Re: [外部邮件] Re: [PATCH v2] PCI/ACPI: Check the found handle in acpi_get_rc_resources()
From: Rafael J. Wysocki (Intel)
Date: Mon Nov 16 2026 - 07:00:51 EST
On Mon, Nov 16, 2026 at 4:05 AM Li,Rongqing <lirongqing@xxxxxxxxx> wrote:
>
> > > From: Li RongQing <lirongqing@xxxxxxxxx>
> > >
> > > acpi_get_devices() returns AE_OK even when the namespace walk finishes
> > > without a match, in which case the return_value argument is not written.
> > > The caller passes an uninitialized acpi_handle and hands it to
> > > acpi_fetch_acpi_dev(), which dereferences it as an ACPI node.
> > >
> > > Initialize the handle and reject a missing match.
> > >
> > > Fixes: 169de969c018 ("PCI/ACPI: Provide acpi_get_rc_resources() for
> > > ARM64 platform")
> > > Signed-off-by: Li RongQing <lirongqing@xxxxxxxxx>
> > > ---
> > > Diff with v1: add fixes tag
> > >
> > > drivers/pci/pci-acpi.c | 5 ++++-
> > > 1 file changed, 4 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/pci/pci-acpi.c b/drivers/pci/pci-acpi.c index
> > > e260089..e90e749 100644
> > > --- a/drivers/pci/pci-acpi.c
> > > +++ b/drivers/pci/pci-acpi.c
> > > @@ -82,7 +82,7 @@ int acpi_get_rc_resources(struct device *dev, const
> > > char *hid, u16 segment, {
> > > struct acpi_device *adev;
> > > acpi_status status;
> > > - acpi_handle handle;
> > > + acpi_handle handle = NULL;
> > > int ret;
> > >
> > > status = acpi_get_devices(hid, acpi_match_rc, &segment,
> > > &handle); @@ -92,6 +92,9 @@ int acpi_get_rc_resources(struct device
> > *dev, const char *hid, u16 segment,
> > > return -ENODEV;
> > > }
> > >
> > > + if (!handle)
> > > + return -ENODEV;
> >
> > acpi_fetch_acpi_dev() will return NULL if handle is NULL, won't it?
> >
>
> Agreed that acpi_fetch_acpi_dev(NULL) is safe today — acpi_get_data_full() rejects NULL up front. But that safety is incidental, not contractual: the handle layer doesn't uniformly reject NULL (acpi_ns_validate_handle() even maps it to the root node), so relying on it makes the fix fragile to future changes. The explicit check states the "no match ⇒ -ENODEV" intent directly and costs 3 lines. If you prefer the minimal diff, I can drop it — the initialization alone is also correct.
Yes, please!