions were consolidated and moved to common/ltc.c.

Bug 2015747

Change-Id: I3f41961b4f94378b954e7502a6b27cf0bc627375
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1593666
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
This commit is contained in:
Alex Waterman
2017-11-07 10:50:49 -08:00
committed by mobile promotions
parent e7c4547889
commit 016231c045
7 changed files with 29 additions and 95 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
drivers/gpu/nvgpu/common/ltc.c
View File

@@ -21,8 +21,10 @@
*/
#include <nvgpu/ltc.h>
#include <nvgpu/dma.h>
#include "gk20a/gk20a.h"
#include "gk20a/gr_gk20a.h"
int nvgpu_init_ltc_support(struct gk20a *g)
{
@@ -46,3 +48,17 @@ void nvgpu_ltc_sync_enabled(struct gk20a *g)
}
nvgpu_spinlock_release(&g->ltc_enabled_lock);
}
int nvgpu_ltc_alloc_cbc(struct gk20a *g, size_t compbit_backing_size)
{
struct gr_gk20a *gr = &g->gr;
unsigned long flags = 0;
if (!nvgpu_iommuable(g))
flags = NVGPU_DMA_FORCE_CONTIGUOUS;
return nvgpu_dma_alloc_flags_sys(g,
flags,
compbit_backing_size,
&gr->compbit_store.mem);
}