On 03/08/ , Christian König wrote:
>
>
> Am 08.03.22 um 04:39 schrieb Lang Yu:
> > It is a hardware issue that VCN can't handle a GTT
> > backing stored TMZ buffer on Raven.
> >
> > Move such a TMZ buffer to VRAM domain before command
> > submission.
> >
> > v2:
> > - Use patch_cs_in_place callback.
> >
> > Suggested-by: Christian König <christian.koenig@xxxxxxx>
> > Signed-off-by: Lang Yu <Lang.Yu@xxxxxxx>
> > ---
> > drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c | 68 +++++++++++++++++++++++++++
> > 1 file changed, 68 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
> > index 7bbb9ba6b80b..810932abd3af 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
> > @@ -24,6 +24,7 @@
> > #include <linux/firmware.h>
> > #include "amdgpu.h"
> > +#include "amdgpu_cs.h"
> > #include "amdgpu_vcn.h"
> > #include "amdgpu_pm.h"
> > #include "soc15.h"
> > @@ -1905,6 +1906,72 @@ static const struct amd_ip_funcs vcn_v1_0_ip_funcs = {
> > .set_powergating_state = vcn_v1_0_set_powergating_state,
> > };
> > +/**
> > + * It is a hardware issue that Raven VCN can't handle a GTT TMZ buffer.
> > + * Move such a GTT TMZ buffer to VRAM domain before command submission.
> > + */
> > +static int vcn_v1_0_validate_bo(struct amdgpu_cs_parser *parser,
> > + struct amdgpu_job *job,
> > + uint64_t addr)
> > +{
> > + struct ttm_operation_ctx ctx = { false, false };
> > + struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
> > + struct amdgpu_vm *vm = &fpriv->vm;
> > + struct amdgpu_bo_va_mapping *mapping;
> > + struct amdgpu_bo *bo;
> > + int r;
> > +
> > + addr &= AMDGPU_GMC_HOLE_MASK;
> > + if (addr & 0x7) {
> > + DRM_ERROR("VCN messages must be 8 byte aligned!\n");
> > + return -EINVAL;
> > + }
> > +
> > + mapping = amdgpu_vm_bo_lookup_mapping(vm, addr/AMDGPU_GPU_PAGE_SIZE);
> > + if (!mapping || !mapping->bo_va || !mapping->bo_va->base.bo)
> > + return -EINVAL;
> > +
> > + bo = mapping->bo_va->base.bo;
> > + if (!(bo->flags & AMDGPU_GEM_CREATE_ENCRYPTED))
> > + return 0;
> > +
> > + amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_VRAM);
> > + r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
> > + if (r) {
> > + DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
> > + return r;
> > + }
>
> Well, exactly that won't work.
>
> The message structure isn't TMZ protected because otherwise the driver won't
> be able to stitch it together.
>
> What is TMZ protected are the surfaces the message structure is pointing to.
> So what you would need to do is to completely parse the structure and then
> move on the relevant buffers into VRAM.
>
> Leo or James, can you help with that?
