<html> <head> <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1"> <meta name="Generator" content="Microsoft Word 15 (filtered medium)"> <style></style> </head> <body lang="EN-US" link="#0563C1" vlink="#954F72"> <div class="WordSection1"> I just try to explain what I want to do here, no real patch formalized yet <o:p></o:p> <o:p> </o:p> <div> _____________________________________<o:p></o:p> Monk Liu|GPU Virtualization Team |AMD<o:p></o:p> <img width="80" height="80" style="width:.8333in;height:.8333in" id="Picture_x0020_1" src="cid:image001.png@01D6172B.627DC310" alt="sig-cloud-gpu"><o:p></o:p> </div> <o:p> </o:p> <div> <div style="border:none;border-top:solid #E1E1E1 1.0pt;padding:3.0pt 0in 0in 0in"> From: He, Jacob <Jacob.He@amd.com> Sent: Monday, April 20, 2020 3:45 PM To: Liu, Monk <Monk.Liu@amd.com>; Koenig, Christian <Christian.Koenig@amd.com> Cc: amd-gfx@lists.freedesktop.org Subject: Re: why we need to do infinite RLC_SPM register setting during VM flush<o:p></o:p> </div> </div> <o:p> </o:p> [AMD Official Use Only - Internal Distribution Only]<o:p></o:p> <o:p> </o:p> <div> <div> Do you miss a file which adds spm_updated to vm structure?<o:p></o:p> </div> <div class="MsoNormal" align="center" style="text-align:center"> <hr size="2" width="98%" align="center"> </div> <div id="divRplyFwdMsg"> From: Liu, Monk <<a href="mailto:Monk.Liu@amd.com">Monk.Liu@amd.com</a>> Sent: Monday, April 20, 2020 3:32 PM To: He, Jacob <<a href="mailto:Jacob.He@amd.com">Jacob.He@amd.com</a>>; Koenig, Christian <<a href="mailto:Christian.Koenig@amd.com">Christian.Koenig@amd.com</a>> Cc: <a href="mailto:amd-gfx@lists.freedesktop.org">amd-gfx@lists.freedesktop.org</a> <<a href="mailto:amd-gfx@lists.freedesktop.org">amd-gfx@lists.freedesktop.org</a>> Subject: why we need to do infinite RLC_SPM register setting during VM flush <o:p></o:p> <div> <o:p></o:p> </div> </div> <div> <div> Hi Jaco & Christian<o:p></o:p> <o:p></o:p> As titled , check below patch:<o:p></o:p> <o:p></o:p> commit 10790a09ea584cc832353a5c2a481012e5e31a13<o:p></o:p> Author: Jacob He <<a href="mailto:jacob.he@amd.com">jacob.he@amd.com</a>><o:p></o:p> Date: Fri Feb 28 20:24:41 2020 +0800<o:p></o:p> <o:p></o:p> drm/amdgpu: Update SPM_VMID with the job's vmid when application reserves the vmid<o:p></o:p> <o:p></o:p> SPM access the video memory according to SPM_VMID. It should be updated<o:p></o:p> with the job's vmid right before the job is scheduled. SPM_VMID is a<o:p></o:p> global resource<o:p></o:p> <o:p></o:p> Change-Id: Id3881908960398f87e7c95026a54ff83ff826700<o:p></o:p> Signed-off-by: Jacob He <<a href="mailto:jacob.he@amd.com">jacob.he@amd.com</a>><o:p></o:p> Reviewed-by: Christian König <<a href="mailto:christian.koenig@amd.com">christian.koenig@amd.com</a>><o:p></o:p> <o:p></o:p> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c<o:p></o:p> index 6e6fc8c..ba2236a 100644<o:p></o:p> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c<o:p></o:p> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c<o:p></o:p> @@ -1056,8 +1056,12 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,<o:p></o:p> struct dma_fence *fence = NULL;<o:p></o:p> bool pasid_mapping_needed = false;<o:p></o:p> unsigned patch_offset = 0;<o:p></o:p> + bool update_spm_vmid_needed = (job->vm && (job->vm->reserved_vmid[vmhub] != NULL));<o:p></o:p> int r;<o:p></o:p> <o:p></o:p> + if (update_spm_vmid_needed && adev->gfx.rlc.funcs->update_spm_vmid)<o:p></o:p> + adev->gfx.rlc.funcs->update_spm_vmid(adev, job->vmid);<o:p></o:p> +<o:p></o:p> if (amdgpu_vmid_had_gpu_reset(adev, id)) {<o:p></o:p> gds_switch_needed = true;<o:p></o:p> vm_flush_needed = true;<o:p></o:p> <o:p></o:p> this update_spm_vmid() looks an completely overkill to me, we only need to do it once for its VM …<o:p></o:p> <o:p></o:p> in SRIOV the register reading/writing for update_spm_vmid() is now carried by KIQ thus there is too much burden on KIQ for such unnecessary jobs …<o:p></o:p> <o:p></o:p> I want to change it to only do it once per VM, like:<o:p></o:p> <o:p></o:p> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c<o:p></o:p> index 6e6fc8c..ba2236a 100644<o:p></o:p> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c<o:p></o:p> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c<o:p></o:p> @@ -1056,8 +1056,12 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,<o:p></o:p> struct dma_fence *fence = NULL;<o:p></o:p> bool pasid_mapping_needed = false;<o:p></o:p> unsigned patch_offset = 0;<o:p></o:p> + bool update_spm_vmid_needed = (job->vm && (job->vm->reserved_vmid[vmhub] != NULL));<o:p></o:p> int r;<o:p></o:p> <o:p></o:p> + if (update_spm_vmid_needed && adev->gfx.rlc.funcs->update_spm_vmid && !vm->spm_updated) {<o:p></o:p> + adev->gfx.rlc.funcs->update_spm_vmid(adev, job->vmid); <o:p></o:p> + vm->spm_updated = true;<o:p></o:p> + }<o:p></o:p> <o:p></o:p> if (amdgpu_vmid_had_gpu_reset(adev, id)) {<o:p></o:p> gds_switch_needed = true;<o:p></o:p> vm_flush_needed = true;<o:p></o:p> <o:p></o:p> what do you think ?<o:p></o:p> <o:p></o:p> P.S.: the best way is to let GFX ring itself to do the update_spm_vmid() instead of let CPU doing it, e.g.: we put more PM4 command in VM-FLUSH packets …. <o:p></o:p> But I prefer the simple way first like I demonstrated above <o:p></o:p> _____________________________________<o:p></o:p> Monk Liu|GPU Virtualization Team |AMD<o:p></o:p> <img border="0" width="80" height="80" style="width:.8333in;height:.8333in" id="x_Picture_x0020_1" src="cid:image001.png@01D6172B.627DC310" alt="sig-cloud-gpu"><o:p></o:p> <o:p></o:p> </div> </div> </div> </div> </body> </html>