On Wed, Sep 27, 2023 at 1:37 AM Christian König
<ckoenig.leichtzumer...@gmail.com> wrote:
>
> I'm still not happy with moving the GART fixed to the end. We abandoned
> this for good reasons.
We didn't abandon it, there is still code which decides where to put
the gart based on the relative sizes of the address space before and
after vram. It really comes down to where the vbios decides to put
vram. If it starts somewhere lower in the future, gart will naturally
end up higher. Also windows always puts gart above vram.
>
> If we really go this way I would prefer to have this as parameter to the
> amdgpu_gmc_gart_location() function and not in the gmc structure.
I can do that.
Alex
>
> Regards,
> Christian.
>
> Am 26.09.23 um 19:30 schrieb Alex Deucher:
> > Ping on this series?
> >
> > On Thu, Sep 21, 2023 at 5:46 PM Alex Deucher <alexander.deuc...@amd.com>
> > wrote:
> >> We normally place GART based on the location of VRAM and the
> >> available address space around that, but provide an option
> >> to force a particular location for hardware that needs it.
> >>
> >> Signed-off-by: Alex Deucher <alexander.deuc...@amd.com>
> >> ---
> >> drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c | 19 +++++++++++++++----
> >> drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h | 8 ++++++++
> >> 2 files changed, 23 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
> >> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
> >> index f74a51a93ebb..d1d98488373b 100644
> >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
> >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
> >> @@ -287,11 +287,22 @@ void amdgpu_gmc_gart_location(struct amdgpu_device
> >> *adev, struct amdgpu_gmc *mc)
> >> mc->gart_size = max(size_bf, size_af);
> >> }
> >>
> >> - if ((size_bf >= mc->gart_size && size_bf < size_af) ||
> >> - (size_af < mc->gart_size))
> >> - mc->gart_start = 0;
> >> - else
> >> + switch (mc->gart_placement) {
> >> + case AMDGPU_GART_PLACEMENT_HIGH:
> >> mc->gart_start = max_mc_address - mc->gart_size + 1;
> >> + break;
> >> + case AMDGPU_GART_PLACEMENT_LOW:
> >> + mc->gart_start = 0;
> >> + break;
> >> + case AMDGPU_GART_PLACEMENT_BEST_FIT:
> >> + default:
> >> + if ((size_bf >= mc->gart_size && size_bf < size_af) ||
> >> + (size_af < mc->gart_size))
> >> + mc->gart_start = 0;
> >> + else
> >> + mc->gart_start = max_mc_address - mc->gart_size +
> >> 1;
> >> + break;
> >> + }
> >>
> >> mc->gart_start &= ~(four_gb - 1);
> >> mc->gart_end = mc->gart_start + mc->gart_size - 1;
> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> >> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> >> index dd0ede75e5d7..fcef057b9213 100644
> >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> >> @@ -199,6 +199,12 @@ struct amdgpu_mem_partition_info {
> >>
> >> #define INVALID_PFN -1
> >>
> >> +enum amdgpu_gart_placement {
> >> + AMDGPU_GART_PLACEMENT_BEST_FIT = 0,
> >> + AMDGPU_GART_PLACEMENT_HIGH,
> >> + AMDGPU_GART_PLACEMENT_LOW,
> >> +};
> >> +
> >> struct amdgpu_gmc {
> >> /* FB's physical address in MMIO space (for CPU to
> >> * map FB). This is different compared to the agp/
> >> @@ -339,6 +345,8 @@ struct amdgpu_gmc {
> >> bool flush_tlb_needs_extra_type_0;
> >> bool flush_tlb_needs_extra_type_2;
> >> bool flush_pasid_uses_kiq;
> >> +
> >> + enum amdgpu_gart_placement gart_placement;
> >> };
> >>
> >> #define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr)
> >> (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr))
> >> --
> >> 2.41.0
> >>
>
