gpu: nvgpu: Remove sg_phys() from GMMU code

Remove the last sg_phys() call from the GMMU code and replace it
with a generic nvgpu_mem API. This new API, nvgpu_mem_get_phys_addr(),
returns the physical address of an nvgpu_mem struct.

Also, implement this new API in the Linux specific nvgpu_mem code
since it requires access to the underlying SGT/SGL.

JIRA NVGPU-68

Change-Id: Idf88701a2a8515464c658c26e0de493c82ff850d
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1542964
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

1 files changed, 25 insertions, 0 deletions

diff --git a/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c b/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c
index 8d8909dd..8cf3011e 100644
--- a/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c
+++ b/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c
@@ -284,6 +284,14 @@ static u64 nvgpu_mem_get_addr_sysmem(struct gk20a *g, struct nvgpu_mem *mem)
  * Return the base address of %mem. Handles whether this is a VIDMEM or SYSMEM
  * allocation.
  *
+ * Note: this API does not make sense to use for _VIDMEM_ buffers with greater
+ * than one scatterlist chunk. If there's more than one scatterlist chunk then
+ * the buffer will not be contiguous. As such the base address probably isn't
+ * very useful. This is true for SYSMEM as well, if there's no IOMMU.
+ *
+ * However! It _is_ OK to use this on discontiguous sysmem buffers _if_ there's
+ * an IOMMU present and enabled for the GPU.
+ *
  * %attrs can be NULL. If it is not NULL then it may be inspected to determine
  * if the address needs to be modified before writing into a PTE.
  */
@@ -306,6 +314,23 @@ u64 nvgpu_mem_get_addr(struct gk20a *g, struct nvgpu_mem *mem)
 }
 
 /*
+ * This should only be used on contiguous buffers regardless of whether
+ * there's an IOMMU present/enabled. This applies to both SYSMEM and
+ * VIDMEM.
+ */
+u64 nvgpu_mem_get_phys_addr(struct gk20a *g, struct nvgpu_mem *mem)
+{
+        /*
+         * For a VIDMEM buf, this is identical to simply get_addr() so just fall
+         * back to that.
+         */
+        if (mem->aperture == APERTURE_VIDMEM)
+                return nvgpu_mem_get_addr(g, mem);
+
+        return sg_phys(mem->priv.sgt->sgl);
+}
+
+/*
  * Be careful how you use this! You are responsible for correctly freeing this
  * memory.
  */