[PATCH v2] PCI/AER: Fix struct pci_dev reference leak in aer_process_err_devices()

[Priyank Rathod <rathodpriyank@google.com>]

When an AER error occurs, candidate error-source devices are added to
e_info->dev[] by add_error_device(), which takes a reference with
pci_dev_get().

A device can be recorded without any error status being present: the
Requester ID fast path in is_error_source(),

	if (e_info->id == pci_dev_id(dev))
		return true;

matches purely on the ID reported by the Root Port and returns true
without reading the device's AER status registers.

In aer_process_err_devices(), handle_error_source() is called only if
aer_get_device_error_info() returns non-zero, i.e. only if an unmasked
error status bit is actually set. It returns 0 if the device has become
inaccessible (status and mask both read as all ones, so
status & ~mask == 0) or if no unmasked status bit is set.

Because handle_error_source() was responsible for calling pci_dev_put(),
skipping it permanently leaks the reference taken in add_error_device().

Decouple reference lifetime from error handling by moving pci_dev_put()
out of handle_error_source() and into the aer_process_err_devices() loop,
so every recorded device is put exactly once.

handle_error_source() is static and aer_process_err_devices() is its only
caller, so no other path is affected.

Fixes: 60271ab044a5 ("PCI/AER: Take reference on error devices")
Signed-off-by: Priyank Rathod <rathodpriyank@google.com>
---
Changes in v2:
- Corrected Fixes tag from 1ab4a3c80508 to 60271ab044a5 ("PCI/AER: Take
  reference on error devices"), which added both the pci_dev_get() in
  add_error_device() and the conditionally-reached pci_dev_put() in
  handle_error_source(). Thanks to Lukas Wunner for catching this.
- Removed the speculative topology list from the commit message. As Lukas
  pointed out, error reporting is not enabled on devices without an AER
  capability (pcie_aer_is_native() bails on !dev->aer_cap), so the
  dev->aer_cap == 0 reasoning was wrong and is gone.
- Explained instead why a device with no error status can be present in
  e_info->dev[]: the Requester ID fast path in is_error_source() matches
  on e_info->id alone, without reading any AER status register.
- Added a Fixes tag; the imbalance dates back to v4.20. I have not added
  Cc: stable, since you indicated the path is an unlikely corner case -
  happy to add it if you think it is warranted.
- Cc: Keith Busch and Sinan Kaya, author and reviewer of 60271ab044a5.
- Rebased onto v7.3-rc3+ (f259f446f519); applies cleanly to pci/next as well.
- Link to v1: https://lore.kernel.org/r/20260830-fix-aer-refcount-leak-v1-1-64e1013add12@google.com
---
 drivers/pci/pcie/aer.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
index d8dcd238fda1..bc761410d56d 100644
--- a/drivers/pci/pcie/aer.c
+++ b/drivers/pci/pcie/aer.c
@@ -1340,7 +1340,6 @@ static void handle_error_source(struct pci_dev *dev, struct aer_err_info *info)
 {
 	cxl_rch_handle_error(dev, info);
 	pci_aer_handle_error(dev, info);
-	pci_dev_put(dev);
 }
 
 #ifdef CONFIG_ACPI_APEI_PCIEAER
@@ -1518,6 +1517,7 @@ static inline void aer_process_err_devices(struct aer_err_info *e_info)
 	for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
 		if (aer_get_device_error_info(e_info, i))
 			handle_error_source(e_info->dev[i], e_info);
+		pci_dev_put(e_info->dev[i]);
 	}
 }
 

---
base-commit: a077be4fde21ee6e751fa70eb641ef5d9bf2fc48
change-id: 20260830-fix-aer-refcount-leak-6378f84d62ba

Best regards,
-- 
Priyank Rathod <rathodpriyank@google.com>