Skip to contentSkip to conversation

platform/x86/amd: pmf: Use device managed allocations

PATCH Superseded Mario Limonciello 3 participants 5 messages May 7, 2025 – May 9, 2025

Subsystems (2)

v2 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 platform/x86/amd: pmf: Use device managed allocations

From: Mario Limonciello <[email protected]>

If setting up smart PC fails for any reason then this can lead to a double free when unloading amd-pmf. This is because dev->buf was freed but never set to NULL and is again freed in amd_pmf_remove().

To avoid subtle allocation bugs in failures leading to a double free change all allocations into device managed allocations.

Fixes 5b1122fc4995f ("platform/x86/amd/pmf: fix cleanup in amd_pmf_init_smart_pc()")

---
v1->v2:
* Correct commit message with correct offending function root cause
* Switch to device managed allocations.

 drivers/platform/x86/amd/pmf/core.c   |  3 +--
 drivers/platform/x86/amd/pmf/tee-if.c | 30 ++++++++-------------------
 2 files changed, 10 insertions(+), 23 deletions(-)
drivers/platform/x86/amd/pmf/core.c +1 -2
index 96821101ec773..395c011e837f1 100644
@@ -280,7 +280,7 @@ int amd_pmf_set_dram_addr(struct amd_pmf_dev *dev, bool alloc_buffer)
 			dev_err(dev->dev, "Invalid CPU id: 0x%x", dev->cpu_id);
 		}
 
-		dev->buf = kzalloc(dev->mtable_size, GFP_KERNEL);
+		dev->buf = devm_kzalloc(dev->dev, dev->mtable_size, GFP_KERNEL);
 		if (!dev->buf)
 			return -ENOMEM;
 	}
@@ -493,7 +493,6 @@ static void amd_pmf_remove(struct platform_device *pdev)
 	mutex_destroy(&dev->lock);
 	mutex_destroy(&dev->update_mutex);
 	mutex_destroy(&dev->cb_mutex);
-	kfree(dev->buf);
 }
 
 static const struct attribute_group *amd_pmf_driver_groups[] = {
drivers/platform/x86/amd/pmf/tee-if.c +9 -21
index d3bd12ad036ae..50c082a78cd9e 100644
@@ -532,13 +532,13 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
 	dev->policy_base = devm_ioremap_resource(dev->dev, dev->res);
 	if (IS_ERR(dev->policy_base)) {
 		ret = PTR_ERR(dev->policy_base);
-		goto err_free_dram_buf;
+		goto err_cancel_work;
 	}
 
-	dev->policy_buf = kzalloc(dev->policy_sz, GFP_KERNEL);
+	dev->policy_buf = devm_kzalloc(dev->dev, dev->policy_sz, GFP_KERNEL);
 	if (!dev->policy_buf) {
 		ret = -ENOMEM;
-		goto err_free_dram_buf;
+		goto err_cancel_work;
 	}
 
 	memcpy_fromio(dev->policy_buf, dev->policy_base, dev->policy_sz);
@@ -546,21 +546,21 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
 	if (!amd_pmf_pb_valid(dev)) {
 		dev_info(dev->dev, "No Smart PC policy present\n");
 		ret = -EINVAL;
-		goto err_free_policy;
+		goto err_cancel_work;
 	}
 
 	amd_pmf_hex_dump_pb(dev);
 
-	dev->prev_data = kzalloc(sizeof(*dev->prev_data), GFP_KERNEL);
+	dev->prev_data = devm_kzalloc(dev->dev, sizeof(*dev->prev_data), GFP_KERNEL);
 	if (!dev->prev_data) {
 		ret = -ENOMEM;
-		goto err_free_policy;
+		goto err_cancel_work;
 	}
 
 	for (i = 0; i < ARRAY_SIZE(amd_pmf_ta_uuid); i++) {
 		ret = amd_pmf_tee_init(dev, &amd_pmf_ta_uuid[i]);
 		if (ret)
-			goto err_free_prev_data;
+			goto err_cancel_work;
 
 		ret = amd_pmf_start_policy_engine(dev);
 		switch (ret) {
@@ -575,7 +575,7 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
 		default:
 			ret = -EINVAL;
 			amd_pmf_tee_deinit(dev);
-			goto err_free_prev_data;
+			goto err_cancel_work;
 		}
 
 		if (status)
@@ -584,7 +584,7 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
 
 	if (!status && !pb_side_load) {
 		ret = -EINVAL;
-		goto err_free_prev_data;
+		goto err_cancel_work;
 	}
 
 	if (pb_side_load)
@@ -600,12 +600,6 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
 	if (pb_side_load && dev->esbin)
 		amd_pmf_remove_pb(dev);
 	amd_pmf_tee_deinit(dev);
-err_free_prev_data:
-	kfree(dev->prev_data);
-err_free_policy:
-	kfree(dev->policy_buf);
-err_free_dram_buf:
-	kfree(dev->buf);
 err_cancel_work:
 	cancel_delayed_work_sync(&dev->pb_work);
 
@@ -621,11 +615,5 @@ void amd_pmf_deinit_smart_pc(struct amd_pmf_dev *dev)
 		amd_pmf_remove_pb(dev);
 
 	cancel_delayed_work_sync(&dev->pb_work);
-	kfree(dev->prev_data);
-	dev->prev_data = NULL;
-	kfree(dev->policy_buf);
-	dev->policy_buf = NULL;
-	kfree(dev->buf);
-	dev->buf = NULL;
 	amd_pmf_tee_deinit(dev);
 }
Signature · 1 line
2.43.0
#1 lore raw
Quoted 9 lines · Mario Limonciello

From: Mario Limonciello <[email protected]>

If setting up smart PC fails for any reason then this can lead to a double free when unloading amd-pmf. This is because dev->buf was freed but never set to NULL and is again freed in amd_pmf_remove().

To avoid subtle allocation bugs in failures leading to a double free change all allocations into device managed allocations.

Quoted 6 lines
Fixes 5b1122fc4995f ("platform/x86/amd/pmf: fix cleanup in amd_pmf_init_smart_pc()")

---
v1->v2:
* Correct commit message with correct offending function root cause
* Switch to device managed allocations.

Yeah, this the best solution. But we still could end up calling amd_pmf_tee_deinit() twice if amd_pmf_register_input_device() fails so there is still a bug.

regards,
dan carpenter

maintainer
#2 lore raw
In reply to Mario Limonciello
Quoted 46 lines · Mario Limonciello

From: Mario Limonciello <[email protected]>

If setting up smart PC fails for any reason then this can lead to a double free when unloading amd-pmf. This is because dev->buf was freed but never set to NULL and is again freed in amd_pmf_remove().

To avoid subtle allocation bugs in failures leading to a double free change all allocations into device managed allocations.

Fixes 5b1122fc4995f ("platform/x86/amd/pmf: fix cleanup in amd_pmf_init_smart_pc()")

---
v1->v2:
* Correct commit message with correct offending function root cause
* Switch to device managed allocations.

 drivers/platform/x86/amd/pmf/core.c   |  3 +--
 drivers/platform/x86/amd/pmf/tee-if.c | 30 ++++++++-------------------
 2 files changed, 10 insertions(+), 23 deletions(-)
drivers/platform/x86/amd/pmf/core.c +1 -2
index 96821101ec773..395c011e837f1 100644
@@ -280,7 +280,7 @@ int amd_pmf_set_dram_addr(struct amd_pmf_dev *dev, bool alloc_buffer)
 			dev_err(dev->dev, "Invalid CPU id: 0x%x", dev->cpu_id);
 		}
 
-		dev->buf = kzalloc(dev->mtable_size, GFP_KERNEL);
+		dev->buf = devm_kzalloc(dev->dev, dev->mtable_size, GFP_KERNEL);
 		if (!dev->buf)
 			return -ENOMEM;
 	}
@@ -493,7 +493,6 @@ static void amd_pmf_remove(struct platform_device *pdev)
 	mutex_destroy(&dev->lock);
 	mutex_destroy(&dev->update_mutex);
 	mutex_destroy(&dev->cb_mutex);
-	kfree(dev->buf);
 }
 
 static const struct attribute_group *amd_pmf_driver_groups[] = {
drivers/platform/x86/amd/pmf/tee-if.c +0 -0
index d3bd12ad036ae..50c082a78cd9e 100644

drivers/platform/x86/amd/pmf/tee-if.c int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev) · line 538

Quoted 9 lines
drivers/platform/x86/amd/pmf/tee-if.c +2 -2
@@ -532,13 +532,13 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
 	dev->policy_base = devm_ioremap_resource(dev->dev, dev->res);
 	if (IS_ERR(dev->policy_base)) {
 		ret = PTR_ERR(dev->policy_base);
-		goto err_free_dram_buf;
+		goto err_cancel_work;
 	}
 
-	dev->policy_buf = kzalloc(dev->policy_sz, GFP_KERNEL);
+	dev->policy_buf = devm_kzalloc(dev->dev, dev->policy_sz, GFP_KERNEL);

Hi Mario,

Isn't ->policy_buf freed and another allocated in amd_pmf_get_pb_data() and this patch lacks any changes around there?? That switch of the buffer has been the reason why I've not suggested using devm_*() for earlier for it.

Please check all related code to the pointers you're changing if there are other similar traps you have not taken into account.

Quoted 77 lines
drivers/platform/x86/amd/pmf/tee-if.c +7 -19
 	if (!dev->policy_buf) {
 		ret = -ENOMEM;
-		goto err_free_dram_buf;
+		goto err_cancel_work;
 	}
 
 	memcpy_fromio(dev->policy_buf, dev->policy_base, dev->policy_sz);
@@ -546,21 +546,21 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
 	if (!amd_pmf_pb_valid(dev)) {
 		dev_info(dev->dev, "No Smart PC policy present\n");
 		ret = -EINVAL;
-		goto err_free_policy;
+		goto err_cancel_work;
 	}
 
 	amd_pmf_hex_dump_pb(dev);
 
-	dev->prev_data = kzalloc(sizeof(*dev->prev_data), GFP_KERNEL);
+	dev->prev_data = devm_kzalloc(dev->dev, sizeof(*dev->prev_data), GFP_KERNEL);
 	if (!dev->prev_data) {
 		ret = -ENOMEM;
-		goto err_free_policy;
+		goto err_cancel_work;
 	}
 
 	for (i = 0; i < ARRAY_SIZE(amd_pmf_ta_uuid); i++) {
 		ret = amd_pmf_tee_init(dev, &amd_pmf_ta_uuid[i]);
 		if (ret)
-			goto err_free_prev_data;
+			goto err_cancel_work;
 
 		ret = amd_pmf_start_policy_engine(dev);
 		switch (ret) {
@@ -575,7 +575,7 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
 		default:
 			ret = -EINVAL;
 			amd_pmf_tee_deinit(dev);
-			goto err_free_prev_data;
+			goto err_cancel_work;
 		}
 
 		if (status)
@@ -584,7 +584,7 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
 
 	if (!status && !pb_side_load) {
 		ret = -EINVAL;
-		goto err_free_prev_data;
+		goto err_cancel_work;
 	}
 
 	if (pb_side_load)
@@ -600,12 +600,6 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
 	if (pb_side_load && dev->esbin)
 		amd_pmf_remove_pb(dev);
 	amd_pmf_tee_deinit(dev);
-err_free_prev_data:
-	kfree(dev->prev_data);
-err_free_policy:
-	kfree(dev->policy_buf);
-err_free_dram_buf:
-	kfree(dev->buf);
 err_cancel_work:
 	cancel_delayed_work_sync(&dev->pb_work);
 
@@ -621,11 +615,5 @@ void amd_pmf_deinit_smart_pc(struct amd_pmf_dev *dev)
 		amd_pmf_remove_pb(dev);
 
 	cancel_delayed_work_sync(&dev->pb_work);
-	kfree(dev->prev_data);
-	dev->prev_data = NULL;
-	kfree(dev->policy_buf);
-	dev->policy_buf = NULL;
-	kfree(dev->buf);
-	dev->buf = NULL;
 	amd_pmf_tee_deinit(dev);
 }
Signature · 1 line
 i.
#3 lore raw
Quoted 60 lines · Ilpo Järvinen
Quoted 46 lines · Mario Limonciello

From: Mario Limonciello <[email protected]>

If setting up smart PC fails for any reason then this can lead to a double free when unloading amd-pmf. This is because dev->buf was freed but never set to NULL and is again freed in amd_pmf_remove().

To avoid subtle allocation bugs in failures leading to a double free change all allocations into device managed allocations.

Fixes 5b1122fc4995f ("platform/x86/amd/pmf: fix cleanup in amd_pmf_init_smart_pc()")

---
v1->v2:
* Correct commit message with correct offending function root cause
* Switch to device managed allocations.

  drivers/platform/x86/amd/pmf/core.c   |  3 +--
  drivers/platform/x86/amd/pmf/tee-if.c | 30 ++++++++-------------------
  2 files changed, 10 insertions(+), 23 deletions(-)
drivers/platform/x86/amd/pmf/core.c +1 -2
index 96821101ec773..395c011e837f1 100644
@@ -280,7 +280,7 @@ int amd_pmf_set_dram_addr(struct amd_pmf_dev *dev, bool alloc_buffer)
  			dev_err(dev->dev, "Invalid CPU id: 0x%x", dev->cpu_id);
  		}
 
-		dev->buf = kzalloc(dev->mtable_size, GFP_KERNEL);
+		dev->buf = devm_kzalloc(dev->dev, dev->mtable_size, GFP_KERNEL);
  		if (!dev->buf)
  			return -ENOMEM;
  	}
@@ -493,7 +493,6 @@ static void amd_pmf_remove(struct platform_device *pdev)
  	mutex_destroy(&dev->lock);
  	mutex_destroy(&dev->update_mutex);
  	mutex_destroy(&dev->cb_mutex);
-	kfree(dev->buf);
  }
 
  static const struct attribute_group *amd_pmf_driver_groups[] = {
drivers/platform/x86/amd/pmf/tee-if.c +0 -0
index d3bd12ad036ae..50c082a78cd9e 100644

drivers/platform/x86/amd/pmf/tee-if.c int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev) · line 538

Quoted 9 lines
drivers/platform/x86/amd/pmf/tee-if.c +2 -2
@@ -532,13 +532,13 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
  	dev->policy_base = devm_ioremap_resource(dev->dev, dev->res);
  	if (IS_ERR(dev->policy_base)) {
  		ret = PTR_ERR(dev->policy_base);
-		goto err_free_dram_buf;
+		goto err_cancel_work;
  	}
 
-	dev->policy_buf = kzalloc(dev->policy_sz, GFP_KERNEL);
+	dev->policy_buf = devm_kzalloc(dev->dev, dev->policy_sz, GFP_KERNEL);

Hi Mario,

Quoted 7 lines

Isn't ->policy_buf freed and another allocated in amd_pmf_get_pb_data() and this patch lacks any changes around there?? That switch of the buffer has been the reason why I've not suggested using devm_*() for earlier for it.

Please check all related code to the pointers you're changing if there are other similar traps you have not taken into account.

Ah, thanks I thought I caught them all but missed that instance.

A few options that come to mind:
1) instead make the very first allocation POLICY_BUF_MAX_SZ, and then never reallocate for the life of the driver.
2) Don't allow sideloading a bigger policy than the original one in the firmware.
3) Don't switch all 3 to device managed like this patch, just switch the two that can and fix the one broken one causing a double free.
4) Switch the case you pointed out to use devm_kfree() and re-allocate at that time.

Anyone with a preference? I lean upon option 4.

Quoted 79 lines
Quoted 78 lines
drivers/platform/x86/amd/pmf/tee-if.c +7 -19
  	if (!dev->policy_buf) {
  		ret = -ENOMEM;
-		goto err_free_dram_buf;
+		goto err_cancel_work;
  	}
 
  	memcpy_fromio(dev->policy_buf, dev->policy_base, dev->policy_sz);
@@ -546,21 +546,21 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
  	if (!amd_pmf_pb_valid(dev)) {
  		dev_info(dev->dev, "No Smart PC policy present\n");
  		ret = -EINVAL;
-		goto err_free_policy;
+		goto err_cancel_work;
  	}
 
  	amd_pmf_hex_dump_pb(dev);
 
-	dev->prev_data = kzalloc(sizeof(*dev->prev_data), GFP_KERNEL);
+	dev->prev_data = devm_kzalloc(dev->dev, sizeof(*dev->prev_data), GFP_KERNEL);
  	if (!dev->prev_data) {
  		ret = -ENOMEM;
-		goto err_free_policy;
+		goto err_cancel_work;
  	}
 
  	for (i = 0; i < ARRAY_SIZE(amd_pmf_ta_uuid); i++) {
  		ret = amd_pmf_tee_init(dev, &amd_pmf_ta_uuid[i]);
  		if (ret)
-			goto err_free_prev_data;
+			goto err_cancel_work;
 
  		ret = amd_pmf_start_policy_engine(dev);
  		switch (ret) {
@@ -575,7 +575,7 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
  		default:
  			ret = -EINVAL;
  			amd_pmf_tee_deinit(dev);
-			goto err_free_prev_data;
+			goto err_cancel_work;
  		}
 
  		if (status)
@@ -584,7 +584,7 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
 
  	if (!status && !pb_side_load) {
  		ret = -EINVAL;
-		goto err_free_prev_data;
+		goto err_cancel_work;
  	}
 
  	if (pb_side_load)
@@ -600,12 +600,6 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
  	if (pb_side_load && dev->esbin)
  		amd_pmf_remove_pb(dev);
  	amd_pmf_tee_deinit(dev);
-err_free_prev_data:
-	kfree(dev->prev_data);
-err_free_policy:
-	kfree(dev->policy_buf);
-err_free_dram_buf:
-	kfree(dev->buf);
  err_cancel_work:
  	cancel_delayed_work_sync(&dev->pb_work);
 
@@ -621,11 +615,5 @@ void amd_pmf_deinit_smart_pc(struct amd_pmf_dev *dev)
  		amd_pmf_remove_pb(dev);
 
  	cancel_delayed_work_sync(&dev->pb_work);
-	kfree(dev->prev_data);
-	dev->prev_data = NULL;
-	kfree(dev->policy_buf);
-	dev->policy_buf = NULL;
-	kfree(dev->buf);
-	dev->buf = NULL;
  	amd_pmf_tee_deinit(dev);
  }
#4 lore raw
Quoted 78 lines · Mario Limonciello
Quoted 57 lines · Ilpo Järvinen
Quoted 44 lines · Mario Limonciello

From: Mario Limonciello <[email protected]>

If setting up smart PC fails for any reason then this can lead to a double free when unloading amd-pmf. This is because dev->buf was freed but never set to NULL and is again freed in amd_pmf_remove().

To avoid subtle allocation bugs in failures leading to a double free change all allocations into device managed allocations.

Fixes 5b1122fc4995f ("platform/x86/amd/pmf: fix cleanup in amd_pmf_init_smart_pc()")

---
v1->v2:
* Correct commit message with correct offending function root cause
* Switch to device managed allocations.

  drivers/platform/x86/amd/pmf/core.c   |  3 +--
  drivers/platform/x86/amd/pmf/tee-if.c | 30 ++++++++-------------------
  2 files changed, 10 insertions(+), 23 deletions(-)
drivers/platform/x86/amd/pmf/core.c +1 -2
index 96821101ec773..395c011e837f1 100644
@@ -280,7 +280,7 @@ int amd_pmf_set_dram_addr(struct amd_pmf_dev *dev, bool alloc_buffer)
  			dev_err(dev->dev, "Invalid CPU id: 0x%x", dev->cpu_id);
  		}
-		dev->buf = kzalloc(dev->mtable_size, GFP_KERNEL);
+		dev->buf = devm_kzalloc(dev->dev, dev->mtable_size, GFP_KERNEL);
  		if (!dev->buf)
  			return -ENOMEM;
  	}
@@ -493,7 +493,6 @@ static void amd_pmf_remove(struct platform_device *pdev)
  	mutex_destroy(&dev->lock);
  	mutex_destroy(&dev->update_mutex);
  	mutex_destroy(&dev->cb_mutex);
-	kfree(dev->buf);
  }
  static const struct attribute_group *amd_pmf_driver_groups[] = {
drivers/platform/x86/amd/pmf/tee-if.c +0 -0
index d3bd12ad036ae..50c082a78cd9e 100644

drivers/platform/x86/amd/pmf/tee-if.c int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev) · line 537

Quoted 8 lines
drivers/platform/x86/amd/pmf/tee-if.c +2 -2
@@ -532,13 +532,13 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
  	dev->policy_base = devm_ioremap_resource(dev->dev, dev->res);
  	if (IS_ERR(dev->policy_base)) {
  		ret = PTR_ERR(dev->policy_base);
-		goto err_free_dram_buf;
+		goto err_cancel_work;
  	}
-	dev->policy_buf = kzalloc(dev->policy_sz, GFP_KERNEL);
+	dev->policy_buf = devm_kzalloc(dev->dev, dev->policy_sz, GFP_KERNEL);

Hi Mario,

Quoted 7 lines

Isn't ->policy_buf freed and another allocated in amd_pmf_get_pb_data() and this patch lacks any changes around there?? That switch of the buffer has been the reason why I've not suggested using devm_*() for earlier for it.

Please check all related code to the pointers you're changing if there are other similar traps you have not taken into account.

Ah, thanks I thought I caught them all but missed that instance.

A few options that come to mind:
1) instead make the very first allocation POLICY_BUF_MAX_SZ, and then never reallocate for the life of the driver.
2) Don't allow sideloading a bigger policy than the original one in the firmware.
3) Don't switch all 3 to device managed like this patch, just switch the two that can and fix the one broken one causing a double free.
4) Switch the case you pointed out to use devm_kfree() and re-allocate at that time.

Quoted 2 lines

Anyone with a preference? I lean upon option 4.

I feel like Option 3 is the worst...

I didn't have a problem with your v1 patch if we had fix the amd_pmf_tee_deinit() bug as well and probably deleted the second kfree().

I think there is a race with the debugfs sideloader, but it's debugfs and root only so if you hit that race, then you deserve it.

regards,
dan carpenter