Skip to contentSkip to conversation

drm/amdkfd: Move the ignore_crat check before the CRAT table get

PATCH Superseded Hanjun Guo 2 participants 4 messages Aug 26, 2020 – Aug 27, 2020

Subsystems (3)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Hanjun Guo

author
#0 lore raw

PATCH v2 1/2 drm/amdkfd: Move the ignore_crat check before the CRAT table get

If the ignore_crat is set to non-zero value, it's no point getting the CRAT table, so just move the ignore_crat check before we get the CRAT table.

---
 drivers/gpu/drm/amd/amdkfd/kfd_crat.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_crat.c +5 -5
index 6a250f8..ed77b109 100644
@@ -764,6 +764,11 @@ int kfd_create_crat_image_acpi(void **crat_image, size_t *size)
 
 	*crat_image = NULL;
 
+	if (ignore_crat) {
+		pr_info("CRAT table disabled by module option\n");
+		return -ENODATA;
+	}
+
 	/* Fetch the CRAT table from ACPI */
 	status = acpi_get_table(CRAT_SIGNATURE, 0, &crat_table);
 	if (status == AE_NOT_FOUND) {
@@ -776,11 +781,6 @@ int kfd_create_crat_image_acpi(void **crat_image, size_t *size)
 		return -EINVAL;
 	}
 
-	if (ignore_crat) {
-		pr_info("CRAT table disabled by module option\n");
-		return -ENODATA;
-	}
-
 	pcrat_image = kmemdup(crat_table, crat_table->length, GFP_KERNEL);
 	if (!pcrat_image)
 		return -ENOMEM;
Signature · 1 line
1.7.12.4
author
#1 lore raw

PATCH v2 2/2 drm/amdkfd: Put ACPI table after using it

The acpi_get_table() should be coupled with acpi_put_table() if the mapped table is not used at runtime to release the table mapping which can prevent the memory leak.

In kfd_create_crat_image_acpi(), crat_table is copied to pcrat_image, and in kfd_create_vcrat_image_cpu(), the acpi_table is only used to get the OEM information, so those two table mappings need to be released after using it.

Fixes 174de876d6d0 ("drm/amdkfd: Group up CRAT related functions")
Fixes 520b8fb755cc ("drm/amdkfd: Add topology support for CPUs")
---
 drivers/gpu/drm/amd/amdkfd/kfd_crat.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_crat.c +9 -4
index ed77b109..c3f3ffb6 100644
@@ -758,6 +758,7 @@ int kfd_create_crat_image_acpi(void **crat_image, size_t *size)
 	struct acpi_table_header *crat_table;
 	acpi_status status;
 	void *pcrat_image;
+	int rc = 0;
 
 	if (!crat_image)
 		return -EINVAL;
@@ -782,13 +783,16 @@ int kfd_create_crat_image_acpi(void **crat_image, size_t *size)
 	}
 
 	pcrat_image = kmemdup(crat_table, crat_table->length, GFP_KERNEL);
-	if (!pcrat_image)
-		return -ENOMEM;
+	if (!pcrat_image) {
+		rc = -ENOMEM;
+		goto out;
+	}
 
 	*crat_image = pcrat_image;
 	*size = crat_table->length;
-
-	return 0;
+out:
+	acpi_put_table(crat_table);
+	return rc;
 }
 
 /* Memory required to create Virtual CRAT.
@@ -972,6 +976,7 @@ static int kfd_create_vcrat_image_cpu(void *pcrat_image, size_t *size)
 				CRAT_OEMID_LENGTH);
 		memcpy(crat_table->oem_table_id, acpi_table->oem_table_id,
 				CRAT_OEMTABLEID_LENGTH);
+		acpi_put_table(acpi_table);
 	}
 	crat_table->total_entries = 0;
 	crat_table->num_domains = 0;
Signature · 1 line
1.7.12.4
maintainer
#2 lore raw
In reply to Hanjun Guo

Am 2020-08-26 um 4:29 a.m. schrieb Hanjun Guo:

Quoted 14 lines

If the ignore_crat is set to non-zero value, it's no point getting the CRAT table, so just move the ignore_crat check before we get the CRAT table.

---
 drivers/gpu/drm/amd/amdkfd/kfd_crat.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_crat.c +0 -0
index 6a250f8..ed77b109 100644

drivers/gpu/drm/amd/amdkfd/kfd_crat.c int kfd_create_crat_image_acpi(void **crat_image, size_t *size) · line 767

Quoted 4 lines
drivers/gpu/drm/amd/amdkfd/kfd_crat.c +1 -0
@@ -764,6 +764,11 @@ int kfd_create_crat_image_acpi(void **crat_image, size_t *size)
 
 	*crat_image = NULL;
 
+	if (ignore_crat) {

A conflicting change in this area was just submitted on Monday to amd-staging-drm-next. You'll need to rebase your patch. It should be straight-forward. ignore_crat was replaced with a function call kfd_ignore_crat().

Other than that, your patch series looks good to me. If I don't see an update from you in a day or two, I'll fix it up myself and add my R-b.

Thanks,
  Felix

Quoted 19 lines
drivers/gpu/drm/amd/amdkfd/kfd_crat.c +4 -5
+		pr_info("CRAT table disabled by module option\n");
+		return -ENODATA;
+	}
+
 	/* Fetch the CRAT table from ACPI */
 	status = acpi_get_table(CRAT_SIGNATURE, 0, &crat_table);
 	if (status == AE_NOT_FOUND) {
@@ -776,11 +781,6 @@ int kfd_create_crat_image_acpi(void **crat_image, size_t *size)
 		return -EINVAL;
 	}
 
-	if (ignore_crat) {
-		pr_info("CRAT table disabled by module option\n");
-		return -ENODATA;
-	}
-
 	pcrat_image = kmemdup(crat_table, crat_table->length, GFP_KERNEL);
 	if (!pcrat_image)
 		return -ENOMEM;
author
#3 lore raw
Quoted 26 lines · Felix Kuehling

Am 2020-08-26 um 4:29 a.m. schrieb Hanjun Guo:

Quoted 14 lines

If the ignore_crat is set to non-zero value, it's no point getting the CRAT table, so just move the ignore_crat check before we get the CRAT table.

---
  drivers/gpu/drm/amd/amdkfd/kfd_crat.c | 10 +++++-----
  1 file changed, 5 insertions(+), 5 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_crat.c +0 -0
index 6a250f8..ed77b109 100644

drivers/gpu/drm/amd/amdkfd/kfd_crat.c int kfd_create_crat_image_acpi(void **crat_image, size_t *size) · line 767

Quoted 4 lines
drivers/gpu/drm/amd/amdkfd/kfd_crat.c +1 -0
@@ -764,6 +764,11 @@ int kfd_create_crat_image_acpi(void **crat_image, size_t *size)
 
  	*crat_image = NULL;
 
+	if (ignore_crat) {

A conflicting change in this area was just submitted on Monday to amd-staging-drm-next. You'll need to rebase your patch. It should be straight-forward. ignore_crat was replaced with a function call kfd_ignore_crat().

Quoted 2 lines

Other than that, your patch series looks good to me. If I don't see an update from you in a day or two, I'll fix it up myself and add my R-b.

Please fix it up by yourself, thanks a lot for the help!

Regards
Hanjun