Skip to contentSkip to conversation

drm/amdgpu: fix vm size and block size for Vega10 VMPT

PATCH In review Junwei Zhang 2 participants 4 messages Mar 31, 2017 – Apr 1, 2017 2 messages not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/2 drm/amdgpu: fix vm size and block size for Vega10 VMPT

From: "Zhang, Jerry" <[email protected]>

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  8 --------
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c      | 15 ++++++++-------
 2 files changed, 8 insertions(+), 15 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0 -8
index 363d73c..98555fd 100644
@@ -1041,14 +1041,6 @@ static bool amdgpu_check_pot_argument(int arg)
 
 static void amdgpu_get_block_size(struct amdgpu_device *adev)
 {
-	/* from AI, asic starts to support multiple level VMPT */
-	if (adev->asic_type >= CHIP_VEGA10) {
-		if (amdgpu_vm_block_size != 9)
-			dev_warn(adev->dev,
-				 "Multi-VMPT limits block size to one page!\n");
-		amdgpu_vm_block_size = 9;
-		return;
-	}
 	/* defines number of bits in page table versus page directory,
 	 * a page is 4KB so we have 12 bits offset, minimum 9 bits in the
 	 * page table and the remaining bits are in the page directory */
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +8 -7
index df69aae..081a676 100644
@@ -546,6 +546,14 @@ static int gmc_v9_0_sw_init(void *handle)
 	} else {
 		/* XXX Don't know how to get VRAM type yet. */
 		adev->mc.vram_type = AMDGPU_VRAM_TYPE_HBM;
+
+		/*
+		 * To fulfill 4-level page support,
+		 * vm size is 256TB (48bit), maximum size of Vega10,
+		 * block size 512 (9bit)
+		 */
+		amdgpu_vm_size = 1U << 18;
+		amdgpu_vm_block_size = 9;
 	}
 
 	/* This interrupt is VMC page fault.*/
@@ -557,13 +565,6 @@ static int gmc_v9_0_sw_init(void *handle)
 	if (r)
 		return r;
 
-	/* Because of four level VMPTs, vm size is at least 512GB.
-	 * The maximum size is 256TB (48bit).
-	 */
-	if (amdgpu_vm_size < 512) {
-		DRM_WARN("VM size is at least 512GB!\n");
-		amdgpu_vm_size = 512;
-	}
 	adev->vm_manager.max_pfn = (uint64_t)amdgpu_vm_size << 18;
 
 	/* Set the internal MC address mask
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 1490928723-2184-1-git-send-email-Jerry.Zhang-5C7GfCeVMHo@public.gmane.org

#2 lore raw

PATCH 2/2 drm/amdgpu: rename amdgpu_get_block_size to amdgpu_check_block_size

Here just check if the block size is reasonable, otherwise set it as default vaule.
The entity actual vaule should be set globally or in gmc.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -2
index 98555fd..714be3c 100644
@@ -1039,7 +1039,7 @@ static bool amdgpu_check_pot_argument(int arg)
 	return (arg & (arg - 1)) == 0;
 }
 
-static void amdgpu_get_block_size(struct amdgpu_device *adev)
+static void amdgpu_check_block_size(struct amdgpu_device *adev)
 {
 	/* defines number of bits in page table versus page directory,
 	 * a page is 4KB so we have 12 bits offset, minimum 9 bits in the
@@ -1131,7 +1131,7 @@ static void amdgpu_check_arguments(struct amdgpu_device *adev)
 
 	amdgpu_check_vm_size(adev);
 
-	amdgpu_get_block_size(adev);
+	amdgpu_check_block_size(adev);
 
 	if (amdgpu_vram_page_split != -1 && (amdgpu_vram_page_split < 16 ||
 	    !amdgpu_check_pot_argument(amdgpu_vram_page_split))) {
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#3 lore raw
In reply to a message not in the archive
Quoted 44 lines

-----Original Message-----
From: Junwei Zhang [mailto:[email protected]]
Sent: Thursday, March 30, 2017 10:52 PM
To: Deucher, Alexander; Koenig, Christian
Cc: [email protected]; Zhang, Jerry
Subject: [PATCH 1/2 v2] drm/amdgpu: fix vm size and block size for Vega10 VMPT

From: "Zhang, Jerry" <[email protected]>

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  8 --------
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c      | 15 ++++++++-------
 2 files changed, 8 insertions(+), 15 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0 -8
index 363d73c..98555fd 100644
@@ -1041,14 +1041,6 @@ static bool amdgpu_check_pot_argument(int arg)
 
 static void amdgpu_get_block_size(struct amdgpu_device *adev)
 {
-	/* from AI, asic starts to support multiple level VMPT */
-	if (adev->asic_type >= CHIP_VEGA10) {
-		if (amdgpu_vm_block_size != 9)
-			dev_warn(adev->dev,
-				 "Multi-VMPT limits block size to one
 page!\n");
-		amdgpu_vm_block_size = 9;
-		return;
-	}
 	/* defines number of bits in page table versus page directory,
 	 * a page is 4KB so we have 12 bits offset, minimum 9 bits in the
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0 -0
 	 * page table and the remaining bits are in the page directory */
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +0 -0
index df69aae..081a676 100644
@@ -546,6 +546,14 @@ static int gmc_v9_0_sw_init(void *handle)
 	} else {

drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c static int gmc_v9_0_sw_init(void *handle) · line 556

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +8 -0
 		/* XXX Don't know how to get VRAM type yet. */
 		adev->mc.vram_type = AMDGPU_VRAM_TYPE_HBM;
+
+		/*
+		 * To fulfill 4-level page support,
+		 * vm size is 256TB (48bit), maximum size of Vega10,
+		 * block size 512 (9bit)
+		 */
+		amdgpu_vm_size = 1U << 18;
+		amdgpu_vm_block_size = 9;

This won't work. You are still changing the amdgpu_vm_* global variables so if you have 2 GPUs in the systems, you'll change them both.

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +0 -0
 	}
 
 	/* This interrupt is VMC page fault.*/
@@ -557,13 +565,6 @@ static int gmc_v9_0_sw_init(void *handle)
 	if (r)

drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c static int gmc_v9_0_sw_init(void *handle) · line 568

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +0 -7
 		return r;
 
-	/* Because of four level VMPTs, vm size is at least 512GB.
-	 * The maximum size is 256TB (48bit).
-	 */
-	if (amdgpu_vm_size < 512) {
-		DRM_WARN("VM size is at least 512GB!\n");
-		amdgpu_vm_size = 512;
-	}
 	adev->vm_manager.max_pfn = (uint64_t)amdgpu_vm_size << 18;

Instead of changing amdgpu_vm_size, change adev->vm_manager.max_pfn directly, that way you won't be changing the global variable and possibly messing up other driver instances. The alternative, and cleaner in my opinion, would be to add new variables to the vm_manager, e.g., adev->vm_manager.size and adev->vm_manager.block_size, and set them to something reasonable based on the module parameters and them use them in the code rather than the global variables.

Alex

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +0 -1
 	/* Set the internal MC address mask
--

1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): CY4PR12MB1653A907A71FB7C02DE39EE5F7370-rpdhrqHFk06apTa93KjAaQdYzm3356FpvxpqHgZTriW3zl9H0oFU5g@public.gmane.org

#5 lore raw
Quoted 65 lines · Alexander
Quoted 44 lines

-----Original Message-----
From: Junwei Zhang [mailto:[email protected]]
Sent: Thursday, March 30, 2017 10:52 PM
To: Deucher, Alexander; Koenig, Christian
Cc: [email protected]; Zhang, Jerry
Subject: [PATCH 1/2 v2] drm/amdgpu: fix vm size and block size for Vega10 VMPT

From: "Zhang, Jerry" <[email protected]>

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  8 --------
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c      | 15 ++++++++-------
 2 files changed, 8 insertions(+), 15 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0 -8
index 363d73c..98555fd 100644
@@ -1041,14 +1041,6 @@ static bool amdgpu_check_pot_argument(int arg)
 
 static void amdgpu_get_block_size(struct amdgpu_device *adev)
 {
-     /* from AI, asic starts to support multiple level VMPT */
-     if (adev->asic_type >= CHIP_VEGA10) {
-             if (amdgpu_vm_block_size != 9)
-                     dev_warn(adev->dev,
-                              "Multi-VMPT limits block size to one
 page!\n");
-             amdgpu_vm_block_size = 9;
-             return;
-     }
       /* defines number of bits in page table versus page directory,
        * a page is 4KB so we have 12 bits offset, minimum 9 bits in the

* page table and the remaining bits are in the page directory */

drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +0 -0
index df69aae..081a676 100644
@@ -546,6 +546,14 @@ static int gmc_v9_0_sw_init(void *handle)
       } else {

drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c static int gmc_v9_0_sw_init(void *handle) · line 556

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +8 -0
               /* XXX Don't know how to get VRAM type yet. */
               adev->mc.vram_type = AMDGPU_VRAM_TYPE_HBM;
+
+             /*
+              * To fulfill 4-level page support,
+              * vm size is 256TB (48bit), maximum size of Vega10,
+              * block size 512 (9bit)
+              */
+             amdgpu_vm_size = 1U << 18;
+             amdgpu_vm_block_size = 9;

This won't work. You are still changing the amdgpu_vm_* global variables so if you have 2 GPUs in the systems, you'll change them both.

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +0 -0
       }
 
       /* This interrupt is VMC page fault.*/
@@ -557,13 +565,6 @@ static int gmc_v9_0_sw_init(void *handle)
       if (r)
               return r;
 
Quoted 15 lines

drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c static int gmc_v9_0_sw_init(void *handle) · line 568

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +0 -7
-     /* Because of four level VMPTs, vm size is at least 512GB.
-      * The maximum size is 256TB (48bit).
-      */
-     if (amdgpu_vm_size < 512) {
-             DRM_WARN("VM size is at least 512GB!\n");
-             amdgpu_vm_size = 512;
-     }
       adev->vm_manager.max_pfn = (uint64_t)amdgpu_vm_size << 18;

Instead of changing amdgpu_vm_size, change adev->vm_manager.max_pfn directly, that way you won't be changing the global variable and possibly messing up other driver instances. The alternative, and cleaner in my opinion, would be to add new variables to the vm_manager, e.g., adev->vm_manager.size and adev->vm_manager.block_size, and set them to something reasonable based on the module parameters and them use them in the code rather than the global variables.

Yeah, thanks to point it out.
I prepared another patch, please check it too.

Jerry

Quoted 8 lines

Alex

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +0 -1
       /* Set the internal MC address mask
--

1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx