Skip to contentSkip to conversation

ACPI: PCI: Clear driver_data on all paths that free the acpi_pci_root

PATCH New Chen Pei 2 participants 2 messages Jul 15

Subsystems (2)
lore
Share & tools
lore

Chen Pei

author
#0 lore raw

PATCH v2 1/1 ACPI: PCI: Clear driver_data on all paths that free the acpi_pci_root

acpi_pci_root_add() assigns the freshly allocated root to device->driver_data before dmar_device_add() and pci_acpi_scan_root(). Both failure paths reach the end: label where root is kfree()'d, but only the pci_acpi_scan_root() path clears driver_data first.

When dmar_device_add() fails during a hot-add, root is freed while device->driver_data still points at it. The ACPI core does not clear driver_data on attach failure, so a later acpi_pci_find_root() call may dereference this dangling pointer.

acpi_pci_root_remove() has the same problem: it frees root without clearing device->driver_data, leaving a dangling pointer behind after the root bridge is removed.

Move the NULL assignment to the shared end: label so every error path in acpi_pci_root_add() clears driver_data before freeing root, and clear it in acpi_pci_root_remove() as well, so the object is never left reachable through driver_data after being freed.

Fixes db89b4f0dbab ("ACPI: catch calls of acpi_driver_data on pointer of wrong type")

---
Changes in v2:
- Also clear device->driver_data in acpi_pci_root_remove() before kfree(root), so the remove path no longer leaves a dangling pointer (Sashiko AI review).

 drivers/acpi/pci_root.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/acpi/pci_root.c +2 -1
index 4c06c3ffd0cb..408ba12362a7 100644
@@ -725,7 +725,6 @@ static int acpi_pci_root_add(struct acpi_device *device,
 		dev_err(&device->dev,
 			"Bus %04x:%02x not present in PCI namespace\n",
 			root->segment, (unsigned int)root->secondary.start);
-		device->driver_data = NULL;
 		result = -ENODEV;
 		goto remove_dmar;
 	}
@@ -765,6 +764,7 @@ static int acpi_pci_root_add(struct acpi_device *device,
 	if (hotadd)
 		dmar_device_remove(handle);
 end:
+	device->driver_data = NULL;
 	kfree(root);
 	return result;
 }
@@ -788,6 +788,7 @@ static void acpi_pci_root_remove(struct acpi_device *device)
 
 	pci_unlock_rescan_remove();
 
+	device->driver_data = NULL;
 	kfree(root);
 }
 
Signature · 1 line
2.50.1
#1 lore raw
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] Clearing device->driver_data = NULL followed immediately by kfree(root) leaves a Use-After-Free race window for concurrent lockless readers.
- [High] Publishing root to device->driver_data before full initialization and without memory barriers allows readers to see uninitialized fields. --

--- Patch [1]: [PATCH v2] ACPI: PCI: Clear driver_data on all paths that free the acpi_pci_root --- commit df0768c14918e7decb436f5ec80561fb6dbd2aed
Author: Chen Pei <[email protected]>

ACPI: PCI: Clear driver_data on all paths that free the acpi_pci_root

This commit fixes dangling pointers in the ACPI PCI root code by ensuring device->driver_data is cleared to NULL before the root bridge structure is freed in both error and removal paths.

drivers/acpi/pci_root.c static int acpi_pci_root_add(struct acpi_device *device, · line 728

Quoted 9 lines
drivers/acpi/pci_root.c +0 -1
index 4c06c3ffd0cbb..408ba12362a79 100644
@@ -725,7 +725,6 @@ static int acpi_pci_root_add(struct acpi_device *device,
 		dev_err(&device->dev,
 			"Bus %04x:%02x not present in PCI namespace\n",
 			root->segment, (unsigned int)root->secondary.start);
-		device->driver_data = NULL;

[Severity: High]
This isn't a bug introduced by this patch, but does the earlier assignment of device->driver_data = root allow readers to see uninitialized fields?

Earlier in acpi_pci_root_add(), the pointer is published to device->driver_data before critical fields like root->bus are initialized later in the function:

acpi_pci_root_add() {
    ...
    device->driver_data = root;
    ...
    root->bus = pci_acpi_scan_root(root);
    ...
}

If a concurrent caller (such as bus_for_each_dev iteration in cxl_acpi_probe()) fetches the root object, could it observe a NULL root->bus and dereference it before initialization is complete?

Furthermore, without memory barriers, could the CPU reorder the initialization such that readers observe partially written state?

Quoted 3 lines
drivers/acpi/pci_root.c +0 -0
 		result = -ENODEV;
 		goto remove_dmar;
 	}

[ ... ]

drivers/acpi/pci_root.c static int acpi_pci_root_add(struct acpi_device *device, · line 770

Quoted 8 lines
drivers/acpi/pci_root.c +1 -0
@@ -765,6 +764,7 @@ static int acpi_pci_root_add(struct acpi_device *device,
 	if (hotadd)
 		dmar_device_remove(handle);
 end:
+	device->driver_data = NULL;
 	kfree(root);
 	return result;
 }

[Severity: High]
This is a pre-existing issue, but does clearing device->driver_data immediately before kfree(root) leave a use-after-free window for concurrent lockless readers?

If a lockless reader like hisi_pcie_handle_error() or add_host_bridge_uport() calls acpi_pci_find_root(), it reads device->driver_data without taking a lock:

acpi_pci_find_root() {
    ...
    root = acpi_driver_data(device);
    ...
}

If the reader fetches the pointer just before device->driver_data is set to NULL, and then this code proceeds to kfree(root), wouldn't the concurrent reader dereference freed memory?

Quoted 7 lines
drivers/acpi/pci_root.c +1 -0
@@ -788,6 +788,7 @@ static void acpi_pci_root_remove(struct acpi_device *device)
 
 	pci_unlock_rescan_remove();
 
+	device->driver_data = NULL;
 	kfree(root);
 }
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1