On Tue, Nov 17, 2020 at 01:44:07PM +0800, Zhenzhong Duan wrote:
> vfio-pci and pci-stub use new_id to bind devices. But one can add same IDs
> multiple times, for example:
>
> # echo "1af4 1041" > /sys/bus/pci/drivers/vfio-pci/new_id
> # echo "1af4 1041" > /sys/bus/pci/drivers/vfio-pci/new_id
> # echo "1af4 1041" > /sys/bus/pci/drivers/vfio-pci/new_id
> # echo "1af4 1041" > /sys/bus/pci/drivers/vfio-pci/remove_id
> # echo "1af4 1041" > /sys/bus/pci/drivers/vfio-pci/remove_id
> # echo "1af4 1041" > /sys/bus/pci/drivers/vfio-pci/remove_id
> # echo "1af4 1041" > /sys/bus/pci/drivers/vfio-pci/remove_id
> -bash: echo: write error: No such device
>
> This doesn't cause user-visible broken behavior, but not user friendly.
> he has to remove same IDs same times to ensure it's completely gone.
>
> Changed to only allow one dynamic entry of the same kind, after fix:
>
> # echo "1af4 1041" > /sys/bus/pci/drivers/vfio-pci/new_id
> # echo "1af4 1041" > /sys/bus/pci/drivers/vfio-pci/new_id
> -bash: echo: write error: File exists
> # echo "1af4 1041" > /sys/bus/pci/drivers/vfio-pci/remove_id
> # echo "1af4 1041" > /sys/bus/pci/drivers/vfio-pci/remove_id
> -bash: echo: write error: No such device
>
>
> v3: add a separate patch to process dependency issue per Bjorn
> make commit log more clear per Bjorn
> v2: revert the export of pci_match_device() per Christoph
> combind PATCH1 and PATCH2 into one.
>
> v2 link:https://lkml.org/lkml/2020/10/25/347
>
> Zhenzhong Duan (2):
> PCI: move pci_match_device() ahead of new_id_store()
> PCI: avoid duplicate IDs in dynamic IDs list
>
> drivers/pci/pci-driver.c | 146
> +++
> 1 file changed, 73 insertions(+), 73 deletions(-)
>
> --
> 1.8.3.1
>
>
> Zhenzhong Duan (2):
> PCI: move pci_match_device() ahead of new_id_store()
> PCI: avoid duplicate IDs in dynamic IDs list
>
> drivers/pci/pci-driver.c | 146
> +++
> 1 file changed, 73 insertions(+), 73 deletions(-)
I corrected the subject lines:
PCI: Move pci_match_device() ahead of new_id_store()
PCI: Avoid duplicate IDs in driver dynamic IDs list
and applied both to pci/enumeration for v5.11, thanks!
