On Tue, Nov 22, 2022 at 12:10 PM Lazar, Lijo <Lijo.Lazar@xxxxxxx> wrote:
>
> [AMD Official Use Only - General]
>
> When only second GPU has valid ATRM handle -
> then it stays inside the loop and in the next call to pci_get_class(), it passes pdev reference to first GPU as the "from" param. That time it drops the reference count of "from" device.
ah, right, that was the part I missed. Thanks.
Alex
>
> Thanks,
> Lijo
> ________________________________
> From: Alex Deucher <alexdeucher@xxxxxxxxx>
> Sent: Tuesday, November 22, 2022 9:55:33 PM
> To: Lazar, Lijo <Lijo.Lazar@xxxxxxx>
> Cc: Xiongfeng Wang <wangxiongfeng2@xxxxxxxxxx>; Deucher, Alexander <Alexander.Deucher@xxxxxxx>; Koenig, Christian <Christian.Koenig@xxxxxxx>; Pan, Xinhui <Xinhui.Pan@xxxxxxx>; airlied@xxxxxxxxx <airlied@xxxxxxxxx>; daniel@xxxxxxxx <daniel@xxxxxxxx>; Zhang, Hawking <Hawking.Zhang@xxxxxxx>; dri-devel@xxxxxxxxxxxxxxxxxxxxx <dri-devel@xxxxxxxxxxxxxxxxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx <amd-gfx@xxxxxxxxxxxxxxxxxxxxx>; yangyingliang@xxxxxxxxxx <yangyingliang@xxxxxxxxxx>
> Subject: Re: [PATCH 1/2] drm/radeon: Fix PCI device refcount leak in radeon_atrm_get_bios()
>
> On Tue, Nov 22, 2022 at 9:59 AM Lazar, Lijo <lijo.lazar@xxxxxxx> wrote:
> >
> >
> >
> > On 11/22/2022 8:19 PM, Alex Deucher wrote:
> > > On Tue, Nov 22, 2022 at 6:12 AM Xiongfeng Wang
> > > <wangxiongfeng2@xxxxxxxxxx> wrote:
> > >>
> > >> As comment of pci_get_class() says, it returns a pci_device with its
> > >> refcount increased and decreased the refcount for the input parameter
> > >> @from if it is not NULL.
> > >>
> > >> If we break the loop in radeon_atrm_get_bios() with 'pdev' not NULL, we
> > >> need to call pci_dev_put() to decrease the refcount. Add the missing
> > >> pci_dev_put() to avoid refcount leak.
> > >
> > > For both patches, I think pci_dev_put() needs to go into the loops.
> > > There are 2 or more GPUs on the systems where this is relevant.
> > >
> >
> > As per the logic, the loop breaks when it finds a valid ATRM handle. So
> > dev_put is required only for that device.
>
> Sure, but what if the handle is on the second DISPLAY_VGA or
> DISPLAY_OTHER class PCI device on the system? We've already called
> pci_get_class() for the first PCI device that matched.
>
> Alex
>
> >
> > When inside the loop this happens - "decreased the refcount for the
> > input parameter @from if it is not NULL"
> >
> > Thanks,
> > Lijo
> >
> > > Alex
> > >
> > >>
> > >> Fixes: d8ade3526b2a ("drm/radeon: handle non-VGA class pci devices with ATRM")
> > >> Fixes: c61e2775873f ("drm/radeon: split ATRM support out from the ATPX handler (v3)")
> > >> Signed-off-by: Xiongfeng Wang <wangxiongfeng2@xxxxxxxxxx>
> > >> ---
> > >> drivers/gpu/drm/radeon/radeon_bios.c | 1 +
> > >> 1 file changed, 1 insertion(+)
> > >>
> > >> diff --git a/drivers/gpu/drm/radeon/radeon_bios.c b/drivers/gpu/drm/radeon/radeon_bios.c
> > >> index 33121655d50b..2df6ce3e32cb 100644
> > >> --- a/drivers/gpu/drm/radeon/radeon_bios.c
> > >> +++ b/drivers/gpu/drm/radeon/radeon_bios.c
> > >> @@ -227,6 +227,7 @@ static bool radeon_atrm_get_bios(struct radeon_device *rdev)
> > >>
> > >> if (!found)
> > >> return false;
> > >> + pci_dev_put(pdev);
> > >>
> > >> rdev->bios = kmalloc(size, GFP_KERNEL);
> > >> if (!rdev->bios) {
> > >> --
> > >> 2.20.1
> > >>
