On 2/23/2026 9:15 AM, Shawn Lin wrote:
Hi Jeff
在 2026/02/24 星期二 0:01, Jeff Hugo 写道:
On 2/23/2026 8:55 AM, Shawn Lin wrote:
pcim_enable_device() no longer automatically manages IRQ vectors via
devres.
Drivers must now manually call pci_free_irq_vectors() for cleanup.
Alternatively,
pcim_alloc_irq_vectors() should be used.
This seems to break bisect. Surely you should update drivers before
you go changing behavior.
Patch 1 introduces pcim_alloc_irq_vectors() and then convert all needed
users to it. Then the final patch(37) remove the implict IRQ vector
management in pci_alloc_irq_vectors(). So the individual patch for each
driver should be safe IMO. And if these patches go via a single
tree with acks, it seems unlikely to break bisect.
There is a conflict here between the patch description and what you have
just described. It seems like the patch description is wrong.
The patch says that pcim_enable_device() has already changed behavior,
and now drivers are broken and must update.
It seems that the description should be updated.
-Jeff
