This patch fixes gem buffer allocation type checking. EXYNOS_BO_CONTIG has 0 so the checking should be fixed to 'if (!(flags & EXYNOS_BO_NONCONTIG))'
Signed-off-by: Inki Dae inki.dae@samsung.com Signed-off-by: Kyungmin Park kyungmin.park@samsung.com --- drivers/gpu/drm/exynos/exynos_drm_buf.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/drivers/gpu/drm/exynos/exynos_drm_buf.c b/drivers/gpu/drm/exynos/exynos_drm_buf.c index 74592d1..911f7fd 100644 --- a/drivers/gpu/drm/exynos/exynos_drm_buf.c +++ b/drivers/gpu/drm/exynos/exynos_drm_buf.c @@ -38,7 +38,7 @@ static int lowlevel_buffer_allocate(struct drm_device *dev, * region will be allocated else physically contiguous * as possible. */ - if (flags & EXYNOS_BO_CONTIG) + if (!(flags & EXYNOS_BO_NONCONTIG)) dma_set_attr(DMA_ATTR_FORCE_CONTIGUOUS, &buf->dma_attrs);
/*
On Thu, Dec 27, 2012 at 4:27 PM, Inki Dae inki.dae@samsung.com wrote:
This patch fixes gem buffer allocation type checking. EXYNOS_BO_CONTIG has 0 so the checking should be fixed to 'if (!(flags & EXYNOS_BO_NONCONTIG))'
Signed-off-by: Inki Dae inki.dae@samsung.com Signed-off-by: Kyungmin Park kyungmin.park@samsung.com
drivers/gpu/drm/exynos/exynos_drm_buf.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/drivers/gpu/drm/exynos/exynos_drm_buf.c b/drivers/gpu/drm/exynos/exynos_drm_buf.c index 74592d1..911f7fd 100644 --- a/drivers/gpu/drm/exynos/exynos_drm_buf.c +++ b/drivers/gpu/drm/exynos/exynos_drm_buf.c @@ -38,7 +38,7 @@ static int lowlevel_buffer_allocate(struct drm_device *dev, * region will be allocated else physically contiguous * as possible. */
if (flags & EXYNOS_BO_CONTIG)
if (!(flags & EXYNOS_BO_NONCONTIG)) dma_set_attr(DMA_ATTR_FORCE_CONTIGUOUS, &buf->dma_attrs);
Hi Mr. Dae,
If iommu is supported, we would have called arm_iommu_attach_device. So dma_alloc_attrs will always call arm_iommu_alloc_attrs.
If iommu is not supported, dma_alloc_attrs will call arm_dma_alloc which will anyway allocate a contiguous buffer if possible.
With this code, we are forcing the root framebuffer (fbdev) to be contiguous since fbdev allocation does not pass noncontig flag.
Why do we need to force the allocation of a contiguous buffer when iommu is supported?
Regards, Prathyush
/*-- 1.7.4.1
dri-devel mailing list dri-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/dri-devel
2012/12/27 Prathyush K prathyush@chromium.org
On Thu, Dec 27, 2012 at 4:27 PM, Inki Dae inki.dae@samsung.com wrote:
This patch fixes gem buffer allocation type checking. EXYNOS_BO_CONTIG has 0 so the checking should be fixed to 'if (!(flags & EXYNOS_BO_NONCONTIG))'
Signed-off-by: Inki Dae inki.dae@samsung.com Signed-off-by: Kyungmin Park kyungmin.park@samsung.com
drivers/gpu/drm/exynos/exynos_drm_buf.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/drivers/gpu/drm/exynos/exynos_drm_buf.c b/drivers/gpu/drm/exynos/exynos_drm_buf.c index 74592d1..911f7fd 100644 --- a/drivers/gpu/drm/exynos/exynos_drm_buf.c +++ b/drivers/gpu/drm/exynos/exynos_drm_buf.c @@ -38,7 +38,7 @@ static int lowlevel_buffer_allocate(struct drm_device *dev, * region will be allocated else physically contiguous * as possible. */
if (flags & EXYNOS_BO_CONTIG)
if (!(flags & EXYNOS_BO_NONCONTIG)) dma_set_attr(DMA_ATTR_FORCE_CONTIGUOUS, &buf->dma_attrs);Hi Mr. Dae,
If iommu is supported, we would have called arm_iommu_attach_device. So dma_alloc_attrs will always call arm_iommu_alloc_attrs.
If iommu is not supported, dma_alloc_attrs will call arm_dma_alloc which will anyway allocate a contiguous buffer if possible.
With this code, we are forcing the root framebuffer (fbdev) to be contiguous since fbdev allocation does not pass noncontig flag.
we could use noncontig flag simply like below,
if (iommu is supported) exynos_drm_gem_create(dev, EXYNOS_BO_NONCONTIG, size); else exynos_drm_gem_create(dev, EXYNOS_BO_CONTIG, size);
Why do we need to force the allocation of a contiguous buffer when iommu is supported?
Simply saying, the reason is that we want to pass boot logo drew by boot loader without iommu to kernel side without memory copy.
Regards, Prathyush
/*-- 1.7.4.1
dri-devel mailing list dri-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/dri-devel
dri-devel mailing list dri-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/dri-devel
dri-devel@lists.freedesktop.org
-
Inki Dae -
Prathyush K