Skip to contentSkip to conversation

drm/amdgpu: add smu_memory_pool_size module parameter

PATCH In review Rex Zhu 3 participants 5 messages Oct 10, 2017 2 messages not in the archive

Subsystems (2)
lore
Share & tools
lore

Rex Zhu

author
#0 lore raw

PATCH 1/3 drm/amdgpu: add smu_memory_pool_size module parameter

used to clamp system memmoy for smu debug usage.

Change-Id I8399f55069b8dcc05a3a79f9285744fcc8549af0
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     | 1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 5 +++++
 2 files changed, 6 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -0
index da48f97..43fbcfe 100644
@@ -132,6 +132,7 @@
 #ifdef CONFIG_DRM_AMDGPU_CIK
 extern int amdgpu_cik_support;
 #endif
+extern uint amdgpu_smu_memory_pool_size;
 
 #define AMDGPU_DEFAULT_GTT_SIZE_MB		3072ULL /* 3GB by default */
 #define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS	        3000
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +5 -0
index ebcba56..d033904 100644
@@ -126,6 +126,7 @@
 int amdgpu_job_hang_limit = 0;
 int amdgpu_lbpw = -1;
 int amdgpu_compute_multipipe = -1;
+uint amdgpu_smu_memory_pool_size = 0;
 
 MODULE_PARM_DESC(vramlimit, "Restrict VRAM for testing, in megabytes");
 module_param_named(vramlimit, amdgpu_vram_limit, int, 0600);
@@ -301,6 +302,10 @@
 module_param_named(cik_support, amdgpu_cik_support, int, 0444);
 #endif
 
+MODULE_PARM_DESC(smu_memory_pool_size,
+	"clamp max system memory for smu debug usage, 0 = disable,"
+		"0x1 = 256Mbyte, 0x2 = 512Mbyte, 0x4 = 1 Gbyte, 0x8 = 2GByte");
+module_param_named(smu_memory_pool_size, amdgpu_smu_memory_pool_size, uint, 0444);
 
 static const struct pci_device_id pciidlist[] = {
 #ifdef  CONFIG_DRM_AMDGPU_SI
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): 1507631973-14165-1-git-send-email-Rex.Zhu-5C7GfCeVMHo@public.gmane.org

author
#2 lore raw

PATCH 2/3 drm/amdgpu: validate module parameter smu_prv_buffer_size

Change-Id I5f7025a357720c7553035c02519b2b6ac2c59358
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 43 ++++++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    |  3 +++
 drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c      |  2 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c      |  3 +++
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c      |  2 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c      |  2 ++
 6 files changed, 55 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +43 -0
index fdfe418..548f4dc 100644
@@ -1152,6 +1152,46 @@ static void amdgpu_check_vm_size(struct amdgpu_device *adev)
 	amdgpu_vm_size = -1;
 }
 
+static void amdgpu_device_check_smu_prv_buffer_size(struct amdgpu_device *adev)
+{
+	struct sysinfo si;
+	bool is_os_64 = (sizeof(void *) == 8) ? true : false;
+	uint64_t total_memory;
+	uint64_t dram_size_seven_GB = 0x1B8000000;
+	uint64_t dram_size_three_GB = 0xB8000000;
+
+	if (amdgpu_smu_memory_pool_size == 0)
+		return;
+
+	if (!is_os_64) {
+		DRM_WARN("Not 64-bit OS, feature not supported\n");
+		goto def_value;
+	}
+	si_meminfo(&si);
+	total_memory = (uint64_t)si.totalram * si.mem_unit;
+
+	if ((amdgpu_smu_memory_pool_size == 1) ||
+		(amdgpu_smu_memory_pool_size == 2)) {
+		if (total_memory < dram_size_three_GB)
+			goto def_value1;
+	} else if ((amdgpu_smu_memory_pool_size == 4) ||
+		(amdgpu_smu_memory_pool_size == 8)) {
+		if (total_memory < dram_size_seven_GB)
+			goto def_value1;
+	} else {
+		DRM_WARN("Smu memory pool size not supported\n");
+		goto def_value;
+	}
+	amdgpu_smu_memory_pool_size = amdgpu_smu_memory_pool_size << 28;
+
+	return;
+
+def_value1:
+	DRM_WARN("No enough system memory\n");
+def_value:
+	amdgpu_smu_memory_pool_size = 0;
+}
+
 /**
  * amdgpu_check_arguments - validate module params
  *
@@ -1193,6 +1233,8 @@ static void amdgpu_check_arguments(struct amdgpu_device *adev)
 		amdgpu_vm_fragment_size = -1;
 	}
 
+	amdgpu_device_check_smu_prv_buffer_size(adev);
+
 	amdgpu_check_vm_size(adev);
 
 	amdgpu_check_block_size(adev);
@@ -2430,6 +2472,7 @@ void amdgpu_device_fini(struct amdgpu_device *adev)
 		release_firmware(adev->firmware.gpu_info_fw);
 		adev->firmware.gpu_info_fw = NULL;
 	}
+
 	adev->accel_working = false;
 	cancel_delayed_work_sync(&adev->late_init_work);
 	/* free i2c buses */
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +3 -0
index fa2781b..9eab064 100644
@@ -1280,6 +1280,9 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 			       adev->mc.mc_vram_size);
 	else
 		gtt_size = (uint64_t)amdgpu_gtt_size << 20;
+
+	gtt_size = gtt_size + amdgpu_smu_memory_pool_size;
+
 	r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_TT, gtt_size >> PAGE_SHIFT);
 	if (r) {
 		DRM_ERROR("Failed initializing GTT heap.\n");
drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c +2 -0
index f4603a7..e1f97b1 100644
@@ -350,6 +350,8 @@ static int gmc_v6_0_mc_init(struct amdgpu_device *adev)
 		adev->mc.gart_size = (u64)amdgpu_gart_size << 20;
 	}
 
+	adev->mc.gart_size += amdgpu_smu_memory_pool_size;
+
 	gmc_v6_0_vram_gtt_location(adev, &adev->mc);
 
 	return 0;
drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +3 -0
index b0528ca..ac481a0 100644
@@ -407,6 +407,9 @@ static int gmc_v7_0_mc_init(struct amdgpu_device *adev)
 		adev->mc.gart_size = (u64)amdgpu_gart_size << 20;
 	}
 
+
+	adev->mc.gart_size += amdgpu_smu_memory_pool_size;
+
 	gmc_v7_0_vram_gtt_location(adev, &adev->mc);
 
 	return 0;
drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +2 -0
index f368cfe..0f5bccc 100644
@@ -582,6 +582,8 @@ static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
 		adev->mc.gart_size = (u64)amdgpu_gart_size << 20;
 	}
 
+	adev->mc.gart_size += amdgpu_smu_memory_pool_size;
+
 	gmc_v8_0_vram_gtt_location(adev, &adev->mc);
 
 	return 0;
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +2 -0
index 6216993..09cf381 100644
@@ -511,6 +511,8 @@ static int gmc_v9_0_mc_init(struct amdgpu_device *adev)
 		adev->mc.gart_size = (u64)amdgpu_gart_size << 20;
 	}
 
+	adev->mc.gart_size += amdgpu_smu_memory_pool_size;
+
 	gmc_v9_0_vram_gtt_location(adev, &adev->mc);
 
 	return 0;
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#3 lore raw

PATCH 3/3 drm/amd/powerplay: add new sysfs pp_alloc_mem_for_smu

In reply to a message not in the archive
Change-Id I3f689b8da288ba739aaf74340c056814f80713a9
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  6 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c     | 89 ++++++++++++++++++++++++++++++
 3 files changed, 96 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 43fbcfe..7139d3b 100644
@@ -1637,6 +1637,8 @@ struct amdgpu_device {
 	bool has_hw_reset;
 	u8				reset_magic[AMDGPU_RESET_MAGIC_NUM];
 
+	struct amdgpu_bo		*smu_prv_buffer;
+
 	/* record last mm index being written through WREG32*/
 	unsigned long last_mm_index;
 	bool                            in_sriov_reset;
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +5 -1
index 548f4dc..6f07a55 100644
@@ -2426,6 +2426,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 			DRM_INFO("amdgpu: acceleration disabled, skipping benchmarks\n");
 	}
 
+
 	/* enable clockgating, etc. after ib tests, etc. since some blocks require
 	 * explicit gating rather than handling it automatically.
 	 */
@@ -2472,7 +2473,10 @@ void amdgpu_device_fini(struct amdgpu_device *adev)
 		release_firmware(adev->firmware.gpu_info_fw);
 		adev->firmware.gpu_info_fw = NULL;
 	}
-
+	if (adev->smu_prv_buffer) {
+		amdgpu_bo_free_kernel(&adev->smu_prv_buffer, NULL, NULL);
+		adev->smu_prv_buffer = NULL;
+	}
 	adev->accel_working = false;
 	cancel_delayed_work_sync(&adev->late_init_work);
 	/* free i2c buses */
drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c +89 -0

107 lines not on this page · raw message

Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#4 lore raw
In reply to a message not in the archive
Quoted 34 lines · Rex Zhu

used to clamp system memmoy for smu debug usage.

Change-Id I8399f55069b8dcc05a3a79f9285744fcc8549af0
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     | 1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 5 +++++
 2 files changed, 6 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -0
index da48f97..43fbcfe 100644
@@ -132,6 +132,7 @@
 #ifdef CONFIG_DRM_AMDGPU_CIK
 extern int amdgpu_cik_support;
 #endif
+extern uint amdgpu_smu_memory_pool_size;
 
 #define AMDGPU_DEFAULT_GTT_SIZE_MB             3072ULL /* 3GB by default */
 #define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS         3000
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +1 -0
index ebcba56..d033904 100644
@@ -126,6 +126,7 @@
 int amdgpu_job_hang_limit = 0;
 int amdgpu_lbpw = -1;
 int amdgpu_compute_multipipe = -1;
+uint amdgpu_smu_memory_pool_size = 0;
 
 MODULE_PARM_DESC(vramlimit, "Restrict VRAM for testing, in megabytes");
 module_param_named(vramlimit, amdgpu_vram_limit, int, 0600);

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c line 308

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +4 -0
@@ -301,6 +302,10 @@
 module_param_named(cik_support, amdgpu_cik_support, int, 0444);
 #endif
 
+MODULE_PARM_DESC(smu_memory_pool_size,
+       "clamp max system memory for smu debug usage, 0 = disable,"
+               "0x1 = 256Mbyte, 0x2 = 512Mbyte, 0x4 = 1 Gbyte, 0x8 = 2GByte");
+module_param_named(smu_memory_pool_size, amdgpu_smu_memory_pool_size, uint, 0444);

Sorry, I should have been more clear. I want to avoid adding another module parameter. I was thinking of a debugfs file (e.g., amdgpu_smu_log_size or something like that) and if you set the file to 0, it would disable smu power logging and if you set it to a different size, it would allocate the memory and then send the smu messages to enable the logging. Within the debugfs handler it would check the size against the current gart size and return an error (-ENOMEM) if the gart wasn't big enough or the allocation failed.

to disable:
echo 0 > /sys/kernel/debug/dri/0/amdgpu_smu_log_size to enable a 64M SMU log:
echo 64 > /sys/kernel/debug/dri/0/amdgpu_smu_log_size

Alex

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +0 -1
 static const struct pci_device_id pciidlist[] = {
 #ifdef  CONFIG_DRM_AMDGPU_SI
--

1.9.1

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

_______________________________________________
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): CADnq5_P4P_A4wNchSb5qk=65c-qJtnDQS+H1yZXzL4gv+qajFw-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org

#6 lore raw

Am 10.10.2017 um 18:33 schrieb Alex Deucher:

Quoted 51 lines · Rex Zhu
Quoted 34 lines

used to clamp system memmoy for smu debug usage.

Change-Id I8399f55069b8dcc05a3a79f9285744fcc8549af0
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h     | 1 +
  drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 5 +++++
  2 files changed, 6 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -0
index da48f97..43fbcfe 100644
@@ -132,6 +132,7 @@
  #ifdef CONFIG_DRM_AMDGPU_CIK
  extern int amdgpu_cik_support;
  #endif
+extern uint amdgpu_smu_memory_pool_size;
 
  #define AMDGPU_DEFAULT_GTT_SIZE_MB             3072ULL /* 3GB by default */
  #define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS         3000
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +1 -0
index ebcba56..d033904 100644
@@ -126,6 +126,7 @@
  int amdgpu_job_hang_limit = 0;
  int amdgpu_lbpw = -1;
  int amdgpu_compute_multipipe = -1;
+uint amdgpu_smu_memory_pool_size = 0;
 
  MODULE_PARM_DESC(vramlimit, "Restrict VRAM for testing, in megabytes");
  module_param_named(vramlimit, amdgpu_vram_limit, int, 0600);

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c line 308

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +4 -0
@@ -301,6 +302,10 @@
  module_param_named(cik_support, amdgpu_cik_support, int, 0444);
  #endif
 
+MODULE_PARM_DESC(smu_memory_pool_size,
+       "clamp max system memory for smu debug usage, 0 = disable,"
+               "0x1 = 256Mbyte, 0x2 = 512Mbyte, 0x4 = 1 Gbyte, 0x8 = 2GByte");
+module_param_named(smu_memory_pool_size, amdgpu_smu_memory_pool_size, uint, 0444);

Sorry, I should have been more clear. I want to avoid adding another module parameter. I was thinking of a debugfs file (e.g., amdgpu_smu_log_size or something like that) and if you set the file to 0, it would disable smu power logging and if you set it to a different size, it would allocate the memory and then send the smu messages to enable the logging. Within the debugfs handler it would check the size against the current gart size and return an error (-ENOMEM) if the gart wasn't big enough or the allocation failed.

Quoted 4 lines

to disable:
echo 0 > /sys/kernel/debug/dri/0/amdgpu_smu_log_size to enable a 64M SMU log:
echo 64 > /sys/kernel/debug/dri/0/amdgpu_smu_log_size

Seconded, assuming that you can send the message to the SMU while it is running that sounds like a much better approach.

You also don't need to really check the size, allocating and pinning a buffer which is bigger than GART should fail no matter what.

Christian.

Quoted 8 lines

Alex

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +0 -1
  static const struct pci_device_id pciidlist[] = {
  #ifdef  CONFIG_DRM_AMDGPU_SI
--

1.9.1

Quoted 8 lines
Quoted 4 lines

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

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

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