[Intel-gfx] [PATCH 3/5] drm/amdgpu: Paper over the drm_driver mangling for virt

Sun Nov 1 09:59:09 UTC 2020

On Sat, Oct 31, 2020 at 6:09 AM Luben Tuikov <luben.tuikov at amd.com> wrote:
>
> On 2020-10-30 08:04, Daniel Vetter wrote:
> > On Fri, Oct 30, 2020 at 11:41 AM Liu, Monk <Monk.Liu at amd.com> wrote:
> >>
> >> [AMD Official Use Only - Internal Distribution Only]
> >>
> >> What's the purpose of the patch sets
> >>
> >> e.g.: what bug can those 5 patches fix or what feature provided
> >>
> >> for this particular one (3/5) I didn't see how it helpful, could you give a background  ?
> >
> > It's good to make function tables const, so that they can be write
> > protected. More resilience against exploits and all that. This patch
> > here is needed to be able to make drm_device->driver const so that all
> > other drivers can make their drm_driver structure const. Would be good
> > to fully fix up amdgpu like in the comment, but I'm not going that in
> > this series here.
> > -Daniel
>
> Hi Daniel,
>
> I feel that that's a good change.
>
> But if you can clarify this for me... Is this leading
> towards a single instance of a struct drm_driver per
> low-level driver, i.e. amdgpu?
>
> And as such, being able to be defined as const?
>
> So that we have many GPU devices driven by one
> low-level driver (amdgpu_drv), represented by one
> const drm_driver (and thus const)?
>
> Which would imply that if varied devices can be handled
> by a single low-level driver, whose struct drm_driver
> settings cannot be shared among subset of devices (say
> very old and new), then the low-level driver
> would have to create more than one "const" struct drm_driver?

This is already the case, minus the const. Which is why it's
problemantic if you change that shared drm_driver instance at runtime
from a specific driver, since you always change it for all instances.
-Daniel

>
> Regards,
> Luben
>
> >
> >>
> >> thanks
> >> _____________________________________
> >> Monk Liu|GPU Virtualization Team |AMD
> >>
> >>
> >> -----Original Message-----
> >> From: Daniel Vetter <daniel.vetter at ffwll.ch>
> >> Sent: Friday, October 30, 2020 6:11 PM
> >> To: DRI Development <dri-devel at lists.freedesktop.org>
> >> Cc: Intel Graphics Development <intel-gfx at lists.freedesktop.org>; Daniel Vetter <daniel.vetter at ffwll.ch>; Deucher, Alexander <Alexander.Deucher at amd.com>; Koenig, Christian <Christian.Koenig at amd.com>; Quan, Evan <Evan.Quan at amd.com>; Kuehling, Felix <Felix.Kuehling at amd.com>; Zhang, Hawking <Hawking.Zhang at amd.com>; Grodzovsky, Andrey <Andrey.Grodzovsky at amd.com>; Tuikov, Luben <Luben.Tuikov at amd.com>; Thomas Zimmermann <tzimmermann at suse.de>; Liu, Monk <Monk.Liu at amd.com>; Yintian Tao <yttao at amd.com>; Li, Dennis <Dennis.Li at amd.com>; Liu, Shaoyun <Shaoyun.Liu at amd.com>; Zhang, Bokun <Bokun.Zhang at amd.com>; Yang, Stanley <Stanley.Yang at amd.com>; Sheng, Wenhui <Wenhui.Sheng at amd.com>; Gong, Curry <Curry.Gong at amd.com>; Daniel Vetter <daniel.vetter at intel.com>
> >> Subject: [PATCH 3/5] drm/amdgpu: Paper over the drm_driver mangling for virt
> >>
> >> Prep work to make drm_device->driver const.
> >>
> >> Signed-off-by: Daniel Vetter <daniel.vetter at ffwll.ch>
> >> Cc: Alex Deucher <alexander.deucher at amd.com>
> >> Cc: "Christian König" <christian.koenig at amd.com>
> >> Cc: Evan Quan <evan.quan at amd.com>
> >> Cc: Felix Kuehling <Felix.Kuehling at amd.com>
> >> Cc: Hawking Zhang <Hawking.Zhang at amd.com>
> >> Cc: Andrey Grodzovsky <andrey.grodzovsky at amd.com>
> >> Cc: Luben Tuikov <luben.tuikov at amd.com>
> >> Cc: Thomas Zimmermann <tzimmermann at suse.de>
> >> Cc: Monk Liu <Monk.Liu at amd.com>
> >> Cc: Yintian Tao <yttao at amd.com>
> >> Cc: Dennis Li <Dennis.Li at amd.com>
> >> Cc: shaoyunl <shaoyun.liu at amd.com>
> >> Cc: Bokun Zhang <Bokun.Zhang at amd.com>
> >> Cc: "Stanley.Yang" <Stanley.Yang at amd.com>
> >> Cc: Wenhui Sheng <Wenhui.Sheng at amd.com>
> >> Cc: chen gong <curry.gong at amd.com>
> >> Signed-off-by: Daniel Vetter <daniel.vetter at intel.com>
> >> ---
> >>  drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c  |  8 ++++----  drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 12 +++++++++++-
> >>  2 files changed, 15 insertions(+), 5 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >> index 024c3b70b1aa..3d337f13ae4e 100644
> >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >> @@ -1093,7 +1093,7 @@ static const struct pci_device_id pciidlist[] = {
> >>
> >>  MODULE_DEVICE_TABLE(pci, pciidlist);
> >>
> >> -static struct drm_driver kms_driver;
> >> +struct drm_driver amdgpu_kms_driver;
> >>
> >>  static int amdgpu_pci_probe(struct pci_dev *pdev,
> >>      const struct pci_device_id *ent) @@ -1164,7 +1164,7 @@ static int amdgpu_pci_probe(struct pci_dev *pdev,
> >>  if (ret)
> >>  return ret;
> >>
> >> -adev = devm_drm_dev_alloc(&pdev->dev, &kms_driver, typeof(*adev), ddev);
> >> +adev = devm_drm_dev_alloc(&pdev->dev, &amdgpu_kms_driver,
> >> +typeof(*adev), ddev);
> >>  if (IS_ERR(adev))
> >>  return PTR_ERR(adev);
> >>
> >> @@ -1508,7 +1508,7 @@ int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv)
> >>  return 0;
> >>  }
> >>
> >> -static struct drm_driver kms_driver = {
> >> +struct drm_driver amdgpu_kms_driver = {
> >>  .driver_features =
> >>      DRIVER_ATOMIC |
> >>      DRIVER_GEM |
> >> @@ -1571,7 +1571,7 @@ static int __init amdgpu_init(void)
> >>  goto error_fence;
> >>
> >>  DRM_INFO("amdgpu kernel modesetting enabled.\n");
> >> -kms_driver.num_ioctls = amdgpu_max_kms_ioctl;
> >> +amdgpu_kms_driver.num_ioctls = amdgpu_max_kms_ioctl;
> >>  amdgpu_register_atpx_handler();
> >>
> >>  /* Ignore KFD init failures. Normal when CONFIG_HSA_AMD is not set. */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c
> >> index d0aea5e39531..dde4c449c284 100644
> >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c
> >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c
> >> @@ -45,13 +45,23 @@ bool amdgpu_virt_mmio_blocked(struct amdgpu_device *adev)
> >>  return RREG32_NO_KIQ(0xc040) == 0xffffffff;  }
> >>
> >> +extern struct drm_driver amdgpu_kms_driver;
> >> +
> >>  void amdgpu_virt_init_setting(struct amdgpu_device *adev)  {
> >>  /* enable virtual display */
> >>  if (adev->mode_info.num_crtc == 0)
> >>  adev->mode_info.num_crtc = 1;
> >>  adev->enable_virtual_display = true;
> >> -adev_to_drm(adev)->driver->driver_features &= ~DRIVER_ATOMIC;
> >> +
> >> +/*
> >> + * FIXME: Either make virt support atomic or make sure you have two
> >> + * drm_driver structs, these kind of tricks are only ok when there's
> >> + * guaranteed only a single device per system. This should also be done
> >> + * before struct drm_device is initialized.
> >> + */
> >> +amdgpu_kms_driver.driver_features &= ~DRIVER_ATOMIC;
> >> +
> >>  adev->cg_flags = 0;
> >>  adev->pg_flags = 0;
> >>  }
> >> --
> >> 2.28.0
> >>
> >
> >
>


-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch