[AMD Public Use]


The full vbios versioning information consists of the version (numeric), build date and the name... I this this interface exposes only the name.

Should we expose those on different sysfs files or just combine all of them in a single file?

David

From: Kees Cook <keescook@chromium.org>
Sent: Saturday, May 8, 2021 2:51 AM
To: Gu, JiaWei (Will) <JiaWei.Gu@amd.com>
Cc: Deucher, Alexander <Alexander.Deucher@amd.com>; StDenis, Tom <Tom.StDenis@amd.com>; Christian König <ckoenig.leichtzumerken@gmail.com>; amd-gfx@lists.freedesktop.org <amd-gfx@lists.freedesktop.org>; Nieto, David M <David.Nieto@amd.com>; linux-next@vger.kernel.org <linux-next@vger.kernel.org>
Subject: Re: [PATCH] drm/amdgpu: Add vbios info ioctl interface
 
On Sat, May 08, 2021 at 06:01:23AM +0000, Gu, JiaWei (Will) wrote:
> [AMD Official Use Only - Internal Distribution Only]
>
> Thanks for this catching Kees.
>
> Yes it should be 20, not 16. I was not aware that serial size had been changed from 16 to 20 in struct amdgpu_device.
> Will submit a fix soon.

You might want to add a BUILD_BUG_ON() to keep those in sync, especially
since it's about to be UAPI.

-Kees

>
> Best regards,
> Jiawei
>
>
> -----Original Message-----
> From: Kees Cook <keescook@chromium.org>
> Sent: Saturday, May 8, 2021 12:28 PM
> To: Gu, JiaWei (Will) <JiaWei.Gu@amd.com>; Deucher, Alexander <Alexander.Deucher@amd.com>
> Cc: StDenis, Tom <Tom.StDenis@amd.com>; Deucher, Alexander <Alexander.Deucher@amd.com>; Christian König <ckoenig.leichtzumerken@gmail.com>; Gu, JiaWei (Will) <JiaWei.Gu@amd.com>; amd-gfx@lists.freedesktop.org; Nieto, David M <David.Nieto@amd.com>; linux-next@vger.kernel.org
> Subject: Re: [PATCH] drm/amdgpu: Add vbios info ioctl interface
>
> Hi!
>
> This patch needs some fixing.
>
> On Thu, Apr 22, 2021 at 10:34:48AM +0800, Jiawei Gu wrote:
> > +           case AMDGPU_INFO_VBIOS_INFO: {
> > +                   struct drm_amdgpu_info_vbios vbios_info = {};
> > +                   struct atom_context *atom_context;
> > +
> > +                   atom_context = adev->mode_info.atom_context;
> > +                   memcpy(vbios_info.name, atom_context->name, sizeof(atom_context->name));
> > +                   vbios_info.dbdf = PCI_DEVID(adev->pdev->bus->number, adev->pdev->devfn);
> > +                   memcpy(vbios_info.vbios_pn, atom_context->vbios_pn, sizeof(atom_context->vbios_pn));
> > +                   vbios_info.version = atom_context->version;
> > +                   memcpy(vbios_info.date, atom_context->date, sizeof(atom_context->date));
> > +                   memcpy(vbios_info.serial, adev->serial, sizeof(adev->serial));
>
> This writes beyond the end of vbios_info.serial.
>
> > +                   vbios_info.dev_id = adev->pdev->device;
> > +                   vbios_info.rev_id = adev->pdev->revision;
> > +                   vbios_info.sub_dev_id = atom_context->sub_dev_id;
> > +                   vbios_info.sub_ved_id = atom_context->sub_ved_id;
>
> Though it gets "repaired" by these writes.
>
> > +
> > +                   return copy_to_user(out, &vbios_info,
> > +                                           min((size_t)size, sizeof(vbios_info))) ? -EFAULT : 0;
> > +           }
>
> sizeof(adev->serial) != sizeof(vbios_info.serial)
>
> adev is struct amdgpu_device:
>
> struct amdgpu_device {
> ...
>         char                            serial[20];
>
>
> > +struct drm_amdgpu_info_vbios {
> > [...]
> > +   __u8 serial[16];
> > +   __u32 dev_id;
> > +   __u32 rev_id;
> > +   __u32 sub_dev_id;
> > +   __u32 sub_ved_id;
> > +};
>
> Is there a truncation issue (20 vs 16) and is this intended to be a NUL-terminated string?
>
> --
> Kees Cook

--
Kees Cook