- 07 Jul, 2016 40 commits
-
-
Chunming Zhou authored
Signed-off-by:
Chunming Zhou <David1.Zhou@amd.com> Reviewed-by:
Christian König <christian.koenig@amd.com> Signed-off-by:
Alex Deucher <alexander.deucher@amd.com>
-
Chunming Zhou authored
V3: directly use pd_addr. Signed-off-by:
-
Chunming Zhou authored
Signed-off-by:
-
Chunming Zhou authored
which avoids job->vm_pd_addr be changed. V2: pass job structure to amdgpu_vm_grab_id and amdgpu_vm_flush directly. Signed-off-by:
-
Chunming Zhou authored
Which is to recover hw jobs when gpu reset. Signed-off-by:
-
Chunming Zhou authored
After all hw jobs are reset, hw fence is meaningless, so force_completion Cc: William Lewis <minutemaidpark@hotmail.com> Signed-off-by:
-
Chunming Zhou authored
Signed-off-by:
-
Chunming Zhou authored
amd_sched_hw_job_reset will remove callback from hw fence. Signed-off-by:
-
Chunming Zhou authored
Parent of sched fence is hw fence which is to signal sched fence. Signed-off-by:
-
Chunming Zhou authored
Previous vm fault is since page talbe losts connection with vmid after gpu reset. Now the issue is fixed by recovery. No need more. If we want to save vram for some EDC card, we will need to consider a complete solution. Signed-off-by:
-
Chunming Zhou authored
Signed-off-by:
-
Christian König authored
We return the fence as part of the job structur anyway, no need to do this twice. Signed-off-by:
-
Christian König authored
Keep the time we don't have a fence associated with the resource smaller. Signed-off-by:
-
Christian König authored
The fence and the sync object are not hardware resources. Signed-off-by:
-
Christian König authored
Same problem as with the VM page tables. The user fence address must be determined before the job is scheduled, not when the IB is executed. This fixes a security problem where user fences could be used to overwrite any part of VRAM. Signed-off-by:
-
Tom St Denis authored
(v2) Added INSTANCE selector (v3) Changed order of bank selectors Signed-off-by:
Tom St Denis <tom.stdenis@amd.com> Reviewed-by:
Nicolai Hähnle <nicolai.haehnle@amd.com> Signed-off-by:
-
Tom St Denis authored
A binary entry that lists GCA configuration data (and can be read by umr). (v2) Use kmalloc instead of vmalloc (v3) Minor indentation correction (v4) agd: Squash in kmalloc fix Signed-off-by:
-
Tom St Denis authored
Add ability to specify instance in select_se_sh callback. Defaults to 0xffffffff all over the driver. (v2) Don't enable INSTANCE_BROADCAST by default (v3) Style changes Signed-off-by:
-
Lyude authored
DRM_CONNECTOR_POLL_CONNECT only enables polling for connections, not disconnections. Because of this, we end up losing hotplug polling for analog connectors once they get connected. Easy way to reproduce: - Grab a machine with an AMD GPU and a VGA port - Plug a monitor into the VGA port, wait for it to update the connector from disconnected to connected - Disconnect the monitor on VGA, a hotplug event is never sent for the removal of the connector. Originally, only using DRM_CONNECTOR_POLL_CONNECT might have been a good idea since doing VGA polling can sometimes result in having to mess with the DAC voltages to figure out whether or not there's actually something there since VGA doesn't have HPD. Doing this would have the potential of showing visible artifacts on the screen every time we ran a poll while a VGA display was connected. Luckily, amdgpu_vga_detect() only resorts to this sort of polling if the poll is forced, and DRM's polling helper doesn't force it's polls. Additionally, this removes some assignments to connector->polled that weren't actually doing anything. Cc: stable@vger.kernel.org Signed-off-by:
Lyude <cpaul@redhat.com> Signed-off-by:
-
Lyude authored
DRM_CONNECTOR_POLL_CONNECT only enables polling for connections, not disconnections. Because of this, we end up losing hotplug polling for analog connectors once they get connected. Easy way to reproduce: - Grab a machine with a radeon GPU and a VGA port - Plug a monitor into the VGA port, wait for it to update the connector from disconnected to connected - Disconnect the monitor on VGA, a hotplug event is never sent for the removal of the connector. Originally, only using DRM_CONNECTOR_POLL_CONNECT might have been a good idea since doing VGA polling can sometimes result in having to mess with the DAC voltages to figure out whether or not there's actually something there since VGA doesn't have HPD. Doing this would have the potential of showing visible artifacts on the screen every time we ran a poll while a VGA display was connected. Luckily, radeon_vga_detect() only resorts to this sort of polling if the poll is forced, and DRM's polling helper doesn't force it's polls. Additionally, this removes some assignments to connector->polled that weren't actually doing anything. Cc: stable@vger.kernel.org Signed-off-by:
-
Chunming Zhou authored
V2: Add wait_for_mc_idle after stopping fb access V3: 1. Remove resume fb access since there is no need to do that for gpu reset. 2. Move stop fb access to amdgpu_gpu_reset function, since it's the same for all asics. Reviewed-by:
-
Rex Zhu authored
Signed-off-by:
Rex Zhu <Rex.Zhu@amd.com> Signed-off-by:
-
Rex Zhu authored
In commit 195567e9, use true/false instead of 1/0 to fix build warning. But the original logic: '0' means true and '1' means false. Reviewed-by:
-
Christian König authored
A regular spin_lock/unlock should do here as well. Signed-off-by:
-
Chunming Zhou authored
Record the gpu reset count in vmid to identify if gpu reset happened. Signed-off-by:
-
Chunming Zhou authored
Signed-off-by:
-
Chunming Zhou authored
Record the gpu reset count in vmid to identify if gpu reset happened. Reviewed-by:
-
Christian König authored
We want to keep the newest fence, not the oldest one. Reviewed-by:
-
Alexandre Demers authored
Signed-off-by:
Alexandre Demers <alexandre.f.demers@gmail.com> Signed-off-by:
-
Alexandre Demers authored
Signed-off-by:
-
Frank Binns authored
This flag was being set unconditionally at runtime so just set it at compile time instead. Reviewed-by:
Frank Binns <frank.binns@imgtec.com> Signed-off-by:
-
Michel Dänzer authored
If reservation_object_get_fences_rcu failed, we'd previously go directly to the cleanup label, so we'd leave the BO pinned. While we're at it, remove two amdgpu_bo_unreserve calls in favour of two new labels. Reviewed-by:
Michel Dänzer <michel.daenzer@amd.com> Signed-off-by:
-
Alex Deucher authored
Reviewed-by:
-
Alex Deucher authored
Reviewed-by:
-
Junwei Zhang authored
Fixed mc stop and resume hardware programming sequence. Signed-off-by:
Junwei Zhang <Jerry.Zhang@amd.com> Reviewed-by:
-
Alex Deucher authored
Skip IP blocks that are not enabled. Reviewed-by:
-
Alex Deucher authored
Saves a few CPU cycles. Reviewed-by:
-
Alex Deucher authored
Helpers to to call the IP functions for the selected IP. Reviewed-by:
-
Alex Deucher authored
Not enabled yet, but missing the call. Signed-off-by: -
Alex Deucher authored
Use UPDATE_IMMEDIATE (update on next data request boundary) rather than UPDATE_H_RETRACE (update on next line boundary). The data request boundary is less than a scanline, so it update will happen sooner. Cc: Vitaly Prosyak <vitaly.prosyak@amd.com> Signed-off-by:
-
