[PATCH 1/2] drm/amdgpu: mask out invalid PTE address bit

[PATCH 1/2] drm/amdgpu: mask out invalid PTE address bit

[Philip Yang]

Before adding the PTE flags, explicitly mask out invalid address bits.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index a1ddf74bbdba..b40c15c9fc81 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1445,6 +1445,8 @@ static void amdgpu_vm_update_flags(struct amdgpu_vm_update_params *params,
 				   uint64_t flags)
 
 {
+	addr &= 0x0000FFFFFFFFF000UL;
+
 	if (level != AMDGPU_VM_PTB) {
 		flags |= AMDGPU_PDE_PTE;
 		amdgpu_gmc_get_vm_pde(params->adev, level, &addr, &flags);
-- 
2.17.1

[PATCH 2/2] drm/amdkfd: fix dma mapping leaking warning

[Philip Yang]

For xnack off, restore work dma unmap previous system memory page, and
dma map the updated system memory page to update GPU mapping, this is
not dma mapping leaking, remove the WARN_ONCE for dma mapping leaking.

prange->dma_addr store the VRAM page pfn after the range migrated to
VRAM, should not dma unmap VRAM page when updating GPU mapping or
remove prange. Add helper svm_is_valid_dma_mapping_addr to check VRAM
page and error cases.

Keep SVM_RANGE_VRAM_DOMAIN flag in dma_addr to diff VRAM page and dma
map system page, amdgpu vm update will mask out this bit before adding
flags to update PTE.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 57318edc4020..763c74d60c8e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -118,6 +118,13 @@ static void svm_range_remove_notifier(struct svm_range *prange)
 		mmu_interval_notifier_remove(&prange->notifier);
 }
 
+static bool
+svm_is_valid_dma_mapping_addr(struct device *dev, dma_addr_t dma_addr)
+{
+	return dma_addr && !dma_mapping_error(dev, dma_addr) &&
+	       !(dma_addr & SVM_RANGE_VRAM_DOMAIN);
+}
+
 static int
 svm_range_dma_map_dev(struct amdgpu_device *adev, struct svm_range *prange,
 		      unsigned long offset, unsigned long npages,
@@ -139,8 +146,7 @@ svm_range_dma_map_dev(struct amdgpu_device *adev, struct svm_range *prange,
 
 	addr += offset;
 	for (i = 0; i < npages; i++) {
-		if (WARN_ONCE(addr[i] && !dma_mapping_error(dev, addr[i]),
-			      "leaking dma mapping\n"))
+		if (svm_is_valid_dma_mapping_addr(dev, addr[i]))
 			dma_unmap_page(dev, addr[i], PAGE_SIZE, dir);
 
 		page = hmm_pfn_to_page(hmm_pfns[i]);
@@ -209,7 +215,7 @@ void svm_range_dma_unmap(struct device *dev, dma_addr_t *dma_addr,
 		return;
 
 	for (i = offset; i < offset + npages; i++) {
-		if (!dma_addr[i] || dma_mapping_error(dev, dma_addr[i]))
+		if (!svm_is_valid_dma_mapping_addr(dev, dma_addr[i]))
 			continue;
 		pr_debug("dma unmapping 0x%llx\n", dma_addr[i] >> PAGE_SHIFT);
 		dma_unmap_page(dev, dma_addr[i], PAGE_SIZE, dir);
@@ -1177,7 +1183,6 @@ svm_range_map_to_gpu(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 
 	for (i = offset; i < offset + npages; i++) {
 		last_domain = dma_addr[i] & SVM_RANGE_VRAM_DOMAIN;
-		dma_addr[i] &= ~SVM_RANGE_VRAM_DOMAIN;
 
 		/* Collect all pages in the same address range and memory domain
 		 * that can be mapped with a single call to update mapping.
-- 
2.17.1

Re: [PATCH 1/2] drm/amdgpu: mask out invalid PTE address bit

[Christian König]

Am 17.09.21 um 02:50 schrieb Philip Yang:
> Before adding the PTE flags, explicitly mask out invalid address bits.
>
> Signed-off-by: Philip Yang <Philip.Yang@amd.com>

Well NAK, we intentionally have a WARN_ON later on to catch errors like 
that.

Are some of the flags set in the address here?

Christian.

> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 2 ++
>   1 file changed, 2 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> index a1ddf74bbdba..b40c15c9fc81 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> @@ -1445,6 +1445,8 @@ static void amdgpu_vm_update_flags(struct amdgpu_vm_update_params *params,
>   				   uint64_t flags)
>   
>   {
> +	addr &= 0x0000FFFFFFFFF000UL;
> +
>   	if (level != AMDGPU_VM_PTB) {
>   		flags |= AMDGPU_PDE_PTE;
>   		amdgpu_gmc_get_vm_pde(params->adev, level, &addr, &flags);