| Message ID | 20161110182345.31777-3-mperttunen@nvidia.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Nov 10, 2016 at 08:23:39PM +0200, Mikko Perttunen wrote: > On 64-bit Tegras, buffer object memory allocation may > return memory above 4G that units behind Host1x cannot > access. Add the GFP_DMA flag to these allocation when > IOMMU is not enabled to ensure units can always access > BO memory. I don't think that's entirely true. The display controller can address more than 32 bits. It's perhaps slightly different from other units that are behind host1x, but maybe we need a special case here to allow frame- buffers to reside in higher memory? Thierry
On 05.12.2016 20:39, Thierry Reding wrote: > On Thu, Nov 10, 2016 at 08:23:39PM +0200, Mikko Perttunen wrote: >> On 64-bit Tegras, buffer object memory allocation may >> return memory above 4G that units behind Host1x cannot >> access. Add the GFP_DMA flag to these allocation when >> IOMMU is not enabled to ensure units can always access >> BO memory. > > I don't think that's entirely true. The display controller can address > more than 32 bits. It's perhaps slightly different from other units that > are behind host1x, but maybe we need a special case here to allow frame- > buffers to reside in higher memory? I checked and looks like all (or at least most) units can actually address 34 bits /almost always/. The thing that tripped me up was that apparently the firmware for VIC still needs be in the lower 32 bits. So we should be able to drop this patch. > > Thierry > Thanks, Mikko.
diff --git a/drivers/gpu/drm/tegra/gem.c b/drivers/gpu/drm/tegra/gem.c index 19bf9cd..e36e6c5 100644 --- a/drivers/gpu/drm/tegra/gem.c +++ b/drivers/gpu/drm/tegra/gem.c @@ -233,7 +233,7 @@ static int tegra_bo_alloc(struct drm_device *drm, struct tegra_bo *bo) size_t size = bo->gem.size; bo->vaddr = dma_alloc_wc(drm->dev, size, &bo->paddr, - GFP_KERNEL | __GFP_NOWARN); + GFP_KERNEL | GFP_DMA | __GFP_NOWARN); if (!bo->vaddr) { dev_err(drm->dev, "failed to allocate buffer of size %zu\n",
On 64-bit Tegras, buffer object memory allocation may return memory above 4G that units behind Host1x cannot access. Add the GFP_DMA flag to these allocation when IOMMU is not enabled to ensure units can always access BO memory. Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com> --- drivers/gpu/drm/tegra/gem.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)