Proposal to use netlink for RAS and Telemetry across drm subsystem
Subsystems (3)
v5 cover Proposal to use netlink for RAS and Telemetry across drm subsystem
Revisiting this patch series to address pending feedback and help move the discussion towards a conclusion. This revision includes updates based on previous comments[1] and aims to clarify outstanding concerns. Specifically added command to facility reporting errors from IP blocks to support AMDGPU driver model of RAS.
[1]: https://lore.kernel.org/all/[email protected]/
I sincerely appreciate everyones patience and thoughtful reviews so far, and I hope this refreshed series facilitates the final evaluation and acceptance.
Please feel free to share any further suggestions or questions.
Thank you for your continued consideration.
----------------------------------------------------------------------
Our hardware supports RAS(Reliability, Availability, Serviceability) by reporting the errors to the host, which the KMD processes and exposes a set of error counters which can be used by observability tools to take corrective actions or repairs. Traditionally there were being exposed via PMU (for relative counters) and sysfs interface (for absolute value) in our internal branch. But, due to the limitations in this approach to use two interfaces and also not able to have an event based reporting or configurability, an alternative approach to try netlink was suggested by community for drm subsystem wide UAPI for RAS and telemetry as discussed in [2].
This [2] is the inspiration to this series. It uses the generic netlink(genl) family subsystem and exposes a set of commands that can be used by every drm driver, the framework provides a means to have custom commands too. Each drm driver instance in this example xe driver instance registers a family and operations to the genl subsystem through which it enumerates and reports the error counters. An event based notification is also supported to which userpace can subscribe to and be notified when any error occurs and read the error counter this avoids continuous polling on error counter. This can also be extended to threshold based notification.
[2]: https://airlied.blogspot.com/2022/09/accelerators-bof-outcomes-summary.html
This series is on top [3] series which introduces error counting infra in Xe driver.
[3]: https://lore.kernel.org/all/[email protected]/
V5:
Add support to read error corresponding to an IP BLOCK
v4:
1. Rebase
2. rename drm_genl_send to drm_genl_reply
3. catch error from xa_store and handle appropriately
4. presently xe_list_errors fills blank data for IGFX, prevent it by having an early check of IS_DGFX (Michael J. Ruhl)
v3:
1. Rebase on latest RAS series for XE
2. drop DRIVER_NETLINK flag and use the driver_genl_ops structure to register to netlink subsystem
v2: define common interfaces to genl netlink subsystem that all drm drivers can leverage.
Below is an example tool drm_ras which demonstrates the use of the
supported commands. The tool will be sent to ML with the subject
"[RFC i-g-t v3 0/1] A tool to demonstrate use of netlink sockets to read RAS error counters"
https://lore.kernel.org/all/[email protected]/
read single error counter:
$ ./drm_ras READ_ONE --device=drm:/dev/dri/card1 --error_id=0x0000000000000005 counter value 0
read all error counters:
$ ./drm_ras READ_ALL --device=drm:/dev/dri/card1 name config-id counter
error-gt0-correctable-guc 0x0000000000000001 0 error-gt0-correctable-slm 0x0000000000000003 0 error-gt0-correctable-eu-ic 0x0000000000000004 0 error-gt0-correctable-eu-grf 0x0000000000000005 0 error-gt0-fatal-guc 0x0000000000000009 0 error-gt0-fatal-slm 0x000000000000000d 0 error-gt0-fatal-eu-grf 0x000000000000000f 0 error-gt0-fatal-fpu 0x0000000000000010 0 error-gt0-fatal-tlb 0x0000000000000011 0 error-gt0-fatal-l3-fabric 0x0000000000000012 0 error-gt0-correctable-subslice 0x0000000000000013 0 error-gt0-correctable-l3bank 0x0000000000000014 0 error-gt0-fatal-subslice 0x0000000000000015 0 error-gt0-fatal-l3bank 0x0000000000000016 0 error-gt0-sgunit-correctable 0x0000000000000017 0 error-gt0-sgunit-nonfatal 0x0000000000000018 0 error-gt0-sgunit-fatal 0x0000000000000019 0 error-gt0-soc-fatal-psf-csc-0 0x000000000000001a 0 error-gt0-soc-fatal-psf-csc-1 0x000000000000001b 0 error-gt0-soc-fatal-psf-csc-2 0x000000000000001c 0 error-gt0-soc-fatal-punit 0x000000000000001d 0 error-gt0-soc-fatal-psf-0 0x000000000000001e 0 error-gt0-soc-fatal-psf-1 0x000000000000001f 0 error-gt0-soc-fatal-psf-2 0x0000000000000020 0 error-gt0-soc-fatal-cd0 0x0000000000000021 0 error-gt0-soc-fatal-cd0-mdfi 0x0000000000000022 0 error-gt0-soc-fatal-mdfi-east 0x0000000000000023 0 error-gt0-soc-fatal-mdfi-south 0x0000000000000024 0 error-gt0-soc-fatal-hbm-ss0-0 0x0000000000000025 0 error-gt0-soc-fatal-hbm-ss0-1 0x0000000000000026 0 error-gt0-soc-fatal-hbm-ss0-2 0x0000000000000027 0 error-gt0-soc-fatal-hbm-ss0-3 0x0000000000000028 0 error-gt0-soc-fatal-hbm-ss0-4 0x0000000000000029 0 error-gt0-soc-fatal-hbm-ss0-5 0x000000000000002a 0 error-gt0-soc-fatal-hbm-ss0-6 0x000000000000002b 0 error-gt0-soc-fatal-hbm-ss0-7 0x000000000000002c 0 error-gt0-soc-fatal-hbm-ss1-0 0x000000000000002d 0 error-gt0-soc-fatal-hbm-ss1-1 0x000000000000002e 0 error-gt0-soc-fatal-hbm-ss1-2 0x000000000000002f 0 error-gt0-soc-fatal-hbm-ss1-3 0x0000000000000030 0 error-gt0-soc-fatal-hbm-ss1-4 0x0000000000000031 0 error-gt0-soc-fatal-hbm-ss1-5 0x0000000000000032 0 error-gt0-soc-fatal-hbm-ss1-6 0x0000000000000033 0 error-gt0-soc-fatal-hbm-ss1-7 0x0000000000000034 0 error-gt0-soc-fatal-hbm-ss2-0 0x0000000000000035 0 error-gt0-soc-fatal-hbm-ss2-1 0x0000000000000036 0 error-gt0-soc-fatal-hbm-ss2-2 0x0000000000000037 0 error-gt0-soc-fatal-hbm-ss2-3 0x0000000000000038 0 error-gt0-soc-fatal-hbm-ss2-4 0x0000000000000039 0 error-gt0-soc-fatal-hbm-ss2-5 0x000000000000003a 0 error-gt0-soc-fatal-hbm-ss2-6 0x000000000000003b 0 error-gt0-soc-fatal-hbm-ss2-7 0x000000000000003c 0 error-gt0-soc-fatal-hbm-ss3-0 0x000000000000003d 0 error-gt0-soc-fatal-hbm-ss3-1 0x000000000000003e 0 error-gt0-soc-fatal-hbm-ss3-2 0x000000000000003f 0 error-gt0-soc-fatal-hbm-ss3-3 0x0000000000000040 0 error-gt0-soc-fatal-hbm-ss3-4 0x0000000000000041 0 error-gt0-soc-fatal-hbm-ss3-5 0x0000000000000042 0 error-gt0-soc-fatal-hbm-ss3-6 0x0000000000000043 0 error-gt0-soc-fatal-hbm-ss3-7 0x0000000000000044 0 error-gt0-gsc-correctable-sram-ecc 0x0000000000000045 0 error-gt0-gsc-nonfatal-mia-shutdown 0x0000000000000046 0 error-gt0-gsc-nonfatal-mia-int 0x0000000000000047 0 error-gt0-gsc-nonfatal-sram-ecc 0x0000000000000048 0 error-gt0-gsc-nonfatal-wdg-timeout 0x0000000000000049 0 error-gt0-gsc-nonfatal-rom-parity 0x000000000000004a 0 error-gt0-gsc-nonfatal-ucode-parity 0x000000000000004b 0 error-gt0-gsc-nonfatal-glitch-det 0x000000000000004c 0 error-gt0-gsc-nonfatal-fuse-pull 0x000000000000004d 0 error-gt0-gsc-nonfatal-fuse-crc-check 0x000000000000004e 0 error-gt0-gsc-nonfatal-selfmbist 0x000000000000004f 0 error-gt0-gsc-nonfatal-aon-parity 0x0000000000000050 0 error-gt1-correctable-guc 0x1000000000000001 0 error-gt1-correctable-slm 0x1000000000000003 0 error-gt1-correctable-eu-ic 0x1000000000000004 0 error-gt1-correctable-eu-grf 0x1000000000000005 0 error-gt1-fatal-guc 0x1000000000000009 0 error-gt1-fatal-slm 0x100000000000000d 0 error-gt1-fatal-eu-grf 0x100000000000000f 0 error-gt1-fatal-fpu 0x1000000000000010 0 error-gt1-fatal-tlb 0x1000000000000011 0 error-gt1-fatal-l3-fabric 0x1000000000000012 0 error-gt1-correctable-subslice 0x1000000000000013 0 error-gt1-correctable-l3bank 0x1000000000000014 0 error-gt1-fatal-subslice 0x1000000000000015 0 error-gt1-fatal-l3bank 0x1000000000000016 0 error-gt1-sgunit-correctable 0x1000000000000017 0 error-gt1-sgunit-nonfatal 0x1000000000000018 0 error-gt1-sgunit-fatal 0x1000000000000019 0 error-gt1-soc-fatal-psf-csc-0 0x100000000000001a 0 error-gt1-soc-fatal-psf-csc-1 0x100000000000001b 0 error-gt1-soc-fatal-psf-csc-2 0x100000000000001c 0 error-gt1-soc-fatal-punit 0x100000000000001d 0 error-gt1-soc-fatal-psf-0 0x100000000000001e 0 error-gt1-soc-fatal-psf-1 0x100000000000001f 0 error-gt1-soc-fatal-psf-2 0x1000000000000020 0 error-gt1-soc-fatal-cd0 0x1000000000000021 0 error-gt1-soc-fatal-cd0-mdfi 0x1000000000000022 0 error-gt1-soc-fatal-mdfi-east 0x1000000000000023 0 error-gt1-soc-fatal-mdfi-south 0x1000000000000024 0 error-gt1-soc-fatal-hbm-ss0-0 0x1000000000000025 0 error-gt1-soc-fatal-hbm-ss0-1 0x1000000000000026 0 error-gt1-soc-fatal-hbm-ss0-2 0x1000000000000027 0 error-gt1-soc-fatal-hbm-ss0-3 0x1000000000000028 0 error-gt1-soc-fatal-hbm-ss0-4 0x1000000000000029 0 error-gt1-soc-fatal-hbm-ss0-5 0x100000000000002a 0 error-gt1-soc-fatal-hbm-ss0-6 0x100000000000002b 0 error-gt1-soc-fatal-hbm-ss0-7 0x100000000000002c 0 error-gt1-soc-fatal-hbm-ss1-0 0x100000000000002d 0 error-gt1-soc-fatal-hbm-ss1-1 0x100000000000002e 0 error-gt1-soc-fatal-hbm-ss1-2 0x100000000000002f 0 error-gt1-soc-fatal-hbm-ss1-3 0x1000000000000030 0 error-gt1-soc-fatal-hbm-ss1-4 0x1000000000000031 0 error-gt1-soc-fatal-hbm-ss1-5 0x1000000000000032 0 error-gt1-soc-fatal-hbm-ss1-6 0x1000000000000033 0 error-gt1-soc-fatal-hbm-ss1-7 0x1000000000000034 0 error-gt1-soc-fatal-hbm-ss2-0 0x1000000000000035 0 error-gt1-soc-fatal-hbm-ss2-1 0x1000000000000036 0 error-gt1-soc-fatal-hbm-ss2-2 0x1000000000000037 0 error-gt1-soc-fatal-hbm-ss2-3 0x1000000000000038 0 error-gt1-soc-fatal-hbm-ss2-4 0x1000000000000039 0 error-gt1-soc-fatal-hbm-ss2-5 0x100000000000003a 0 error-gt1-soc-fatal-hbm-ss2-6 0x100000000000003b 0 error-gt1-soc-fatal-hbm-ss2-7 0x100000000000003c 0 error-gt1-soc-fatal-hbm-ss3-0 0x100000000000003d 0 error-gt1-soc-fatal-hbm-ss3-1 0x100000000000003e 0 error-gt1-soc-fatal-hbm-ss3-2 0x100000000000003f 0 error-gt1-soc-fatal-hbm-ss3-3 0x1000000000000040 0 error-gt1-soc-fatal-hbm-ss3-4 0x1000000000000041 0 error-gt1-soc-fatal-hbm-ss3-5 0x1000000000000042 0 error-gt1-soc-fatal-hbm-ss3-6 0x1000000000000043 0 error-gt1-soc-fatal-hbm-ss3-7 0x1000000000000044 0
wait on a error event:
$ ./drm_ras WAIT_ON_EVENT --device=drm:/dev/dri/card1 waiting for error event error event received counter value 0
list all errors:
$ ./drm_ras LIST_ERRORS --device=drm:/dev/dri/card1 name config-id
error-gt0-correctable-guc 0x0000000000000001 error-gt0-correctable-slm 0x0000000000000003 error-gt0-correctable-eu-ic 0x0000000000000004 error-gt0-correctable-eu-grf 0x0000000000000005 error-gt0-fatal-guc 0x0000000000000009 error-gt0-fatal-slm 0x000000000000000d error-gt0-fatal-eu-grf 0x000000000000000f error-gt0-fatal-fpu 0x0000000000000010 error-gt0-fatal-tlb 0x0000000000000011 error-gt0-fatal-l3-fabric 0x0000000000000012 error-gt0-correctable-subslice 0x0000000000000013 error-gt0-correctable-l3bank 0x0000000000000014 error-gt0-fatal-subslice 0x0000000000000015 error-gt0-fatal-l3bank 0x0000000000000016 error-gt0-sgunit-correctable 0x0000000000000017 error-gt0-sgunit-nonfatal 0x0000000000000018 error-gt0-sgunit-fatal 0x0000000000000019 error-gt0-soc-fatal-psf-csc-0 0x000000000000001a error-gt0-soc-fatal-psf-csc-1 0x000000000000001b error-gt0-soc-fatal-psf-csc-2 0x000000000000001c error-gt0-soc-fatal-punit 0x000000000000001d error-gt0-soc-fatal-psf-0 0x000000000000001e error-gt0-soc-fatal-psf-1 0x000000000000001f error-gt0-soc-fatal-psf-2 0x0000000000000020 error-gt0-soc-fatal-cd0 0x0000000000000021 error-gt0-soc-fatal-cd0-mdfi 0x0000000000000022 error-gt0-soc-fatal-mdfi-east 0x0000000000000023 error-gt0-soc-fatal-mdfi-south 0x0000000000000024 error-gt0-soc-fatal-hbm-ss0-0 0x0000000000000025 error-gt0-soc-fatal-hbm-ss0-1 0x0000000000000026 error-gt0-soc-fatal-hbm-ss0-2 0x0000000000000027 error-gt0-soc-fatal-hbm-ss0-3 0x0000000000000028 error-gt0-soc-fatal-hbm-ss0-4 0x0000000000000029 error-gt0-soc-fatal-hbm-ss0-5 0x000000000000002a error-gt0-soc-fatal-hbm-ss0-6 0x000000000000002b error-gt0-soc-fatal-hbm-ss0-7 0x000000000000002c error-gt0-soc-fatal-hbm-ss1-0 0x000000000000002d error-gt0-soc-fatal-hbm-ss1-1 0x000000000000002e error-gt0-soc-fatal-hbm-ss1-2 0x000000000000002f error-gt0-soc-fatal-hbm-ss1-3 0x0000000000000030 error-gt0-soc-fatal-hbm-ss1-4 0x0000000000000031 error-gt0-soc-fatal-hbm-ss1-5 0x0000000000000032 error-gt0-soc-fatal-hbm-ss1-6 0x0000000000000033 error-gt0-soc-fatal-hbm-ss1-7 0x0000000000000034 error-gt0-soc-fatal-hbm-ss2-0 0x0000000000000035 error-gt0-soc-fatal-hbm-ss2-1 0x0000000000000036 error-gt0-soc-fatal-hbm-ss2-2 0x0000000000000037 error-gt0-soc-fatal-hbm-ss2-3 0x0000000000000038 error-gt0-soc-fatal-hbm-ss2-4 0x0000000000000039 error-gt0-soc-fatal-hbm-ss2-5 0x000000000000003a error-gt0-soc-fatal-hbm-ss2-6 0x000000000000003b error-gt0-soc-fatal-hbm-ss2-7 0x000000000000003c error-gt0-soc-fatal-hbm-ss3-0 0x000000000000003d error-gt0-soc-fatal-hbm-ss3-1 0x000000000000003e error-gt0-soc-fatal-hbm-ss3-2 0x000000000000003f error-gt0-soc-fatal-hbm-ss3-3 0x0000000000000040 error-gt0-soc-fatal-hbm-ss3-4 0x0000000000000041 error-gt0-soc-fatal-hbm-ss3-5 0x0000000000000042 error-gt0-soc-fatal-hbm-ss3-6 0x0000000000000043 error-gt0-soc-fatal-hbm-ss3-7 0x0000000000000044 error-gt0-gsc-correctable-sram-ecc 0x0000000000000045 error-gt0-gsc-nonfatal-mia-shutdown 0x0000000000000046 error-gt0-gsc-nonfatal-mia-int 0x0000000000000047 error-gt0-gsc-nonfatal-sram-ecc 0x0000000000000048 error-gt0-gsc-nonfatal-wdg-timeout 0x0000000000000049 error-gt0-gsc-nonfatal-rom-parity 0x000000000000004a error-gt0-gsc-nonfatal-ucode-parity 0x000000000000004b error-gt0-gsc-nonfatal-glitch-det 0x000000000000004c error-gt0-gsc-nonfatal-fuse-pull 0x000000000000004d error-gt0-gsc-nonfatal-fuse-crc-check 0x000000000000004e error-gt0-gsc-nonfatal-selfmbist 0x000000000000004f error-gt0-gsc-nonfatal-aon-parity 0x0000000000000050 error-gt1-correctable-guc 0x1000000000000001 error-gt1-correctable-slm 0x1000000000000003 error-gt1-correctable-eu-ic 0x1000000000000004 error-gt1-correctable-eu-grf 0x1000000000000005 error-gt1-fatal-guc 0x1000000000000009 error-gt1-fatal-slm 0x100000000000000d error-gt1-fatal-eu-grf 0x100000000000000f error-gt1-fatal-fpu 0x1000000000000010 error-gt1-fatal-tlb 0x1000000000000011 error-gt1-fatal-l3-fabric 0x1000000000000012 error-gt1-correctable-subslice 0x1000000000000013 error-gt1-correctable-l3bank 0x1000000000000014 error-gt1-fatal-subslice 0x1000000000000015 error-gt1-fatal-l3bank 0x1000000000000016 error-gt1-sgunit-correctable 0x1000000000000017 error-gt1-sgunit-nonfatal 0x1000000000000018 error-gt1-sgunit-fatal 0x1000000000000019 error-gt1-soc-fatal-psf-csc-0 0x100000000000001a error-gt1-soc-fatal-psf-csc-1 0x100000000000001b error-gt1-soc-fatal-psf-csc-2 0x100000000000001c error-gt1-soc-fatal-punit 0x100000000000001d error-gt1-soc-fatal-psf-0 0x100000000000001e error-gt1-soc-fatal-psf-1 0x100000000000001f error-gt1-soc-fatal-psf-2 0x1000000000000020 error-gt1-soc-fatal-cd0 0x1000000000000021 error-gt1-soc-fatal-cd0-mdfi 0x1000000000000022 error-gt1-soc-fatal-mdfi-east 0x1000000000000023 error-gt1-soc-fatal-mdfi-south 0x1000000000000024 error-gt1-soc-fatal-hbm-ss0-0 0x1000000000000025 error-gt1-soc-fatal-hbm-ss0-1 0x1000000000000026 error-gt1-soc-fatal-hbm-ss0-2 0x1000000000000027 error-gt1-soc-fatal-hbm-ss0-3 0x1000000000000028 error-gt1-soc-fatal-hbm-ss0-4 0x1000000000000029 error-gt1-soc-fatal-hbm-ss0-5 0x100000000000002a error-gt1-soc-fatal-hbm-ss0-6 0x100000000000002b error-gt1-soc-fatal-hbm-ss0-7 0x100000000000002c error-gt1-soc-fatal-hbm-ss1-0 0x100000000000002d error-gt1-soc-fatal-hbm-ss1-1 0x100000000000002e error-gt1-soc-fatal-hbm-ss1-2 0x100000000000002f error-gt1-soc-fatal-hbm-ss1-3 0x1000000000000030 error-gt1-soc-fatal-hbm-ss1-4 0x1000000000000031 error-gt1-soc-fatal-hbm-ss1-5 0x1000000000000032 error-gt1-soc-fatal-hbm-ss1-6 0x1000000000000033 error-gt1-soc-fatal-hbm-ss1-7 0x1000000000000034 error-gt1-soc-fatal-hbm-ss2-0 0x1000000000000035 error-gt1-soc-fatal-hbm-ss2-1 0x1000000000000036 error-gt1-soc-fatal-hbm-ss2-2 0x1000000000000037 error-gt1-soc-fatal-hbm-ss2-3 0x1000000000000038 error-gt1-soc-fatal-hbm-ss2-4 0x1000000000000039 error-gt1-soc-fatal-hbm-ss2-5 0x100000000000003a error-gt1-soc-fatal-hbm-ss2-6 0x100000000000003b error-gt1-soc-fatal-hbm-ss2-7 0x100000000000003c error-gt1-soc-fatal-hbm-ss3-0 0x100000000000003d error-gt1-soc-fatal-hbm-ss3-1 0x100000000000003e error-gt1-soc-fatal-hbm-ss3-2 0x100000000000003f error-gt1-soc-fatal-hbm-ss3-3 0x1000000000000040 error-gt1-soc-fatal-hbm-ss3-4 0x1000000000000041 error-gt1-soc-fatal-hbm-ss3-5 0x1000000000000042 error-gt1-soc-fatal-hbm-ss3-6 0x1000000000000043 error-gt1-soc-fatal-hbm-ss3-7 0x1000000000000044
Aravind Iddamsetty (5):
drm/netlink: Add netlink infrastructure drm/xe/RAS: Register netlink capability drm/xe/RAS: Expose the error counters drm/netlink: Define multicast groups
drm/xe/RAS: send multicast event on occurrence of an error
drivers/gpu/drm/Makefile | 1 + drivers/gpu/drm/drm_drv.c | 7 + drivers/gpu/drm/drm_netlink.c | 219 +++++++++++ drivers/gpu/drm/xe/Makefile | 2 + drivers/gpu/drm/xe/xe_device.c | 6 + drivers/gpu/drm/xe/xe_device_types.h | 1 + drivers/gpu/drm/xe/xe_hw_error.c | 56 ++- drivers/gpu/drm/xe/xe_netlink.c | 531 +++++++++++++++++++++++++++ include/drm/drm_device.h | 10 + include/drm/drm_drv.h | 7 + include/drm/drm_netlink.h | 46 +++ include/uapi/drm/drm_netlink.h | 105 ++++++ include/uapi/drm/xe_drm.h | 85 +++++ 13 files changed, 1071 insertions(+), 5 deletions(-) create mode 100644 drivers/gpu/drm/drm_netlink.c create mode 100644 drivers/gpu/drm/xe/xe_netlink.c create mode 100644 include/drm/drm_netlink.h create mode 100644 include/uapi/drm/drm_netlink.h
Signature · 1 line
2.25.1
RFC v5 1/5 drm/netlink: Add netlink infrastructure
Define the netlink registration interface and commands, attributes that can be commonly used across by drm drivers. This patch intends to use the generic netlink family to expose various stats of device. At present it defines some commands that shall be used to expose RAS error counters.
v2:
define common interfaces to genl netlink subsystem that all drm drivers can leverage.(Tomer Tayar)
v3: drop DRIVER_NETLINK flag and use the driver_genl_ops structure to register to netlink subsystem (Daniel Vetter)
v4:(Michael J. Ruhl)
1. rename drm_genl_send to drm_genl_reply
2. catch error from xa_store and handle appropriately
v5:
1. compile only if CONFIG_NET is enabled
V6: Add support for reading an IP block errors
--- drivers/gpu/drm/Makefile | 1 + drivers/gpu/drm/drm_drv.c | 7 ++ drivers/gpu/drm/drm_netlink.c | 212 +++++++++++++++++++++++++++++++++ include/drm/drm_device.h | 10 ++ include/drm/drm_drv.h | 7 ++ include/drm/drm_netlink.h | 41 +++++++ include/uapi/drm/drm_netlink.h | 101 ++++++++++++++++ 7 files changed, 379 insertions(+) create mode 100644 drivers/gpu/drm/drm_netlink.c create mode 100644 include/drm/drm_netlink.h create mode 100644 include/uapi/drm/drm_netlink.h
drivers/gpu/drm/Makefile +1 -0
@@ -77,6 +77,7 @@ drm-$(CONFIG_DRM_CLIENT) += \ drm_client.o \ drm_client_event.o \ drm_client_modeset.o +drm-$(CONFIG_NET) += drm_netlink.o drm-$(CONFIG_DRM_LIB_RANDOM) += lib/drm_random.o drm-$(CONFIG_COMPAT) += drm_ioc32.o drm-$(CONFIG_DRM_PANEL) += drm_panel.o
drivers/gpu/drm/drm_drv.c +7 -0
@@ -1088,6 +1088,12 @@ int drm_dev_register(struct drm_device *dev, unsigned long flags)if (ret) goto err_minors; +if (driver->genl_ops) { + ret =drm_genl_register (dev); +if (ret) + goto err_minors; + } + ret =create_compat_control_link (dev);if (ret) goto err_minors;
@@ -1229,6 +1235,7 @@ static void drm_core_exit(void)drm_privacy_screen_lookup_exit ();drm_panic_exit ();accel_core_exit (); +drm_genl_exit ();unregister_chrdev (DRM_MAJOR ,"drm" );debugfs_remove (drm_debugfs_root);drm_sysfs_destroy ();
drivers/gpu/drm/drm_netlink.c new file +212 -0
212 lines not on this page · raw message
include/drm/drm_device.h +10 -0
@@ -8,6 +8,7 @@#include <linux/sched.h> #include <drm/drm_mode_config.h> +#include <drm/drm_netlink.h> struct drm_driver ;struct drm_minor ;
@@ -22,6 +23,8 @@ struct inode;struct pci_dev ;struct pci_controller ; +struct genl_family ; +/* * Recovery methods for wedged device in order of less to more side-effects. * To be used with drm_dev_wedged_event() as recovery @method. Callers can
@@ -356,6 +359,13 @@ struct drm_device {* Root directory for debugfs files. */ struct dentry *debugfs_root; + +/** +* @drm_genl_family: +* +* Generic netlink family registration structure. +*/ +struct genl_family *drm_genl_family; };void drm_dev_set_dma_dev (struct drm_device *dev,struct device *dma_dev);
include/drm/drm_drv.h +7 -0
@@ -431,6 +431,13 @@ struct drm_driver {* some examples. */ const struct file_operations *fops; + +/** +* @genl_ops: +* +* Drivers private callback to genl commands +*/ +const struct driver_genl_ops *genl_ops; }; void *__devm_drm_dev_alloc(struct device *parent,
include/drm/drm_netlink.h new file +41 -0
41 lines not on this page · raw message
include/uapi/drm/drm_netlink.h new file +101 -0
101 lines not on this page · raw message
Signature · 1 line
2.25.1
Re: [RFC v5 1/5] drm/netlink: Add netlink infrastructure
Quoted 5 lines · Aravind Iddamsetty
drivers/gpu/drm/drm_netlink.c new file +5
-0+static void drm_genl_family_init (struct drm_device *dev) +{ + dev->drm_genl_family =drmm_kzalloc (dev,sizeof (struct genl_family ), +GFP_KERNEL ); +
drivers/gpu/drm/drm_netlink.c line 160
Quoted 10 lines
drivers/gpu/drm/drm_netlink.c new file +10
-0+/* Use drm primary node name eg: card0 to name the genl family */ +snprintf (dev->drm_genl_family->name,sizeof (dev->drm_genl_family->name), +"%s" , dev->primary->kdev->kobj.name); + dev->drm_genl_family->version =DRM_GENL_VERSION ; + dev->drm_genl_family->parallel_ops = true; + dev->drm_genl_family->ops = drm_genl_ops; + dev->drm_genl_family->n_ops =ARRAY_SIZE (drm_genl_ops); + dev->drm_genl_family->maxattr =DRM_ATTR_MAX ; + dev->drm_genl_family->module = dev->dev->driver->owner; +}
We are interested in using this infrastructure at Qualcomm to communicate telemetry information for the AI100 accelerators. It would be nice if this function could support drm_minor accel nodes (dev->accel) as well.
Thanks,
Zack
Re: [RFC v5 1/5] drm/netlink: Add netlink infrastructure
Quoted 15 lines · Zack McKevitt
Quoted 14 lines · Aravind Iddamsetty
drivers/gpu/drm/drm_netlink.c new file +3
-0+static void drm_genl_family_init (struct drm_device *dev) +{ + dev->drm_genl_family = drmm_kzalloc(dev, sizeof(struct
genl_family),
+GFP_KERNEL);
+
+ /* Use drm primary node name eg: card0 to name the genl family */ + snprintf(dev->drm_genl_family->name,sizeof(dev->drm_genl_family->name),
+ "%s",dev->primary->kdev->kobj.name);
+dev->drm_genl_family->version=DRM_GENL_VERSION;
+dev->drm_genl_family->parallel_ops= true;
+dev->drm_genl_family->ops=drm_genl_ops;
+dev->drm_genl_family->n_ops=ARRAY_SIZE(drm_genl_ops);
Quoted 8 lines
Quoted 3 lines
+
dev->drm_genl_family->maxattr=DRM_ATTR_MAX; +dev->drm_genl_family->module=dev->dev->driver->owner; +}We are interested in using this infrastructure at Qualcomm to communicate telemetry information for the AI100 accelerators. It would be nice if this function could support
drm_minoraccel nodes (dev->accel) as well.
Glad to know the interest, at present the code does create drm netlink family for accel device as well, but it is tries to register with the drm primary node name which might not be present for dev->accel, checking for the "DRIVER_COMPUTE_ACCEL" and registering with the name will fix that.
But also to bring to your attention the current series focuses on reporting RAS errors and hence the commands are as well oriented towards errors, anything specific to telemetry needs to be added. Do you have any thought as to what kind of data and what kind of command support you might need.
Thanks,
Aravind.
Quoted 4 lines
Thanks,
Zack
Re: [RFC v5 1/5] drm/netlink: Add netlink infrastructure
Quoted 5 lines · Aravind Iddamsetty
Glad to know the interest, at present the code does create drm netlink family for accel device as well, but it is tries to register with the drm primary node name which might not be present for
dev->accel, checking for the "DRIVER_COMPUTE_ACCEL" and registering with the name will fix that.
This is correct, trying to access dev->primary->kdev->kobj.name through an accel device will cause a fault. I believe your solution will work, and the node name can instead be retrieved via dev->accel->kdev->kobj.name.
Quoted 5 lines
But also to bring to your attention the current series focuses on reporting RAS errors and hence the commands are as well oriented towards errors, anything specific to telemetry needs to be added. Do you have any thought as to what kind of data and what kind of command support you might need.
Understood. We will likely be interested in RAS functionality in the future but thought this would be a good avenue for telemetry as well since our device currently has a functional RAS implementation.
An early prototype for this might add a new command for telemetry and a new policy with 4 new attributes: the type of telemetry to read/write (e.g., SoC temp), whether the request is a read or write from/to the device, the status of the request, and the telemetry value read or written. As actual telemetry fields are likely device specific, these can be defined in the driver's uapi header and passed opaquely through the netlink interface.
The above description was sufficient for an initial prototype on top of our driver. Mostly, however, we want to reiterate our interest in these changes and will keep an eye out for a future patchset that incorporates generated boilerplate from a YAML description.
Thanks again,
Zack
Quoted 2 lines
Thanks,
Aravind.
Rodrigo Vivi
maintainerRe: [RFC v5 1/5] drm/netlink: Add netlink infrastructure
Quoted 29 lines · Zack McKevitt
Quoted 5 lines · Aravind Iddamsetty
Glad to know the interest, at present the code does create drm netlink family for accel device as well, but it is tries to register with the drm primary node name which might not be present for
dev->accel, checking for the "DRIVER_COMPUTE_ACCEL" and registering with the name will fix that.This is correct, trying to access
dev->primary->kdev->kobj.namethrough an accel device will cause a fault. I believe your solution will work, and the node name can instead be retrieved viadev->accel->kdev->kobj.name.Quoted 5 lines
But also to bring to your attention the current series focuses on reporting RAS errors and hence the commands are as well oriented towards errors, anything specific to telemetry needs to be added. Do you have any thought as to what kind of data and what kind of command support you might need.
Understood. We will likely be interested in RAS functionality in the future but thought this would be a good avenue for telemetry as well since our device currently has a functional RAS implementation.
An early prototype for this might add a new command for telemetry and a new policy with 4 new attributes: the type of telemetry to read/write (e.g., SoC temp), whether the request is a read or write from/to the device, the status of the request, and the telemetry value read or written. As actual telemetry fields are likely device specific, these can be defined in the driver's uapi header and passed opaquely through the netlink interface.
Quoted 6 lines
The above description was sufficient for an initial prototype on top of our driver. Mostly, however, we want to reiterate our interest in these changes and will keep an eye out for a future patchset that incorporates generated boilerplate from a YAML description.
Thanks again,
Hi Zack,
thanks for your interest,
could you please take a look in the new version and help with reviews and feedback there to see if that is flexible enough to be later expanded to your needs:
https://lore.kernel.org/dri-devel/[email protected]/
Thanks,
Rodrigo.
Quoted 5 lines
Zack
Quoted 2 lines
Thanks,
Aravind.
Rodrigo Vivi
maintainerRe: [RFC v5 1/5] drm/netlink: Add netlink infrastructure
In reply to Aravind IddamsettyQuoted 53 lines · Aravind Iddamsetty
Define the netlink registration interface and commands, attributes that can be commonly used across by drm drivers. This patch intends to use the generic netlink family to expose various stats of device. At present it defines some commands that shall be used to expose RAS error counters.
v2:
define common interfaces to genl netlink subsystem that all drm drivers can leverage.(Tomer Tayar)v3: drop
DRIVER_NETLINKflag and use thedriver_genl_opsstructure to register to netlink subsystem (Daniel Vetter)v4:(Michael J. Ruhl)
1. renamedrm_genl_sendtodrm_genl_reply
2. catch error fromxa_storeand handle appropriatelyv5:
1. compile only ifCONFIG_NETis enabledV6: Add support for reading an IP block errors
Reviewed-by Michael J. Ruhl <[email protected]> #v4Signed-off-by Aravind Iddamsetty <[email protected]>--- drivers/gpu/drm/Makefile | 1 + drivers/gpu/drm/drm_drv.c | 7 ++ drivers/gpu/drm/drm_netlink.c | 212 +++++++++++++++++++++++++++++++++ include/drm/drm_device.h | 10 ++ include/drm/drm_drv.h | 7 ++ include/drm/drm_netlink.h | 41 +++++++ include/uapi/drm/drm_netlink.h | 101 ++++++++++++++++ 7 files changed, 379 insertions(+) create mode 100644 drivers/gpu/drm/drm_netlink.c create mode 100644 include/drm/drm_netlink.h create mode 100644 include/uapi/drm/drm_netlink.hdrivers/gpu/drm/Makefile +1
-0@@ -77,6 +77,7 @@ drm-$(CONFIG_DRM_CLIENT) += \ drm_client.o \ drm_client_event.o \ drm_client_modeset.o +drm-$(CONFIG_NET) += drm_netlink.o drm-$(CONFIG_DRM_LIB_RANDOM) += lib/drm_random.o drm-$(CONFIG_COMPAT) += drm_ioc32.o drm-$(CONFIG_DRM_PANEL) += drm_panel.odrivers/gpu/drm/drm_drv.c +0
-0
drivers/gpu/drm/drm_drv.c int drm_dev_register(struct drm_device *dev, unsigned long flags) · line 1095
Quoted 8 lines
drivers/gpu/drm/drm_drv.c +5
-0@@ -1088,6 +1088,12 @@ int drm_dev_register(struct drm_device *dev, unsigned long flags)if (ret) goto err_minors; +if (driver->genl_ops) { + ret =drm_genl_register (dev); +if (ret) + goto err_minors; + }
Even if we don't go with multiple 'groups' I believe that the driver should explicitly call the netlink registration.
drivers/gpu/drm/drm_netlink.c
Quoted 17 lines
drivers/gpu/drm/drm_drv.c +2
-0+ ret =create_compat_control_link (dev);if (ret) goto err_minors;@@ -1229,6 +1235,7 @@ static void drm_core_exit(void)drm_privacy_screen_lookup_exit ();drm_panic_exit ();accel_core_exit (); +drm_genl_exit ();unregister_chrdev (DRM_MAJOR ,"drm" );debugfs_remove (drm_debugfs_root);drm_sysfs_destroy ();drivers/gpu/drm/drm_netlink.c new file +0
-0
drm_ras.c ?
drivers/gpu/drm/drm_netlink.c line 3
Quoted 4 lines
drivers/gpu/drm/drm_netlink.c new file +3
-0@@ -0,0 +1,212 @@ +// SPDX-License-Identifier: MIT +/* +* Copyright © 2023 Intel Corporation
2025 here and in any other file
drivers/gpu/drm/drm_netlink.c line 7
Quoted 4 lines
drivers/gpu/drm/drm_netlink.c new file +4
-0+*/ + +#include <net/genetlink.h> +#include <uapi/drm/drm_netlink.h>
uapi/drm/drm_ras.h ?!
like we don't have a drm_ioctl.h but drm_mode.h
Quoted 10 lines
drivers/gpu/drm/drm_netlink.c new file +10
-0+ +#include <drm/drm_device.h> +#include <drm/drm_drv.h> +#include <drm/drm_file.h> +#include <drm/drm_managed.h> +#include <drm/drm_netlink.h> +#include <drm/drm_print.h> + +DEFINE_XARRAY (drm_dev_xarray); +
drivers/gpu/drm/drm_netlink.c line 27
Quoted 10 lines
drivers/gpu/drm/drm_netlink.c new file +10
-0+/** +* drm_genl_reply - response to a request +* @msg: socket buffer +* @info: receiver information +* @usrhdr: pointer to user specific header in the message buffer +* +* RETURNS: +* 0 on success and negative error code on failure +*/ +int drm_genl_reply (struct sk_buff *msg,struct genl_info *info,void *usrhdr)
drm_ras_reply so we standardize in a single namespace everywhere ?!
and same for all other functions and structs, except for things that are declared outside drm
Quoted 116 lines
drivers/gpu/drm/drm_netlink.c new file +116
-0+{ +int ret; + +genlmsg_end (msg, usrhdr); + + ret =genlmsg_reply (msg, info); +if (ret) +nlmsg_free (msg); + +return ret; +} +EXPORT_SYMBOL (drm_genl_reply); + +/** +* drm_genl_alloc_msg - allocate genl message buffer +* @dev: drm_device for which the message is being allocated +* @info: receiver information +* @msg_size: size of the msg buffer that needs to be allocated +* @usrhdr: pointer to user specific header in the message buffer +* +* RETURNS: +* pointer to new allocated buffer on success, NULL on failure +*/ +struct sk_buff * +drm_genl_alloc_msg (struct drm_device *dev, +struct genl_info *info, +size_t msg_size,void **usrhdr) +{ +struct sk_buff *new_msg; + + new_msg =genlmsg_new (msg_size,GFP_KERNEL ); +if (!new_msg) +return new_msg; + + *usrhdr =genlmsg_put_reply (new_msg, info, dev->drm_genl_family,0 , info->genlhdr->cmd); +if (!*usrhdr) { +nlmsg_free (new_msg); + new_msg =NULL ; + } + +return new_msg; +} +EXPORT_SYMBOL (drm_genl_alloc_msg); + +static struct drm_device *genl_to_dev (struct genl_info *info) +{ +return xa_load (&drm_dev_xarray, info->nlhdr->nlmsg_type); +} + +static int drm_genl_list_errors (struct sk_buff *msg,struct genl_info *info) +{ +struct drm_device *dev =genl_to_dev (info); + +if (info->genlhdr->cmd ==DRM_RAS_CMD_READ_ALL ) { +if (GENL_REQ_ATTR_CHECK (info,DRM_RAS_ATTR_READ_ALL )) +return -EINVAL ; + }else { +if (GENL_REQ_ATTR_CHECK (info,DRM_RAS_ATTR_QUERY )) +return -EINVAL ; + } + +if (WARN_ON (!dev->driver->genl_ops[info->genlhdr->cmd].doit)) +return -EOPNOTSUPP ; + +return dev->driver->genl_ops[info->genlhdr->cmd].doit (dev, msg, info); +} + +static int drm_genl_read_error (struct sk_buff *msg,struct genl_info *info) +{ +struct drm_device *dev =genl_to_dev (info); + +if (GENL_REQ_ATTR_CHECK (info,DRM_RAS_ATTR_ERROR_ID )) +return -EINVAL ; + +if (WARN_ON (!dev->driver->genl_ops[info->genlhdr->cmd].doit)) +return -EOPNOTSUPP ; + +return dev->driver->genl_ops[info->genlhdr->cmd].doit (dev, msg, info); +} + +/* attribute policies */ +static const struct nla_policy drm_attr_policy_query[DRM_ATTR_MAX +1 ] = { + [DRM_RAS_ATTR_QUERY ] = { .type =NLA_U8 }, +}; + +static const struct nla_policy drm_attr_policy_read_one[DRM_ATTR_MAX +1 ] = { + [DRM_RAS_ATTR_ERROR_ID ] = { .type =NLA_U64 }, +}; + +static const struct nla_policy drm_attr_policy_read_all[DRM_ATTR_MAX +1 ] = { + [DRM_RAS_ATTR_READ_ALL ] = { .type =NLA_U8 }, +}; + +/* drm genl operations definition */ +const struct genl_ops drm_genl_ops[] = { + { + .cmd =DRM_RAS_CMD_QUERY , + .doit = drm_genl_list_errors, + .policy = drm_attr_policy_query, + }, + { + .cmd =DRM_RAS_CMD_READ_ONE , + .doit = drm_genl_read_error, + .policy = drm_attr_policy_read_one, + }, + { + .cmd =DRM_RAS_CMD_READ_ALL , + .doit = drm_genl_list_errors, + .policy = drm_attr_policy_read_all, + }, + { + .cmd =DRM_RAS_CMD_READ_BLOCK , + .doit = drm_genl_read_error, + .policy = drm_attr_policy_read_one, + }, +
drivers/gpu/drm/drm_netlink.c line 153
Quoted 10 lines
drivers/gpu/drm/drm_netlink.c new file +10
-0+}; + +static void drm_genl_family_init (struct drm_device *dev) +{ + dev->drm_genl_family =drmm_kzalloc (dev,sizeof (struct genl_family ), +GFP_KERNEL ); + +/* Use drm primary node name eg: card0 to name the genl family */ +snprintf (dev->drm_genl_family->name,sizeof (dev->drm_genl_family->name), +"%s" , dev->primary->kdev->kobj.name);
for the family name I believe we deserve the 'drmras', then the card minor number, then the group name.
For instance, but not necessarily suggesting xe to do it:
drmras-0-gt
drmras-0-soc
.....
driver can select their own name...
drivers/gpu/drm/drm_netlink.c line 154
Quoted 1 line
drivers/gpu/drm/drm_netlink.c new file +1
-0+ dev->drm_genl_family->version =DRM_GENL_VERSION ;
I believe driver could control their own version so if something changes in the group names for instance or supported commands they can change it.
Quoted 80 lines
drivers/gpu/drm/drm_netlink.c new file +58
-0+ dev->drm_genl_family->parallel_ops = true; + dev->drm_genl_family->ops = drm_genl_ops; + dev->drm_genl_family->n_ops =ARRAY_SIZE (drm_genl_ops); + dev->drm_genl_family->maxattr =DRM_ATTR_MAX ; + dev->drm_genl_family->module = dev->dev->driver->owner; +} + +static void drm_genl_deregister (struct drm_device *dev,void *arg) +{ +drm_dbg_driver (dev,"unregistering genl family %s\n" , dev->drm_genl_family->name); + +xa_erase (&drm_dev_xarray, dev->drm_genl_family->id); + +genl_unregister_family (dev->drm_genl_family); +} + +/** +* drm_genl_register - Register genl family +* @dev: drm_device for which genl family needs to be registered +* +* RETURNS: +* 0 on success and negative error code on failure +*/ +int drm_genl_register (struct drm_device *dev) +{ +int ret; + +drm_genl_family_init (dev); + + ret =genl_register_family (dev->drm_genl_family); +if (ret <0 ) { +drm_warn (dev,"genl family registration failed\n" ); +return ret; + } + +drm_dbg_driver (dev,"genl family id %d and name %s\n" , dev->drm_genl_family->id, + dev->drm_genl_family->name); + + ret =xa_err (xa_store (&drm_dev_xarray, dev->drm_genl_family->id, dev,GFP_KERNEL )); +if (ret) + goto genl_unregister; + + ret =drmm_add_action_or_reset (dev, drm_genl_deregister,NULL ); + +return ret; + +genl_unregister: +genl_unregister_family (dev->drm_genl_family); +return ret; +} + +/** +* drm_genl_exit: destroy drm_dev_xarray +*/ +void drm_genl_exit (void ) +{ +xa_destroy (&drm_dev_xarray); +}include/drm/drm_device.h +3
-0@@ -8,6 +8,7 @@#include <linux/sched.h> #include <drm/drm_mode_config.h> +#include <drm/drm_netlink.h> struct drm_driver ;struct drm_minor ;@@ -22,6 +23,8 @@ struct inode;struct pci_dev ;struct pci_controller ; +struct genl_family ; +/* * Recovery methods for wedged device in order of less to more side-effects. * To be used with drm_dev_wedged_event() as recovery @method. Callers can
include/drm/drm_device.h struct drm_device { · line 368
Quoted 10 lines
include/drm/drm_device.h +7
-0@@ -356,6 +359,13 @@ struct drm_device {* Root directory for debugfs files. */ struct dentry *debugfs_root; + +/** +* @drm_genl_family: +* +* Generic netlink family registration structure. +*/ +struct genl_family *drm_genl_family;
we should probably have this inside a struct drm_ras and without the 1-1 tie here
Quoted 8 lines
include/drm/drm_device.h +0
-0};void drm_dev_set_dma_dev (struct drm_device *dev,struct device *dma_dev);include/drm/drm_drv.h +0
-0
include/drm/drm_drv.h struct drm_driver { · line 440
Quoted 10 lines
include/drm/drm_drv.h +7
-0@@ -431,6 +431,13 @@ struct drm_driver {* some examples. */ const struct file_operations *fops; + +/** +* @genl_ops: +* +* Drivers private callback to genl commands +*/ +const struct driver_genl_ops *genl_ops;
as well the ops should be encapsulated in the drm_ras struct
Quoted 15 lines
include/drm/drm_drv.h +0
-0}; void *__devm_drm_dev_alloc(struct device *parent,include/drm/drm_netlink.h new file +6
-0@@ -0,0 +1,41 @@ +/* SPDX-License-Identifier: MIT */ +/* +* Copyright © 2023 Intel Corporation +*/ + +#ifndef __DRM_NETLINK_H__
include/drm/drm_netlink.h line 16
Quoted 10 lines
include/drm/drm_netlink.h new file +10
-0+#define __DRM_NETLINK_H__ + +#include <linux/types.h> + +struct drm_device ; +struct genl_info ; +struct sk_buff ; + +struct driver_genl_ops { + int (*doit)(struct drm_device *dev,
when I first saw the doit I was going to complain about it, until I learned this is part of netlink definition :)
Quoted 43 lines
include/drm/drm_netlink.h new file +25
-0+struct sk_buff *skb, + struct genl_info *info); +}; + +#if IS_ENABLED (CONFIG_NET ) +int drm_genl_register (struct drm_device *dev); +void drm_genl_exit (void ); +int drm_genl_reply (struct sk_buff *msg,struct genl_info *info,void *usrhdr); +struct sk_buff * +drm_genl_alloc_msg (struct drm_device *dev, +struct genl_info *info, +size_t msg_size,void **usrhdr); +#else +static inline int drm_genl_register (struct drm_device *dev) {return 0 ; } +static inline void drm_genl_exit (void ) {} +static inline int drm_genl_reply (struct sk_buff *msg, +struct genl_info *info, +void *usrhdr) {return 0 ; } +static inline struct skb_buff * +drm_genl_alloc_msg (struct drm_device *dev, +struct genl_info *info, +size_t msg_size,void **usrhdr) {return NULL ; } +#endif + +#endif include/uapi/drm/drm_netlink.h new file +12
-0@@ -0,0 +1,101 @@ +/* SPDX-License-Identifier: MIT */ +/* +* Copyright 2023 Intel Corporation +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next
include/uapi/drm/drm_netlink.h line 22
Quoted 10 lines
include/uapi/drm/drm_netlink.h new file +10
-0+* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* VA LINUX SYSTEMS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR +* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, +* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR +* OTHER DEALINGS IN THE SOFTWARE.
This header kind of conflicts/overlaps the MIT SPDX above. We should remove it and go only with the SPDX imho, unless I'm missing something
Quoted 12 lines
include/uapi/drm/drm_netlink.h new file +12
-0+*/ + +#ifndef _DRM_NETLINK_H_ +#define _DRM_NETLINK_H_ + +#define DRM_GENL_VERSION 1 + +#if defined(__cplusplus) +extern "C" { +#endif + +/**
include/uapi/drm/drm_netlink.h line 44
Quoted 10 lines
include/uapi/drm/drm_netlink.h new file +10
-0+* enum drm_genl_error_cmds - Supported error commands +* +*/ +enum drm_genl_error_cmds { +DRM_CMD_UNSPEC , +/** +* @DRM_RAS_CMD_QUERY: Command to list all errors names with config-id in verbose mode. +* In normal mode will list IP blocks, total instances available and error types supported +*/ +DRM_RAS_CMD_QUERY ,
here is the part where naming inconsistency is more visible, file has one namespacing, struct has another, and command has even a third one.
drm_ras everywhere to solve this please.
include/uapi/drm/drm_netlink.h line 50
Quoted 6 lines
include/uapi/drm/drm_netlink.h new file +6
-0+/** @DRM_RAS_CMD_READ_ONE: Command to get a counter for a specific error */ +DRM_RAS_CMD_READ_ONE , +/** @DRM_RAS_CMD_READ_BLOCK: Command to get a counter of specific error type from an IP +* block +*/ + DRM_RAS_CMD_READ_BLOCK,
here is the part that I believe this API already shows how it is not expansible. you had to create an argument to filter the type of errors instead of declaring the errors per ip block like AMD folks had asked for.
include/uapi/drm/drm_netlink.h line 61
Quoted 11 lines
include/uapi/drm/drm_netlink.h new file +11
-0+/** @DRM_RAS_CMD_READ_ALL: Command to get counters of all errors */ +DRM_RAS_CMD_READ_ALL , + + __DRM_CMD_MAX, + DRM_CMD_MAX = __DRM_CMD_MAX - 1, +}; + +enum drm_cmd_request_type { +DRM_RAS_CMD_QUERY_VERBOSE =1 , +DRM_RAS_CMD_QUERY_NORMAL =2 , +};
I don't understand why we need verbose vs normal. Perhaps this should be a separate path or explain with examples?
it took me a while to realize that the drm_ras igt tool would only list my available errors if I was using --verbose, otherwise we would return in the begin of the list_error functions in xe...
Quoted 20 lines
include/uapi/drm/drm_netlink.h new file +20
-0+ +/** +* enum drm_error_attr - Attributes to use with drm_genl_error_cmds +* +*/ +enum drm_error_attr { +DRM_ATTR_UNSPEC , +DRM_ATTR_PAD =DRM_ATTR_UNSPEC , +/** +* @DRM_RAS_ATTR_QUERY: Should be used with DRM_RAS_CMD_QUERY, +* DRM_RAS_CMD_READ_ALL +*/ +DRM_RAS_ATTR_QUERY ,/* NLA_U8 */ +/** +* @DRM_RAS_ATTR_READ_ALL: Should be used with DRM_RAS_CMD_READ_ALL +*/ +DRM_RAS_ATTR_READ_ALL ,/* NLA_U8 */ +/** +* @DRM_RAS_ATTR_QUERY_REPLY: First Nested attributed sent as a +* response to DRM_RAS_CMD_QUERY, DRM_RAS_CMD_READ_ALL commands.
include/uapi/drm/drm_netlink.h line 91
Quoted 10 lines
include/uapi/drm/drm_netlink.h new file +10
-0+*/ +DRM_RAS_ATTR_QUERY_REPLY ,/* NLA_NESTED */ +/** @DRM_RAS_ATTR_ERROR_NAME: Used to pass error name */ +DRM_RAS_ATTR_ERROR_NAME ,/* NLA_NUL_STRING */ +/** @DRM_RAS_ATTR_ERROR_ID: Used to pass error id, should be used with +* DRM_RAS_CMD_READ_ONE, DRM_RAS_CMD_READ_BLOCK +*/ +DRM_RAS_ATTR_ERROR_ID ,/* NLA_U64 */ +/** @DRM_RAS_ATTR_ERROR_VALUE: Used to pass error value */ + DRM_RAS_ATTR_ERROR_VALUE, /* NLA_U64 */
I'm also confused on all of the errors here and why we would need them and it also looks not expansible...
Quoted 13 lines
include/uapi/drm/drm_netlink.h new file +10
-0+ + __DRM_ATTR_MAX, + DRM_ATTR_MAX = __DRM_ATTR_MAX - 1, +}; + +#if defined(__cplusplus) +} +#endif + +#endif Signature · 1 line
2.25.1
Re: [RFC v5 1/5] drm/netlink: Add netlink infrastructure
Quoted 57 lines · Rodrigo Vivi
Quoted 56 lines · Aravind Iddamsetty
Define the netlink registration interface and commands, attributes that can be commonly used across by drm drivers. This patch intends to use the generic netlink family to expose various stats of device. At present it defines some commands that shall be used to expose RAS error counters.
v2:
define common interfaces to genl netlink subsystem that all drm drivers can leverage.(Tomer Tayar)v3: drop
DRIVER_NETLINKflag and use thedriver_genl_opsstructure to register to netlink subsystem (Daniel Vetter)v4:(Michael J. Ruhl)
1. renamedrm_genl_sendtodrm_genl_reply
2. catch error fromxa_storeand handle appropriatelyv5:
1. compile only ifCONFIG_NETis enabledV6: Add support for reading an IP block errors
Reviewed-by Michael J. Ruhl <[email protected]> #v4Signed-off-by Aravind Iddamsetty <[email protected]>--- drivers/gpu/drm/Makefile | 1 + drivers/gpu/drm/drm_drv.c | 7 ++ drivers/gpu/drm/drm_netlink.c | 212 +++++++++++++++++++++++++++++++++ include/drm/drm_device.h | 10 ++ include/drm/drm_drv.h | 7 ++ include/drm/drm_netlink.h | 41 +++++++ include/uapi/drm/drm_netlink.h | 101 ++++++++++++++++ 7 files changed, 379 insertions(+) create mode 100644 drivers/gpu/drm/drm_netlink.c create mode 100644 include/drm/drm_netlink.h create mode 100644 include/uapi/drm/drm_netlink.hdrivers/gpu/drm/Makefile +1
-0@@ -77,6 +77,7 @@ drm-$(CONFIG_DRM_CLIENT) += \ drm_client.o \ drm_client_event.o \ drm_client_modeset.o +drm-$(CONFIG_NET) += drm_netlink.o drm-$(CONFIG_DRM_LIB_RANDOM) += lib/drm_random.o drm-$(CONFIG_COMPAT) += drm_ioc32.o drm-$(CONFIG_DRM_PANEL) += drm_panel.odrivers/gpu/drm/drm_drv.c +0
-0@@ -1088,6 +1088,12 @@ int drm_dev_register(struct drm_device *dev, unsigned long flags)if (ret) goto err_minors;
Quoted 7 lines
drivers/gpu/drm/drm_drv.c int drm_dev_register(struct drm_device *dev, unsigned long flags) · line 1095
Quoted 5 lines
drivers/gpu/drm/drm_drv.c +5
-0+if (driver->genl_ops) { + ret =drm_genl_register (dev); +if (ret) + goto err_minors; + }Even if we don't go with multiple 'groups' I believe that the driver should explicitly call the netlink registration.
the check is if the driver is supporting netlink callbacks or not and only then register the netlink family for the device.
Also, I believe we want drivers to explicitly call drm_ras_register, see below
Quoted 11 lines
Quoted 11 lines
drivers/gpu/drm/drm_drv.c +2
-0+ ret =create_compat_control_link (dev);if (ret) goto err_minors;@@ -1229,6 +1235,7 @@ static void drm_core_exit(void)drm_privacy_screen_lookup_exit ();drm_panic_exit ();accel_core_exit (); +drm_genl_exit ();unregister_chrdev (DRM_MAJOR ,"drm" );
Quoted 8 lines
drivers/gpu/drm/drm_netlink.c
Quoted 7 lines
drivers/gpu/drm/drm_drv.c +0
-0debugfs_remove (drm_debugfs_root);drm_sysfs_destroy ();drivers/gpu/drm/drm_netlink.c new file +0
-0
drm_ras.c?
ok
Quoted 6 lines
drivers/gpu/drm/drm_netlink.c line 3
Quoted 5 lines
drivers/gpu/drm/drm_netlink.c new file +3
-0@@ -0,0 +1,212 @@ +// SPDX-License-Identifier: MIT +/* +* Copyright © 2023 Intel Corporation 2025 here and in any other file
by bad i just respined the old series without checking for this will fix it.
Quoted 8 lines
drivers/gpu/drm/drm_netlink.c line 7
Quoted 5 lines
drivers/gpu/drm/drm_netlink.c new file +4
-0+*/ + +#include <net/genetlink.h> +#include <uapi/drm/drm_netlink.h>
uapi/drm/drm_ras.h?!like we don't have a
drm_ioctl.hbutdrm_mode.h
ok makes sense.
Quoted 23 lines
Quoted 11 lines
drivers/gpu/drm/drm_netlink.c new file +10
-0+ +#include <drm/drm_device.h> +#include <drm/drm_drv.h> +#include <drm/drm_file.h> +#include <drm/drm_managed.h> +#include <drm/drm_netlink.h> +#include <drm/drm_print.h> + +DEFINE_XARRAY (drm_dev_xarray); +drivers/gpu/drm/drm_netlink.c line 27
Quoted 10 lines
drivers/gpu/drm/drm_netlink.c new file +10
-0+/** +* drm_genl_reply - response to a request +* @msg: socket buffer +* @info: receiver information +* @usrhdr: pointer to user specific header in the message buffer +* +* RETURNS: +* 0 on success and negative error code on failure +*/ +int drm_genl_reply (struct sk_buff *msg,struct genl_info *info,void *usrhdr)
drm_ras_replyso we standardize in a single namespace everywhere ?!
Quoted 2 lines
and same for all other functions and structs, except for things that are declared outside drm
ok so the way i understood based on your comment in the first patch is netlink constructs
should be abstracted out from drm drivers so this interface should not expect sk_buff or genl
things and handle all that things internally.
Quoted 130 lines
Quoted 117 lines
drivers/gpu/drm/drm_netlink.c new file +116
-0+{ +int ret; + +genlmsg_end (msg, usrhdr); + + ret =genlmsg_reply (msg, info); +if (ret) +nlmsg_free (msg); + +return ret; +} +EXPORT_SYMBOL (drm_genl_reply); + +/** +* drm_genl_alloc_msg - allocate genl message buffer +* @dev: drm_device for which the message is being allocated +* @info: receiver information +* @msg_size: size of the msg buffer that needs to be allocated +* @usrhdr: pointer to user specific header in the message buffer +* +* RETURNS: +* pointer to new allocated buffer on success, NULL on failure +*/ +struct sk_buff * +drm_genl_alloc_msg (struct drm_device *dev, +struct genl_info *info, +size_t msg_size,void **usrhdr) +{ +struct sk_buff *new_msg; + + new_msg =genlmsg_new (msg_size,GFP_KERNEL ); +if (!new_msg) +return new_msg; + + *usrhdr =genlmsg_put_reply (new_msg, info, dev->drm_genl_family,0 , info->genlhdr->cmd); +if (!*usrhdr) { +nlmsg_free (new_msg); + new_msg =NULL ; + } + +return new_msg; +} +EXPORT_SYMBOL (drm_genl_alloc_msg); + +static struct drm_device *genl_to_dev (struct genl_info *info) +{ +return xa_load (&drm_dev_xarray, info->nlhdr->nlmsg_type); +} + +static int drm_genl_list_errors (struct sk_buff *msg,struct genl_info *info) +{ +struct drm_device *dev =genl_to_dev (info); + +if (info->genlhdr->cmd ==DRM_RAS_CMD_READ_ALL ) { +if (GENL_REQ_ATTR_CHECK (info,DRM_RAS_ATTR_READ_ALL )) +return -EINVAL ; + }else { +if (GENL_REQ_ATTR_CHECK (info,DRM_RAS_ATTR_QUERY )) +return -EINVAL ; + } + +if (WARN_ON (!dev->driver->genl_ops[info->genlhdr->cmd].doit)) +return -EOPNOTSUPP ; + +return dev->driver->genl_ops[info->genlhdr->cmd].doit (dev, msg, info); +} + +static int drm_genl_read_error (struct sk_buff *msg,struct genl_info *info) +{ +struct drm_device *dev =genl_to_dev (info); + +if (GENL_REQ_ATTR_CHECK (info,DRM_RAS_ATTR_ERROR_ID )) +return -EINVAL ; + +if (WARN_ON (!dev->driver->genl_ops[info->genlhdr->cmd].doit)) +return -EOPNOTSUPP ; + +return dev->driver->genl_ops[info->genlhdr->cmd].doit (dev, msg, info); +} + +/* attribute policies */ +static const struct nla_policy drm_attr_policy_query[DRM_ATTR_MAX +1 ] = { + [DRM_RAS_ATTR_QUERY ] = { .type =NLA_U8 }, +}; + +static const struct nla_policy drm_attr_policy_read_one[DRM_ATTR_MAX +1 ] = { + [DRM_RAS_ATTR_ERROR_ID ] = { .type =NLA_U64 }, +}; + +static const struct nla_policy drm_attr_policy_read_all[DRM_ATTR_MAX +1 ] = { + [DRM_RAS_ATTR_READ_ALL ] = { .type =NLA_U8 }, +}; + +/* drm genl operations definition */ +const struct genl_ops drm_genl_ops[] = { + { + .cmd =DRM_RAS_CMD_QUERY , + .doit = drm_genl_list_errors, + .policy = drm_attr_policy_query, + }, + { + .cmd =DRM_RAS_CMD_READ_ONE , + .doit = drm_genl_read_error, + .policy = drm_attr_policy_read_one, + }, + { + .cmd =DRM_RAS_CMD_READ_ALL , + .doit = drm_genl_list_errors, + .policy = drm_attr_policy_read_all, + }, + { + .cmd =DRM_RAS_CMD_READ_BLOCK , + .doit = drm_genl_read_error, + .policy = drm_attr_policy_read_one, + }, +drivers/gpu/drm/drm_netlink.c line 153
Quoted 10 lines
drivers/gpu/drm/drm_netlink.c new file +10
-0+}; + +static void drm_genl_family_init (struct drm_device *dev) +{ + dev->drm_genl_family =drmm_kzalloc (dev,sizeof (struct genl_family ), +GFP_KERNEL ); + +/* Use drm primary node name eg: card0 to name the genl family */ +snprintf (dev->drm_genl_family->name,sizeof (dev->drm_genl_family->name), +"%s" , dev->primary->kdev->kobj.name);for the family name I believe we deserve the 'drmras', then the card minor number, then the group name.
Quoted 8 lines
For instance, but not necessarily suggesting xe to do it:
drmras-0-gt
drmras-0-soc.....
driver can select their own name...
ok the drm_ras_register that we will introduce shall take a name. As I mentioned earlier we needn't expose IP as group.
Quoted 5 lines
drivers/gpu/drm/drm_netlink.c line 154
Quoted 3 lines
drivers/gpu/drm/drm_netlink.c new file +1
-0+ dev->drm_genl_family->version =DRM_GENL_VERSION ;I believe driver could control their own version so if something changes in the group names for instance or supported commands they can change it.
but the commands itself are common, I'm not sure if we want drivers to have their private set of commands.
Quoted 85 lines
Quoted 85 lines
drivers/gpu/drm/drm_netlink.c new file +58
-0+ dev->drm_genl_family->parallel_ops = true; + dev->drm_genl_family->ops = drm_genl_ops; + dev->drm_genl_family->n_ops =ARRAY_SIZE (drm_genl_ops); + dev->drm_genl_family->maxattr =DRM_ATTR_MAX ; + dev->drm_genl_family->module = dev->dev->driver->owner; +} + +static void drm_genl_deregister (struct drm_device *dev,void *arg) +{ +drm_dbg_driver (dev,"unregistering genl family %s\n" , dev->drm_genl_family->name); + +xa_erase (&drm_dev_xarray, dev->drm_genl_family->id); + +genl_unregister_family (dev->drm_genl_family); +} + +/** +* drm_genl_register - Register genl family +* @dev: drm_device for which genl family needs to be registered +* +* RETURNS: +* 0 on success and negative error code on failure +*/ +int drm_genl_register (struct drm_device *dev) +{ +int ret; + +drm_genl_family_init (dev); + + ret =genl_register_family (dev->drm_genl_family); +if (ret <0 ) { +drm_warn (dev,"genl family registration failed\n" ); +return ret; + } + +drm_dbg_driver (dev,"genl family id %d and name %s\n" , dev->drm_genl_family->id, + dev->drm_genl_family->name); + + ret =xa_err (xa_store (&drm_dev_xarray, dev->drm_genl_family->id, dev,GFP_KERNEL )); +if (ret) + goto genl_unregister; + + ret =drmm_add_action_or_reset (dev, drm_genl_deregister,NULL ); + +return ret; + +genl_unregister: +genl_unregister_family (dev->drm_genl_family); +return ret; +} + +/** +* drm_genl_exit: destroy drm_dev_xarray +*/ +void drm_genl_exit (void ) +{ +xa_destroy (&drm_dev_xarray); +}include/drm/drm_device.h +4
-0@@ -8,6 +8,7 @@#include <linux/sched.h> #include <drm/drm_mode_config.h> +#include <drm/drm_netlink.h> struct drm_driver ;struct drm_minor ;@@ -22,6 +23,8 @@ struct inode;struct pci_dev ;struct pci_controller ; +struct genl_family ; +/* * Recovery methods for wedged device in order of less to more side-effects. * To be used with drm_dev_wedged_event() as recovery @method. Callers can @@ -356,6 +359,13 @@ struct drm_device {* Root directory for debugfs files. */ struct dentry *debugfs_root; +
Quoted 8 lines
include/drm/drm_device.h struct drm_device { · line 368
Quoted 6 lines
include/drm/drm_device.h +6
-0+/** +* @drm_genl_family: +* +* Generic netlink family registration structure. +*/ +struct genl_family *drm_genl_family;we should probably have this inside a struct
drm_rasand without the 1-1 tie here
Ok.
Quoted 13 lines
Quoted 13 lines
include/drm/drm_device.h +0
-0};void drm_dev_set_dma_dev (struct drm_device *dev,struct device *dma_dev);include/drm/drm_drv.h +0
-0@@ -431,6 +431,13 @@ struct drm_driver {* some examples. */ const struct file_operations *fops;
Quoted 8 lines
include/drm/drm_drv.h struct drm_driver { · line 440
Quoted 7 lines
include/drm/drm_drv.h +7
-0+ +/** +* @genl_ops: +* +* Drivers private callback to genl commands +*/ +const struct driver_genl_ops *genl_ops;as well the ops should be encapsulated in the
drm_rasstruct
got it.
Quoted 76 lines
Quoted 16 lines
include/drm/drm_drv.h +0
-0}; void *__devm_drm_dev_alloc(struct device *parent,include/drm/drm_netlink.h new file +6
-0@@ -0,0 +1,41 @@ +/* SPDX-License-Identifier: MIT */ +/* +* Copyright © 2023 Intel Corporation +*/ + +#ifndef __DRM_NETLINK_H__include/drm/drm_netlink.h line 16
Quoted 10 lines
include/drm/drm_netlink.h new file +10
-0+#define __DRM_NETLINK_H__ + +#include <linux/types.h> + +struct drm_device ; +struct genl_info ; +struct sk_buff ; + +struct driver_genl_ops { + int (*doit)(struct drm_device *dev,when I first saw the doit I was going to complain about it, until I learned this is part of netlink definition :)
Quoted 47 lines
include/drm/drm_netlink.h new file +25
-0+struct sk_buff *skb, + struct genl_info *info); +}; + +#if IS_ENABLED (CONFIG_NET ) +int drm_genl_register (struct drm_device *dev); +void drm_genl_exit (void ); +int drm_genl_reply (struct sk_buff *msg,struct genl_info *info,void *usrhdr); +struct sk_buff * +drm_genl_alloc_msg (struct drm_device *dev, +struct genl_info *info, +size_t msg_size,void **usrhdr); +#else +static inline int drm_genl_register (struct drm_device *dev) {return 0 ; } +static inline void drm_genl_exit (void ) {} +static inline int drm_genl_reply (struct sk_buff *msg, +struct genl_info *info, +void *usrhdr) {return 0 ; } +static inline struct skb_buff * +drm_genl_alloc_msg (struct drm_device *dev, +struct genl_info *info, +size_t msg_size,void **usrhdr) {return NULL ; } +#endif + +#endif include/uapi/drm/drm_netlink.h new file +16
-0@@ -0,0 +1,101 @@ +/* SPDX-License-Identifier: MIT */ +/* +* Copyright 2023 Intel Corporation +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
Quoted 8 lines
include/uapi/drm/drm_netlink.h line 22
Quoted 6 lines
include/uapi/drm/drm_netlink.h new file +6
-0+* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* VA LINUX SYSTEMS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR +* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, +* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR +* OTHER DEALINGS IN THE SOFTWARE. This header kind of conflicts/overlaps the MIT SPDX above. We should remove it and go only with the SPDX imho, unless I'm missing something
My bad may be i clubbed the two taking from some header.
Quoted 26 lines
Quoted 13 lines
include/uapi/drm/drm_netlink.h new file +12
-0+*/ + +#ifndef _DRM_NETLINK_H_ +#define _DRM_NETLINK_H_ + +#define DRM_GENL_VERSION 1 + +#if defined(__cplusplus) +extern "C" { +#endif + +/**include/uapi/drm/drm_netlink.h line 44
Quoted 10 lines
include/uapi/drm/drm_netlink.h new file +10
-0+* enum drm_genl_error_cmds - Supported error commands +* +*/ +enum drm_genl_error_cmds { +DRM_CMD_UNSPEC , +/** +* @DRM_RAS_CMD_QUERY: Command to list all errors names with config-id in verbose mode. +* In normal mode will list IP blocks, total instances available and error types supported +*/ +DRM_RAS_CMD_QUERY ,here is the part where naming inconsistency is more visible, file has one namespacing, struct has another, and command has even a third one.
Quoted 1 line
drm_raseverywhere to solve this please.
Ok makes sense.
Quoted 10 lines
include/uapi/drm/drm_netlink.h line 50
Quoted 7 lines
include/uapi/drm/drm_netlink.h new file +6
-0+/** @DRM_RAS_CMD_READ_ONE: Command to get a counter for a specific error */ +DRM_RAS_CMD_READ_ONE , +/** @DRM_RAS_CMD_READ_BLOCK: Command to get a counter of specific error type from an IP +* block +*/ + DRM_RAS_CMD_READ_BLOCK,here is the part that I believe this API already shows how it is not expansible. you had to create an argument to filter the type of errors instead of declaring the errors per ip block like AMD folks had asked for.
Based on the comment here
https://lore.kernel.org/all/[email protected]/
they want to extract the details from a command itself using bitmask. Let me know if you think i misunderstood.
Quoted 15 lines
include/uapi/drm/drm_netlink.h line 61
Quoted 12 lines
include/uapi/drm/drm_netlink.h new file +11
-0+/** @DRM_RAS_CMD_READ_ALL: Command to get counters of all errors */ +DRM_RAS_CMD_READ_ALL , + + __DRM_CMD_MAX, + DRM_CMD_MAX = __DRM_CMD_MAX - 1, +}; + +enum drm_cmd_request_type { +DRM_RAS_CMD_QUERY_VERBOSE =1 , +DRM_RAS_CMD_QUERY_NORMAL =2 , +};I don't understand why we need verbose vs normal. Perhaps this should be a separate path or explain with examples?
Quoted 3 lines
it took me a while to realize that the
drm_rasigt tool would only list my available errors if I was using --verbose, otherwise we would return in the begin of thelist_errorfunctions in xe...
The VERBOSE is to support the current error model in Xe where each error is enumerated separately and NORMAL for reporting groups for AMDs model.
By default drm_ras tool uses normal that is why it doesn't work with xe unless --verbose is passed.
Quoted 26 lines
Quoted 26 lines
include/uapi/drm/drm_netlink.h new file +25
-0+ +/** +* enum drm_error_attr - Attributes to use with drm_genl_error_cmds +* +*/ +enum drm_error_attr { +DRM_ATTR_UNSPEC , +DRM_ATTR_PAD =DRM_ATTR_UNSPEC , +/** +* @DRM_RAS_ATTR_QUERY: Should be used with DRM_RAS_CMD_QUERY, +* DRM_RAS_CMD_READ_ALL +*/ +DRM_RAS_ATTR_QUERY ,/* NLA_U8 */ +/** +* @DRM_RAS_ATTR_READ_ALL: Should be used with DRM_RAS_CMD_READ_ALL +*/ +DRM_RAS_ATTR_READ_ALL ,/* NLA_U8 */ +/** +* @DRM_RAS_ATTR_QUERY_REPLY: First Nested attributed sent as a +* response to DRM_RAS_CMD_QUERY, DRM_RAS_CMD_READ_ALL commands. +*/ +DRM_RAS_ATTR_QUERY_REPLY ,/* NLA_NESTED */ +/** @DRM_RAS_ATTR_ERROR_NAME: Used to pass error name */ +DRM_RAS_ATTR_ERROR_NAME ,/* NLA_NUL_STRING */ +/** @DRM_RAS_ATTR_ERROR_ID: Used to pass error id, should be used with
Quoted 8 lines
include/uapi/drm/drm_netlink.h line 91
Quoted 5 lines
include/uapi/drm/drm_netlink.h new file +5
-0+* DRM_RAS_CMD_READ_ONE, DRM_RAS_CMD_READ_BLOCK +*/ +DRM_RAS_ATTR_ERROR_ID ,/* NLA_U64 */ +/** @DRM_RAS_ATTR_ERROR_VALUE: Used to pass error value */ + DRM_RAS_ATTR_ERROR_VALUE, /* NLA_U64 */I'm also confused on all of the errors here and why we would need them and it also looks not expansible...
Netlink commands have attributes which enforce a policy and type checking the command can accept.
and attributes can be specific to the command.
These attributes are used in request and response commands as mentioned in the comments.
/* attribute policies */
static const struct nla_policy drm_attr_policy_query[DRM_ATTR_MAX + 1] = { [DRM_RAS_ATTR_QUERY] = { .type = NLA_U8 },
};
static const struct nla_policy drm_attr_policy_read_one[DRM_ATTR_MAX + 1] = {
[DRM_RAS_ATTR_ERROR_ID] = { .type = NLA_U64 },
};
static const struct nla_policy drm_attr_policy_read_all[DRM_ATTR_MAX + 1] = {
[DRM_RAS_ATTR_READ_ALL] = { .type = NLA_U8 },
};
Please let me know if it is unclear.
Thanks,
Aravind.
Quoted 14 lines
Quoted 14 lines
include/uapi/drm/drm_netlink.h new file +10
-0+ + __DRM_ATTR_MAX, + DRM_ATTR_MAX = __DRM_ATTR_MAX - 1, +}; + +#if defined(__cplusplus) +} +#endif + +#endif Signature · 1 line
2.25.1
RFC v5 2/5 drm/xe/RAS: Register netlink capability
In reply to Aravind IddamsettyRegister netlink capability with the DRM and register the driver callbacks to DRM RAS netlink commands.
v2:
Move the netlink registration parts to DRM susbsytem (Tomer Tayar)
v3: compile only if CONFIG_NET is enabled
--- drivers/gpu/drm/xe/Makefile | 2 ++ drivers/gpu/drm/xe/xe_device.c | 6 ++++++ drivers/gpu/drm/xe/xe_device_types.h | 1 + drivers/gpu/drm/xe/xe_netlink.c | 26 ++++++++++++++++++++++++++ 4 files changed, 35 insertions(+) create mode 100644 drivers/gpu/drm/xe/xe_netlink.c
drivers/gpu/drm/xe/Makefile +2 -0
@@ -304,6 +304,8 @@ xe-$(CONFIG_DRM_XE_DISPLAY) += \ i915-display/skl_universal_plane.o \ i915-display/skl_watermark.o +xe-$(CONFIG_NET) += xe_netlink.o + ifeq ($(CONFIG_ACPI),y) xe-$(CONFIG_DRM_XE_DISPLAY) += \ i915-display/intel_acpi.o \
drivers/gpu/drm/xe/xe_device.c +6 -0
@@ -363,6 +363,8 @@ static const struct file_operations xe_driver_fops = { .fop_flags = FOP_UNSIGNED_OFFSET, }; +extern const struct driver_genl_ops xe_genl_ops[]; +static struct drm_driver driver = {/* Don't use MTRRs here; the Xserver or userspace app should * deal with them for Intel hardware.
@@ -381,6 +383,10 @@ static struct drm_driver driver = {#ifdef CONFIG_PROC_FS .show_fdinfo = xe_drm_client_fdinfo,#endif +#ifdef CONFIG_NET + .genl_ops = xe_genl_ops, +#endif + .ioctls = xe_ioctls, .num_ioctls = ARRAY_SIZE(xe_ioctls), .fops = &xe_driver_fops,
drivers/gpu/drm/xe/xe_device_types.h +1 -0
@@ -10,6 +10,7 @@#include <drm/drm_device.h> #include <drm/drm_file.h> +#include <drm/drm_netlink.h> #include <drm/ttm/ttm_device.h> #include "xe_devcoredump_types.h"
drivers/gpu/drm/xe/xe_netlink.c new file +26 -0
@@ -0,0 +1,26 @@ +// SPDX-License-Identifier: MIT +/* +* Copyright © 2023 Intel Corporation +*/ + +#include <net/genetlink.h> +#include <uapi/drm/drm_netlink.h> + +#include "xe_device.h" + +static int xe_genl_list_errors (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info) +{ +return 0 ; +} + +static int xe_genl_read_error (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info) +{ +return 0 ; +} + +/* driver callbacks to DRM netlink commands*/ +const struct driver_genl_ops xe_genl_ops[] = { + [DRM_RAS_CMD_QUERY ] = { .doit = xe_genl_list_errors }, + [DRM_RAS_CMD_READ_ONE ] = { .doit = xe_genl_read_error }, + [DRM_RAS_CMD_READ_ALL ] = { .doit = xe_genl_list_errors, }, +};
Signature · 1 line
2.25.1
Rodrigo Vivi
maintainerRe: [RFC v5 2/5] drm/xe/RAS: Register netlink capability
Quoted 45 lines · Aravind Iddamsetty
Register netlink capability with the DRM and register the driver callbacks to DRM RAS netlink commands.
v2:
Move the netlink registration parts to DRM susbsytem (Tomer Tayar)v3: compile only if
CONFIG_NETis enabledReviewed-by Michael J. Ruhl <[email protected]> #v2Signed-off-by Aravind Iddamsetty <[email protected]>--- drivers/gpu/drm/xe/Makefile | 2 ++ drivers/gpu/drm/xe/xe_device.c | 6 ++++++ drivers/gpu/drm/xe/xe_device_types.h | 1 + drivers/gpu/drm/xe/xe_netlink.c | 26 ++++++++++++++++++++++++++ 4 files changed, 35 insertions(+) create mode 100644 drivers/gpu/drm/xe/xe_netlink.cdrivers/gpu/drm/xe/Makefile +2
-0@@ -304,6 +304,8 @@ xe-$(CONFIG_DRM_XE_DISPLAY) += \ i915-display/skl_universal_plane.o \ i915-display/skl_watermark.o +xe-$(CONFIG_NET) += xe_netlink.o + ifeq ($(CONFIG_ACPI),y) xe-$(CONFIG_DRM_XE_DISPLAY) += \ i915-display/intel_acpi.o \drivers/gpu/drm/xe/xe_device.c +2
-0@@ -363,6 +363,8 @@ static const struct file_operations xe_driver_fops = { .fop_flags = FOP_UNSIGNED_OFFSET, }; +extern const struct driver_genl_ops xe_genl_ops[]; +static struct drm_driver driver = {/* Don't use MTRRs here; the Xserver or userspace app should * deal with them for Intel hardware.
drivers/gpu/drm/xe/xe_device.c static struct drm_driver driver = { · line 389
Quoted 7 lines
drivers/gpu/drm/xe/xe_device.c +4
-0@@ -381,6 +383,10 @@ static struct drm_driver driver = {#ifdef CONFIG_PROC_FS .show_fdinfo = xe_drm_client_fdinfo,#endif +#ifdef CONFIG_NET + .genl_ops = xe_genl_ops, +#endif +
we should definitely have a drm function to register it instead of hard-coding it here, regardless if we go with the group split or not.
It is not okay forcing this to every platform, even the ones without any RAS available for instance.
Quoted 37 lines
drivers/gpu/drm/xe/xe_device.c +0
-0.ioctls = xe_ioctls, .num_ioctls = ARRAY_SIZE(xe_ioctls), .fops = &xe_driver_fops,drivers/gpu/drm/xe/xe_device_types.h +1
-0@@ -10,6 +10,7 @@#include <drm/drm_device.h> #include <drm/drm_file.h> +#include <drm/drm_netlink.h> #include <drm/ttm/ttm_device.h> #include "xe_devcoredump_types.h" drivers/gpu/drm/xe/xe_netlink.c new file +16
-0@@ -0,0 +1,26 @@ +// SPDX-License-Identifier: MIT +/* +* Copyright © 2023 Intel Corporation +*/ + +#include <net/genetlink.h> +#include <uapi/drm/drm_netlink.h> + +#include "xe_device.h" + +static int xe_genl_list_errors (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info) +{ +return 0 ; +} + +static int xe_genl_read_error (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info)
drivers/gpu/drm/xe/xe_netlink.c line 26
Quoted 10 lines
drivers/gpu/drm/xe/xe_netlink.c new file +10
-0+{ +return 0 ; +} + +/* driver callbacks to DRM netlink commands*/ +const struct driver_genl_ops xe_genl_ops[] = { + [DRM_RAS_CMD_QUERY ] = { .doit = xe_genl_list_errors }, + [DRM_RAS_CMD_READ_ONE ] = { .doit = xe_genl_read_error }, + [DRM_RAS_CMD_READ_ALL ] = { .doit = xe_genl_list_errors, }, +};
this is another space that is strange. you declare it here and drm magically uses it. Another reason for more explicity registration. and with the struct drm_ras where these commands are part of that. as well as the group name, etc.
Quoted 3 lines
Signature · 1 line
2.25.1
Re: [RFC v5 2/5] drm/xe/RAS: Register netlink capability
Quoted 49 lines · Rodrigo Vivi
Quoted 48 lines · Aravind Iddamsetty
Register netlink capability with the DRM and register the driver callbacks to DRM RAS netlink commands.
v2:
Move the netlink registration parts to DRM susbsytem (Tomer Tayar)v3: compile only if
CONFIG_NETis enabledReviewed-by Michael J. Ruhl <[email protected]> #v2Signed-off-by Aravind Iddamsetty <[email protected]>--- drivers/gpu/drm/xe/Makefile | 2 ++ drivers/gpu/drm/xe/xe_device.c | 6 ++++++ drivers/gpu/drm/xe/xe_device_types.h | 1 + drivers/gpu/drm/xe/xe_netlink.c | 26 ++++++++++++++++++++++++++ 4 files changed, 35 insertions(+) create mode 100644 drivers/gpu/drm/xe/xe_netlink.cdrivers/gpu/drm/xe/Makefile +2
-0@@ -304,6 +304,8 @@ xe-$(CONFIG_DRM_XE_DISPLAY) += \ i915-display/skl_universal_plane.o \ i915-display/skl_watermark.o +xe-$(CONFIG_NET) += xe_netlink.o + ifeq ($(CONFIG_ACPI),y) xe-$(CONFIG_DRM_XE_DISPLAY) += \ i915-display/intel_acpi.o \drivers/gpu/drm/xe/xe_device.c +2
-0@@ -363,6 +363,8 @@ static const struct file_operations xe_driver_fops = { .fop_flags = FOP_UNSIGNED_OFFSET, }; +extern const struct driver_genl_ops xe_genl_ops[]; +static struct drm_driver driver = {/* Don't use MTRRs here; the Xserver or userspace app should * deal with them for Intel hardware. @@ -381,6 +383,10 @@ static struct drm_driver driver = {#ifdef CONFIG_PROC_FS .show_fdinfo = xe_drm_client_fdinfo,#endif
Quoted 8 lines
drivers/gpu/drm/xe/xe_device.c static struct drm_driver driver = { · line 389
Quoted 4 lines
drivers/gpu/drm/xe/xe_device.c +4
-0+#ifdef CONFIG_NET + .genl_ops = xe_genl_ops, +#endif +we should definitely have a drm function to register it instead of hard-coding it here, regardless if we go with the group split or not.
It is not okay forcing this to every platform, even the ones without any RAS available for instance.
ok.
Quoted 43 lines
Quoted 43 lines
drivers/gpu/drm/xe/xe_device.c +0
-0.ioctls = xe_ioctls, .num_ioctls = ARRAY_SIZE(xe_ioctls), .fops = &xe_driver_fops,drivers/gpu/drm/xe/xe_device_types.h +1
-0@@ -10,6 +10,7 @@#include <drm/drm_device.h> #include <drm/drm_file.h> +#include <drm/drm_netlink.h> #include <drm/ttm/ttm_device.h> #include "xe_devcoredump_types.h" drivers/gpu/drm/xe/xe_netlink.c new file +22
-0@@ -0,0 +1,26 @@ +// SPDX-License-Identifier: MIT +/* +* Copyright © 2023 Intel Corporation +*/ + +#include <net/genetlink.h> +#include <uapi/drm/drm_netlink.h> + +#include "xe_device.h" + +static int xe_genl_list_errors (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info) +{ +return 0 ; +} + +static int xe_genl_read_error (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info) +{ +return 0 ; +} + +/* driver callbacks to DRM netlink commands*/ +const struct driver_genl_ops xe_genl_ops[] = {
Quoted 8 lines
drivers/gpu/drm/xe/xe_netlink.c line 26
Quoted 4 lines
drivers/gpu/drm/xe/xe_netlink.c new file +4
-0+ [DRM_RAS_CMD_QUERY] = { .doit = xe_genl_list_errors }, + [DRM_RAS_CMD_READ_ONE] = { .doit = xe_genl_read_error }, + [DRM_RAS_CMD_READ_ALL] = { .doit = xe_genl_list_errors, }, +};this is another space that is strange. you declare it here and drm magically uses it. Another reason for more explicity registration. and with the struct
drm_raswhere these commands are part of that. as well as the group name, etc.
agree, this shall be part of explicit registration.
Thanks,
Aravind.
Quoted 4 lines
Quoted 4 lines
Signature · 2 lines
-- 2.25.1
RFC v5 3/5 drm/xe/RAS: Expose the error counters
In reply to Aravind IddamsettyWe expose the various error counters supported on a hardware via genl subsytem through the registered commands to userspace. The DRM_RAS_CMD_QUERY lists the error names with config id, DRM_RAD_CMD_READ_ONE returns the counter value for the requested config id and the DRM_RAS_CMD_READ_ALL lists the counters for all errors along with their names and config ids.
v2: Rebase
v3:
1. presently xe_list_errors fills blank data for IGFX, prevent it by having an early check of IS_DGFX (Michael J. Ruhl)
2. update errors from all sources
v4: Check QUERY param, if its normal return not supported.
--- drivers/gpu/drm/xe/xe_hw_error.c | 15 +- drivers/gpu/drm/xe/xe_netlink.c | 509 ++++++++++++++++++++++++++++++- include/uapi/drm/xe_drm.h | 85 ++++++ 3 files changed, 602 insertions(+), 7 deletions(-)
drivers/gpu/drm/xe/xe_hw_error.c +10 -5
@@ -531,16 +531,21 @@ static void xe_clear_all_soc_errors(struct xe_device *xe)while (hw_err <HARDWARE_ERROR_MAX ) {for (i =0 ; i <XE_SOC_NUM_IEH ; i++) -xe_mmio_write32 (>->tile->mmio,SOC_GSYSEVTCTL_REG (base, slave_base, i), +xe_mmio_write32 (>->tile->mmio, +SOC_GSYSEVTCTL_REG (base, slave_base, i), ~REG_BIT (hw_err)); -xe_mmio_write32 (>->tile->mmio,SOC_GLOBAL_ERR_STAT_MASTER_REG (base, hw_err), +xe_mmio_write32 (>->tile->mmio, +SOC_GLOBAL_ERR_STAT_MASTER_REG (base, hw_err),REG_GENMASK (31 ,0 )); -xe_mmio_write32 (>->tile->mmio,SOC_LOCAL_ERR_STAT_MASTER_REG (base, hw_err), +xe_mmio_write32 (>->tile->mmio, +SOC_LOCAL_ERR_STAT_MASTER_REG (base, hw_err),REG_GENMASK (31 ,0 )); -xe_mmio_write32 (>->tile->mmio,SOC_GLOBAL_ERR_STAT_SLAVE_REG (slave_base, hw_err), +xe_mmio_write32 (>->tile->mmio, +SOC_GLOBAL_ERR_STAT_SLAVE_REG (slave_base, hw_err),REG_GENMASK (31 ,0 )); -xe_mmio_write32 (>->tile->mmio,SOC_LOCAL_ERR_STAT_SLAVE_REG (slave_base, hw_err), +xe_mmio_write32 (>->tile->mmio, +SOC_LOCAL_ERR_STAT_SLAVE_REG (slave_base, hw_err),REG_GENMASK (31 ,0 )); hw_err++; }
drivers/gpu/drm/xe/xe_netlink.c +507 -2
523 lines not on this page · raw message
include/uapi/drm/xe_drm.h +85 -0
91 lines not on this page · raw message
Signature · 1 line
2.25.1
Rodrigo Vivi
maintainerRe: [RFC v5 3/5] drm/xe/RAS: Expose the error counters
Quoted 43 lines · Aravind Iddamsetty
We expose the various error counters supported on a hardware via genl subsytem through the registered commands to userspace. The
DRM_RAS_CMD_QUERYlists the error names with config id,DRM_RAD_CMD_READ_ONEreturns the counter value for the requested config id and theDRM_RAS_CMD_READ_ALLlists the counters for all errors along with their names and config ids.v2: Rebase
v3:
1. presentlyxe_list_errorsfills blank data for IGFX, prevent it by having an early check ofIS_DGFX(Michael J. Ruhl)
2. update errors from all sourcesv4: Check QUERY param, if its normal return not supported.
Reviewed-by Michael J. Ruhl <[email protected]>Signed-off-by Aravind Iddamsetty <[email protected]>--- drivers/gpu/drm/xe/xe_hw_error.c | 15 +- drivers/gpu/drm/xe/xe_netlink.c | 509 ++++++++++++++++++++++++++++++- include/uapi/drm/xe_drm.h | 85 ++++++ 3 files changed, 602 insertions(+), 7 deletions(-)drivers/gpu/drm/xe/xe_hw_error.c +5
-3@@ -531,16 +531,21 @@ static void xe_clear_all_soc_errors(struct xe_device *xe)while (hw_err <HARDWARE_ERROR_MAX ) {for (i =0 ; i <XE_SOC_NUM_IEH ; i++) -xe_mmio_write32 (>->tile->mmio,SOC_GSYSEVTCTL_REG (base, slave_base, i), +xe_mmio_write32 (>->tile->mmio, +SOC_GSYSEVTCTL_REG (base, slave_base, i), ~REG_BIT (hw_err)); -xe_mmio_write32 (>->tile->mmio,SOC_GLOBAL_ERR_STAT_MASTER_REG (base, hw_err), +xe_mmio_write32 (>->tile->mmio, +SOC_GLOBAL_ERR_STAT_MASTER_REG (base, hw_err),REG_GENMASK (31 ,0 )); - xe_mmio_write32(>->tile->mmio, SOC_LOCAL_ERR_STAT_MASTER_REG(base, hw_err), + xe_mmio_write32(>->tile->mmio,
drivers/gpu/drm/xe/xe_hw_error.c static void xe_clear_all_soc_errors(struct xe_device *xe) · line 549
Quoted 10 lines
drivers/gpu/drm/xe/xe_hw_error.c +5
-2+SOC_LOCAL_ERR_STAT_MASTER_REG (base, hw_err), REG_GENMASK(31, 0)); -xe_mmio_write32 (>->tile->mmio,SOC_GLOBAL_ERR_STAT_SLAVE_REG (slave_base, hw_err), +xe_mmio_write32 (>->tile->mmio, +SOC_GLOBAL_ERR_STAT_SLAVE_REG (slave_base, hw_err),REG_GENMASK (31 ,0 )); -xe_mmio_write32 (>->tile->mmio,SOC_LOCAL_ERR_STAT_SLAVE_REG (slave_base, hw_err), +xe_mmio_write32 (>->tile->mmio, +SOC_LOCAL_ERR_STAT_SLAVE_REG (slave_base, hw_err),REG_GENMASK (31 ,0 ));
probably a fixup for the patch in the other series?
Which btw it was hard to understand the dependency. We should make this series indepentent of the other one.
Quoted 272 lines
drivers/gpu/drm/xe/xe_hw_error.c +0
-0hw_err++; }drivers/gpu/drm/xe/xe_netlink.c +259
-1@@ -6,16 +6,521 @@#include <net/genetlink.h> #include <uapi/drm/drm_netlink.h> +#include <drm/xe_drm.h> +#include "xe_device.h" -static int xe_genl_list_errors (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info) +#define MAX_ERROR_NAME 100 + +static const char *const xe_hw_error_events[] = { + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3_SNG ] ="correctable-l3-sng" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_GUC ] ="correctable-guc" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SAMPLER ] ="correctable-sampler" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SLM ] ="correctable-slm" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_IC ] ="correctable-eu-ic" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_GRF ] ="correctable-eu-grf" , + [DRM_XE_GENL_GT_ERROR_FATAL_ARR_BIST ] ="fatal-array-bist" , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_DOUB ] ="fatal-l3-double" , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_ECC_CHK ] ="fatal-l3-ecc-checker" , + [DRM_XE_GENL_GT_ERROR_FATAL_GUC ] ="fatal-guc" , + [DRM_XE_GENL_GT_ERROR_FATAL_IDI_PAR ] ="fatal-idi-parity" , + [DRM_XE_GENL_GT_ERROR_FATAL_SQIDI ] ="fatal-sqidi" , + [DRM_XE_GENL_GT_ERROR_FATAL_SAMPLER ] ="fatal-sampler" , + [DRM_XE_GENL_GT_ERROR_FATAL_SLM ] ="fatal-slm" , + [DRM_XE_GENL_GT_ERROR_FATAL_EU_IC ] ="fatal-eu-ic" , + [DRM_XE_GENL_GT_ERROR_FATAL_EU_GRF ] ="fatal-eu-grf" , + [DRM_XE_GENL_GT_ERROR_FATAL_FPU ] ="fatal-fpu" , + [DRM_XE_GENL_GT_ERROR_FATAL_TLB ] ="fatal-tlb" , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_FABRIC ] ="fatal-l3-fabric" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SUBSLICE ] ="correctable-subslice" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3BANK ] ="correctable-l3bank" , + [DRM_XE_GENL_GT_ERROR_FATAL_SUBSLICE ] ="fatal-subslice" , + [DRM_XE_GENL_GT_ERROR_FATAL_L3BANK ] ="fatal-l3bank" , + [DRM_XE_GENL_SGUNIT_ERROR_CORRECTABLE ] ="sgunit-correctable" , + [DRM_XE_GENL_SGUNIT_ERROR_NONFATAL ] ="sgunit-nonfatal" , + [DRM_XE_GENL_SGUNIT_ERROR_FATAL ] ="sgunit-fatal" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMD ] ="soc-nonfatal-csc-psf-cmd-parity" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMP ] ="soc-nonfatal-csc-psf-unexpected-completion" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_REQ ] ="soc-nonfatal-csc-psf-unsupported-request" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_ANR_MDFI ] ="soc-nonfatal-anr-mdfi" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2T ] ="soc-nonfatal-mdfi-t2t" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2C ] ="soc-nonfatal-mdfi-t2c" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,0 )] ="soc-nonfatal-hbm-ss0-0" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,1 )] ="soc-nonfatal-hbm-ss0-1" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,2 )] ="soc-nonfatal-hbm-ss0-2" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,3 )] ="soc-nonfatal-hbm-ss0-3" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,4 )] ="soc-nonfatal-hbm-ss0-4" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,5 )] ="soc-nonfatal-hbm-ss0-5" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,6 )] ="soc-nonfatal-hbm-ss0-6" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,7 )] ="soc-nonfatal-hbm-ss0-7" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,8 )] ="soc-nonfatal-hbm-ss1-0" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,9 )] ="soc-nonfatal-hbm-ss1-1" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,10 )] ="soc-nonfatal-hbm-ss1-2" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,11 )] ="soc-nonfatal-hbm-ss1-3" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,12 )] ="soc-nonfatal-hbm-ss1-4" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,13 )] ="soc-nonfatal-hbm-ss1-5" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,14 )] ="soc-nonfatal-hbm-ss1-6" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,15 )] ="soc-nonfatal-hbm-ss1-7" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,0 )] ="soc-nonfatal-hbm-ss2-0" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,1 )] ="soc-nonfatal-hbm-ss2-1" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,2 )] ="soc-nonfatal-hbm-ss2-2" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,3 )] ="soc-nonfatal-hbm-ss2-3" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,4 )] ="soc-nonfatal-hbm-ss2-4" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,5 )] ="soc-nonfatal-hbm-ss2-5" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,6 )] ="soc-nonfatal-hbm-ss2-6" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,7 )] ="soc-nonfatal-hbm-ss2-7" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,8 )] ="soc-nonfatal-hbm-ss3-0" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,9 )] ="soc-nonfatal-hbm-ss3-1" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,10 )] ="soc-nonfatal-hbm-ss3-2" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,11 )] ="soc-nonfatal-hbm-ss3-3" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,12 )] ="soc-nonfatal-hbm-ss3-4" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,13 )] ="soc-nonfatal-hbm-ss3-5" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,14 )] ="soc-nonfatal-hbm-ss3-6" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,15 )] ="soc-nonfatal-hbm-ss3-7" , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMD ] ="soc-fatal-csc-psf-cmd-parity" , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMP ] ="soc-fatal-csc-psf-unexpected-completion" , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_REQ ] ="soc-fatal-csc-psf-unsupported-request" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PUNIT ] ="soc-fatal-punit" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMD ] ="soc-fatal-pcie-psf-command-parity" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMP ] ="soc-fatal-pcie-psf-unexpected-completion" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_REQ ] ="soc-fatal-pcie-psf-unsupported-request" , + [DRM_XE_GENL_SOC_ERROR_FATAL_ANR_MDFI ] ="soc-fatal-anr-mdfi" , + [DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2T ] ="soc-fatal-mdfi-t2t" , + [DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2C ] ="soc-fatal-mdfi-t2c" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_AER ] ="soc-fatal-malformed-pcie-aer" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_ERR ] ="soc-fatal-malformed-pcie-err" , + [DRM_XE_GENL_SOC_ERROR_FATAL_UR_COND ] ="soc-fatal-ur-condition-ieh" , + [DRM_XE_GENL_SOC_ERROR_FATAL_SERR_SRCS ] ="soc-fatal-from-serr-sources" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,0 )] ="soc-fatal-hbm-ss0-0" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,1 )] ="soc-fatal-hbm-ss0-1" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,2 )] ="soc-fatal-hbm-ss0-2" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,3 )] ="soc-fatal-hbm-ss0-3" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,4 )] ="soc-fatal-hbm-ss0-4" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,5 )] ="soc-fatal-hbm-ss0-5" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,6 )] ="soc-fatal-hbm-ss0-6" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,7 )] ="soc-fatal-hbm-ss0-7" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,8 )] ="soc-fatal-hbm-ss1-0" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,9 )] ="soc-fatal-hbm-ss1-1" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,10 )] ="soc-fatal-hbm-ss1-2" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,11 )] ="soc-fatal-hbm-ss1-3" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,12 )] ="soc-fatal-hbm-ss1-4" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,13 )] ="soc-fatal-hbm-ss1-5" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,14 )] ="soc-fatal-hbm-ss1-6" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,15 )] ="soc-fatal-hbm-ss1-7" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,0 )] ="soc-fatal-hbm-ss2-0" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,1 )] ="soc-fatal-hbm-ss2-1" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,2 )] ="soc-fatal-hbm-ss2-2" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,3 )] ="soc-fatal-hbm-ss2-3" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,4 )] ="soc-fatal-hbm-ss2-4" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,5 )] ="soc-fatal-hbm-ss2-5" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,6 )] ="soc-fatal-hbm-ss2-6" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,7 )] ="soc-fatal-hbm-ss2-7" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,8 )] ="soc-fatal-hbm-ss3-0" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,9 )] ="soc-fatal-hbm-ss3-1" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,10 )] ="soc-fatal-hbm-ss3-2" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,11 )] ="soc-fatal-hbm-ss3-3" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,12 )] ="soc-fatal-hbm-ss3-4" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,13 )] ="soc-fatal-hbm-ss3-5" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,14 )] ="soc-fatal-hbm-ss3-6" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,15 )] ="soc-fatal-hbm-ss3-7" , + [DRM_XE_GENL_GSC_ERROR_CORRECTABLE_SRAM_ECC ] ="gsc-correctable-sram-ecc" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_SHUTDOWN ] ="gsc-nonfatal-mia-shutdown" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_INTERNAL ] ="gsc-nonfatal-mia-internal" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_SRAM_ECC ] ="gsc-nonfatal-sram-ecc" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_WDG_TIMEOUT ] ="gsc-nonfatal-wdg-timeout" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_ROM_PARITY ] ="gsc-nonfatal-rom-parity" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_UCODE_PARITY ] ="gsc-nonfatal-ucode-parity" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_VLT_GLITCH ] ="gsc-nonfatal-vlt-glitch" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_PULL ] ="gsc-nonfatal-fuse-pull" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_CRC_CHECK ] ="gsc-nonfatal-fuse-crc-check" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_SELF_MBIST ] ="gsc-nonfatal-self-mbist" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_AON_RF_PARITY ] ="gsc-nonfatal-aon-parity" , + [DRM_XE_GENL_SGGI_ERROR_NONFATAL ] ="sggi-nonfatal-data-parity" , + [DRM_XE_GENL_SGLI_ERROR_NONFATAL ] ="sgli-nonfatal-data-parity" , + [DRM_XE_GENL_SGCI_ERROR_NONFATAL ] ="sgci-nonfatal-data-parity" , + [DRM_XE_GENL_MERT_ERROR_NONFATAL ] ="mert-nonfatal-data-parity" , + [DRM_XE_GENL_SGGI_ERROR_FATAL ] ="sggi-fatal-data-parity" , + [DRM_XE_GENL_SGLI_ERROR_FATAL ] ="sgli-fatal-data-parity" , + [DRM_XE_GENL_SGCI_ERROR_FATAL ] ="sgci-fatal-data-parity" , + [DRM_XE_GENL_MERT_ERROR_FATAL ] ="mert-nonfatal-data-parity" , +}; + +static const unsigned long xe_hw_error_map[] = { + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3_SNG ] =XE_HW_ERR_GT_CORR_L3_SNG , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_GUC ] =XE_HW_ERR_GT_CORR_GUC , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SAMPLER ] =XE_HW_ERR_GT_CORR_SAMPLER , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SLM ] =XE_HW_ERR_GT_CORR_SLM , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_IC ] =XE_HW_ERR_GT_CORR_EU_IC , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_GRF ] =XE_HW_ERR_GT_CORR_EU_GRF , + [DRM_XE_GENL_GT_ERROR_FATAL_ARR_BIST ] =XE_HW_ERR_GT_FATAL_ARR_BIST , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_DOUB ] =XE_HW_ERR_GT_FATAL_L3_DOUB , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_ECC_CHK ] =XE_HW_ERR_GT_FATAL_L3_ECC_CHK , + [DRM_XE_GENL_GT_ERROR_FATAL_GUC ] =XE_HW_ERR_GT_FATAL_GUC , + [DRM_XE_GENL_GT_ERROR_FATAL_IDI_PAR ] =XE_HW_ERR_GT_FATAL_IDI_PAR , + [DRM_XE_GENL_GT_ERROR_FATAL_SQIDI ] =XE_HW_ERR_GT_FATAL_SQIDI , + [DRM_XE_GENL_GT_ERROR_FATAL_SAMPLER ] =XE_HW_ERR_GT_FATAL_SAMPLER , + [DRM_XE_GENL_GT_ERROR_FATAL_SLM ] =XE_HW_ERR_GT_FATAL_SLM , + [DRM_XE_GENL_GT_ERROR_FATAL_EU_IC ] =XE_HW_ERR_GT_FATAL_EU_IC , + [DRM_XE_GENL_GT_ERROR_FATAL_EU_GRF ] =XE_HW_ERR_GT_FATAL_EU_GRF , + [DRM_XE_GENL_GT_ERROR_FATAL_FPU ] =XE_HW_ERR_GT_FATAL_FPU , + [DRM_XE_GENL_GT_ERROR_FATAL_TLB ] =XE_HW_ERR_GT_FATAL_TLB , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_FABRIC ] =XE_HW_ERR_GT_FATAL_L3_FABRIC , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SUBSLICE ] =XE_HW_ERR_GT_CORR_SUBSLICE , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3BANK ] =XE_HW_ERR_GT_CORR_L3BANK , + [DRM_XE_GENL_GT_ERROR_FATAL_SUBSLICE ] =XE_HW_ERR_GT_FATAL_SUBSLICE , + [DRM_XE_GENL_GT_ERROR_FATAL_L3BANK ] =XE_HW_ERR_GT_FATAL_L3BANK , + [DRM_XE_GENL_SGUNIT_ERROR_CORRECTABLE ] =XE_HW_ERR_TILE_CORR_SGUNIT , + [DRM_XE_GENL_SGUNIT_ERROR_NONFATAL ] =XE_HW_ERR_TILE_NONFATAL_SGUNIT , + [DRM_XE_GENL_SGUNIT_ERROR_FATAL ] =XE_HW_ERR_TILE_FATAL_SGUNIT , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMD ] =XE_HW_ERR_SOC_NONFATAL_CSC_PSF_CMD , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMP ] =XE_HW_ERR_SOC_NONFATAL_CSC_PSF_CMP , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_REQ ] =XE_HW_ERR_SOC_NONFATAL_CSC_PSF_REQ , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_ANR_MDFI ] =XE_HW_ERR_SOC_NONFATAL_ANR_MDFI , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2T ] =XE_HW_ERR_SOC_NONFATAL_MDFI_T2T , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2C ] =XE_HW_ERR_SOC_NONFATAL_MDFI_T2C , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,0 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,1 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,2 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,3 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,4 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,5 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,6 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,7 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,8 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,9 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,10 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,11 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,12 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,13 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,14 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,15 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,0 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,1 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,2 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,3 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,4 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,5 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,6 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,7 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,8 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,9 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,10 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,11 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,12 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,13 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,14 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,15 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMD ] =XE_HW_ERR_SOC_FATAL_CSC_PSF_CMD , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMP ] =XE_HW_ERR_SOC_FATAL_CSC_PSF_CMP , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_REQ ] =XE_HW_ERR_SOC_FATAL_CSC_PSF_REQ , + [DRM_XE_GENL_SOC_ERROR_FATAL_PUNIT ] =XE_HW_ERR_SOC_FATAL_PUNIT , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMD ] =XE_HW_ERR_SOC_FATAL_PCIE_PSF_CMD , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMP ] =XE_HW_ERR_SOC_FATAL_PCIE_PSF_CMP , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_REQ ] =XE_HW_ERR_SOC_FATAL_PCIE_PSF_REQ , + [DRM_XE_GENL_SOC_ERROR_FATAL_ANR_MDFI ] =XE_HW_ERR_SOC_FATAL_ANR_MDFI , + [DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2T ] =XE_HW_ERR_SOC_FATAL_MDFI_T2T , + [DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2C ] =XE_HW_ERR_SOC_FATAL_MDFI_T2C , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_AER ] =XE_HW_ERR_SOC_FATAL_PCIE_AER , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_ERR ] =XE_HW_ERR_SOC_FATAL_PCIE_ERR , + [DRM_XE_GENL_SOC_ERROR_FATAL_UR_COND ] =XE_HW_ERR_SOC_FATAL_UR_COND , + [DRM_XE_GENL_SOC_ERROR_FATAL_SERR_SRCS ] =XE_HW_ERR_SOC_FATAL_SERR_SRCS , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,0 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,1 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,2 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,3 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,4 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,5 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,6 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,7 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,8 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,9 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,10 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,11 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,12 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,13 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,14 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,15 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,0 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,1 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,2 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,3 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,4 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,5 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,6 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,7 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,8 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,9 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,10 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,11 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,12 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,13 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,14 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,15 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL7 , + [DRM_XE_GENL_GSC_ERROR_CORRECTABLE_SRAM_ECC ] =XE_HW_ERR_GSC_CORR_SRAM , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_SHUTDOWN ] =XE_HW_ERR_GSC_NONFATAL_MIA_SHUTDOWN , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_INTERNAL ] =XE_HW_ERR_GSC_NONFATAL_MIA_INTERNAL , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_SRAM_ECC ] =XE_HW_ERR_GSC_NONFATAL_SRAM , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_WDG_TIMEOUT ] =XE_HW_ERR_GSC_NONFATAL_WDG , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_ROM_PARITY ] =XE_HW_ERR_GSC_NONFATAL_ROM_PARITY , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_UCODE_PARITY ] =XE_HW_ERR_GSC_NONFATAL_UCODE_PARITY , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_VLT_GLITCH ] =XE_HW_ERR_GSC_NONFATAL_VLT_GLITCH , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_PULL ] =XE_HW_ERR_GSC_NONFATAL_FUSE_PULL , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_CRC_CHECK ] =XE_HW_ERR_GSC_NONFATAL_FUSE_CRC , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_SELF_MBIST ] =XE_HW_ERR_GSC_NONFATAL_SELF_MBIST ,
drivers/gpu/drm/xe/xe_netlink.c line 279
Quoted 10 lines
drivers/gpu/drm/xe/xe_netlink.c +10
-0+ [DRM_XE_GENL_GSC_ERROR_NONFATAL_AON_RF_PARITY] = XE_HW_ERR_GSC_NONFATAL_AON_RF_PARITY, + [DRM_XE_GENL_SGGI_ERROR_NONFATAL] = XE_HW_ERR_TILE_NONFATAL_SGGI, + [DRM_XE_GENL_SGLI_ERROR_NONFATAL] = XE_HW_ERR_TILE_NONFATAL_SGLI, + [DRM_XE_GENL_SGCI_ERROR_NONFATAL] = XE_HW_ERR_TILE_NONFATAL_SGCI, + [DRM_XE_GENL_MERT_ERROR_NONFATAL] = XE_HW_ERR_TILE_NONFATAL_MERT, + [DRM_XE_GENL_SGGI_ERROR_FATAL] = XE_HW_ERR_TILE_FATAL_SGGI, + [DRM_XE_GENL_SGLI_ERROR_FATAL] = XE_HW_ERR_TILE_FATAL_SGLI, + [DRM_XE_GENL_SGCI_ERROR_FATAL] = XE_HW_ERR_TILE_FATAL_SGCI, + [DRM_XE_GENL_MERT_ERROR_FATAL] = XE_HW_ERR_TILE_FATAL_MERT, +};
probably deserves a separate header file?
Quoted 236 lines
drivers/gpu/drm/xe/xe_netlink.c +229
-1+ +static unsigned int config_gt_id (const u64 config) +{ +return config >> __XE_GENL_GT_SHIFT; +} + +static u64 config_counter (const u64 config) +{ +return config & ~(~0ULL << __XE_GENL_GT_SHIFT); +} + +static bool is_gt_error (const u64 config) +{ +unsigned int error; + + error =config_counter (config); +if (error <=DRM_XE_GENL_GT_ERROR_FATAL_FPU ) +return true; + +return false; +} + +static bool is_gt_vector_error (const u64 config) +{ +unsigned int error; + + error =config_counter (config); +if (error >=DRM_XE_GENL_GT_ERROR_FATAL_TLB && + error <=DRM_XE_GENL_GT_ERROR_FATAL_L3BANK ) +return true; + +return false; +} + +static bool is_pvc_invalid_gt_errors (const u64 config) +{ +switch (config_counter (config)) { +case DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3_SNG : +case DRM_XE_GENL_GT_ERROR_CORRECTABLE_SAMPLER : +case DRM_XE_GENL_GT_ERROR_FATAL_ARR_BIST : +case DRM_XE_GENL_GT_ERROR_FATAL_L3_DOUB : +case DRM_XE_GENL_GT_ERROR_FATAL_L3_ECC_CHK : +case DRM_XE_GENL_GT_ERROR_FATAL_IDI_PAR : +case DRM_XE_GENL_GT_ERROR_FATAL_SQIDI : +case DRM_XE_GENL_GT_ERROR_FATAL_SAMPLER : +case DRM_XE_GENL_GT_ERROR_FATAL_EU_IC : +return true; +default : +return false; + } +} + +static bool is_gsc_hw_error (const u64 config) +{ +if (config_counter (config) >=DRM_XE_GENL_GSC_ERROR_CORRECTABLE_SRAM_ECC && +config_counter (config) <=DRM_XE_GENL_GSC_ERROR_NONFATAL_AON_RF_PARITY ) +return true; + +return false; +} + +static bool is_soc_error (const u64 config) { +if (config_counter (config) >=DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMD && +config_counter (config) <=DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,15 )) +return true; + +return false; +} + +static int +config_status (struct xe_device *xe,u64 config) +{ +unsigned int gt_id =config_gt_id (config); +struct xe_gt *gt =xe_device_get_gt (xe, gt_id); + +if (!IS_DGFX (xe)) +return -ENODEV ; + +if (gt->info.type ==XE_GT_TYPE_UNINITIALIZED ) +return -ENOENT ; + +/* GSC HW ERRORS are present on root tile of +* platform supporting MEMORY SPARING only +*/ +if (is_gsc_hw_error (config) && !(xe->info.platform ==XE_PVC && !gt_id)) +return -ENODEV ; + +/* GT vectors error are valid on Platforms supporting error vectors only */ +if (is_gt_vector_error (config) && xe->info.platform !=XE_PVC ) +return -ENODEV ; + +/* Skip gt errors not supported on pvc */ +if (is_pvc_invalid_gt_errors (config) && xe->info.platform ==XE_PVC ) +return -ENODEV ; + +/* FATAL FPU error is valid on PVC only */ +if (config_counter (config) ==DRM_XE_GENL_GT_ERROR_FATAL_FPU && + !(xe->info.platform ==XE_PVC )) +return -ENODEV ; + +if (is_soc_error (config) && !(xe->info.platform ==XE_PVC )) +return -ENODEV ; + +return (config_counter (config) >= +ARRAY_SIZE (xe_hw_error_map)) ? -ENOENT :0 ; +} + +static u64 get_counter_value (struct xe_device *xe,u64 config) +{ +const unsigned int gt_id =config_gt_id (config); +struct xe_gt *gt =xe_device_get_gt (xe, gt_id); +unsigned int id =config_counter (config); + +if (is_gt_error (config) ||is_gt_vector_error (config)) +return xa_to_value (xa_load (>->errors.hw_error, xe_hw_error_map[id])); + +return xa_to_value (xa_load (>->tile->errors.hw_error, xe_hw_error_map[id])); +} + +static int fill_error_details (struct xe_device *xe,struct genl_info *info,struct sk_buff *new_msg) +{ +struct nlattr *entry_attr; +bool counter = false; +struct xe_gt *gt; +int i, j; + +BUILD_BUG_ON (ARRAY_SIZE (xe_hw_error_events) != +ARRAY_SIZE (xe_hw_error_map)); + +if (info->genlhdr->cmd ==DRM_RAS_CMD_READ_ALL ) + counter = true; + + entry_attr =nla_nest_start (new_msg,DRM_RAS_ATTR_QUERY_REPLY ); +if (!entry_attr) +return -EMSGSIZE ; + +for_each_gt (gt, xe, j) { +char str[MAX_ERROR_NAME ]; +u64 val; + +for (i =0 ; i <ARRAY_SIZE (xe_hw_error_events); i++) { +u64 config =DRM_XE_HW_ERROR (j, i); + +if (config_status (xe, config)) +continue ; + +/* should this be cleared everytime */ +snprintf (str,sizeof (str),"error-gt%d-%s" , j, xe_hw_error_events[i]); + +if (nla_put_string (new_msg,DRM_RAS_ATTR_ERROR_NAME , str)) + goto err; +if (nla_put_u64_64bit (new_msg,DRM_RAS_ATTR_ERROR_ID , config,DRM_ATTR_PAD )) + goto err; +if (counter) { + val =get_counter_value (xe, config); +if (nla_put_u64_64bit (new_msg,DRM_RAS_ATTR_ERROR_VALUE , val, +DRM_ATTR_PAD )) + goto err; + } + } + } + +nla_nest_end (new_msg, entry_attr); +return 0 ; +err: +drm_dbg_driver (&xe->drm,"msg buff is small\n" ); +nla_nest_cancel (new_msg, entry_attr); +nlmsg_free (new_msg); + +return -EMSGSIZE ; +} + +static int xe_genl_list_errors (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info) +{ +struct xe_device *xe =to_xe_device (drm); +size_t msg_size =NLMSG_DEFAULT_SIZE ; +enum drm_cmd_request_type query_type; +struct sk_buff *new_msg; +int retries =2 ; +void *usrhdr; +int ret =0 ; + +if (!IS_DGFX (xe)) +return -ENODEV ; + +/* Support verbose only errors */ + query_type =nla_get_u8 (info->attrs[DRM_RAS_ATTR_QUERY ]); +if (query_type ==DRM_RAS_CMD_QUERY_NORMAL ) +return -EOPNOTSUPP ; + +do { + new_msg =drm_genl_alloc_msg (drm, info, msg_size, &usrhdr); +if (!new_msg) +return -ENOMEM ; + + ret =fill_error_details (xe, info, new_msg); +if (!ret) +break ; + + msg_size +=NLMSG_DEFAULT_SIZE ; + }while (retries--); + +if (!ret) + ret =drm_genl_reply (new_msg, info, usrhdr); + +return ret; }static int xe_genl_read_error (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info) { -return 0 ; +struct xe_device *xe =to_xe_device (drm); +size_t msg_size =NLMSG_DEFAULT_SIZE ; +struct sk_buff *new_msg; +void *usrhdr; +int ret =0 ; +int retries =2 ; +u64 config, val; + +if (info->genlhdr->cmd ==DRM_RAS_CMD_READ_BLOCK ) +return -EOPNOTSUPP ; + + config =nla_get_u64 (info->attrs[DRM_RAS_ATTR_ERROR_ID ]); + ret =config_status (xe, config); +if (ret) +return ret; + do { + new_msg = drm_genl_alloc_msg(drm, info, msg_size, &usrhdr); + if (!new_msg) + return -ENOMEM; + + val = get_counter_value(xe, config); + if (nla_put_u64_64bit(new_msg, DRM_RAS_ATTR_ERROR_VALUE, val, DRM_ATTR_PAD)) { + msg_size += NLMSG_DEFAULT_SIZE;
drivers/gpu/drm/xe/xe_netlink.c line 524
Quoted 10 lines
drivers/gpu/drm/xe/xe_netlink.c +9
-0+continue ; + } + +break ; + }while (retries--); + + ret =drm_genl_reply (new_msg, info, usrhdr); + +return ret; }
this .c without any public function and no .h was what draw my attention that there was something wrong with the registration process. We need to have init and finish per component.
Quoted 7 lines
drivers/gpu/drm/xe/xe_netlink.c +0
-0/* driver callbacks to DRM netlink commands*/ include/uapi/drm/xe_drm.h +0
-0
include/uapi/drm/xe_drm.h struct drm_xe_query_eu_stall { · line 1983
Quoted 10 lines
include/uapi/drm/xe_drm.h +7
-0@@ -1974,6 +1974,91 @@ struct drm_xe_query_eu_stall {__u64 sampling_rates[]; }; +/* +* Top bits of every counter are GT id. +*/ +#define __XE_GENL_GT_SHIFT (56) +/** +* DOC: XE GENL netlink event IDs +* TODO: Add more details
yes, please it is hard to understand why here.
And also I have the feeling that this deserves to be together with the other definitions above all together in a separate header. perhaps even per platform?!
Quoted 84 lines
include/uapi/drm/xe_drm.h +78
-0+*/ +#define DRM_XE_HW_ERROR (gt, id) \ + ((id) | ((__u64)(gt) << __XE_GENL_GT_SHIFT)) + +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3_SNG (0) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_GUC (1) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_SAMPLER (2) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_SLM (3) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_IC (4) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_GRF (5) +#define DRM_XE_GENL_GT_ERROR_FATAL_ARR_BIST (6) +#define DRM_XE_GENL_GT_ERROR_FATAL_L3_DOUB (7) +#define DRM_XE_GENL_GT_ERROR_FATAL_L3_ECC_CHK (8) +#define DRM_XE_GENL_GT_ERROR_FATAL_GUC (9) +#define DRM_XE_GENL_GT_ERROR_FATAL_IDI_PAR (10) +#define DRM_XE_GENL_GT_ERROR_FATAL_SQIDI (11) +#define DRM_XE_GENL_GT_ERROR_FATAL_SAMPLER (12) +#define DRM_XE_GENL_GT_ERROR_FATAL_SLM (13) +#define DRM_XE_GENL_GT_ERROR_FATAL_EU_IC (14) +#define DRM_XE_GENL_GT_ERROR_FATAL_EU_GRF (15) +#define DRM_XE_GENL_GT_ERROR_FATAL_FPU (16) +#define DRM_XE_GENL_GT_ERROR_FATAL_TLB (17) +#define DRM_XE_GENL_GT_ERROR_FATAL_L3_FABRIC (18) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_SUBSLICE (19) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3BANK (20) +#define DRM_XE_GENL_GT_ERROR_FATAL_SUBSLICE (21) +#define DRM_XE_GENL_GT_ERROR_FATAL_L3BANK (22) +#define DRM_XE_GENL_SGUNIT_ERROR_CORRECTABLE (23) +#define DRM_XE_GENL_SGUNIT_ERROR_NONFATAL (24) +#define DRM_XE_GENL_SGUNIT_ERROR_FATAL (25) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMD (26) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMP (27) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_REQ (28) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_ANR_MDFI (29) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2T (30) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2C (31) +#define DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMD (32) +#define DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMP (33) +#define DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_REQ (34) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PUNIT (35) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMD (36) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMP (37) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_REQ (38) +#define DRM_XE_GENL_SOC_ERROR_FATAL_ANR_MDFI (39) +#define DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2T (40) +#define DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2C (41) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_AER (42) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_ERR (43) +#define DRM_XE_GENL_SOC_ERROR_FATAL_UR_COND (44) +#define DRM_XE_GENL_SOC_ERROR_FATAL_SERR_SRCS (45) + +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (ss, n)\ + (DRM_XE_GENL_SOC_ERROR_FATAL_SERR_SRCS + 0x1 + (ss) * 0x10 + (n)) +#define DRM_XE_GENL_SOC_ERROR_FATAL_HBM (ss, n)\ + (DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM(1, 15) + 0x1 + (ss) * 0x10 + (n)) + +/* 109 is the last ID used by SOC errors */ +#define DRM_XE_GENL_GSC_ERROR_CORRECTABLE_SRAM_ECC (110) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_SHUTDOWN (111) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_INTERNAL (112) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_SRAM_ECC (113) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_WDG_TIMEOUT (114) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_ROM_PARITY (115) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_UCODE_PARITY (116) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_VLT_GLITCH (117) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_PULL (118) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_CRC_CHECK (119) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_SELF_MBIST (120) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_AON_RF_PARITY (121) +#define DRM_XE_GENL_SGGI_ERROR_NONFATAL (122) +#define DRM_XE_GENL_SGLI_ERROR_NONFATAL (123) +#define DRM_XE_GENL_SGCI_ERROR_NONFATAL (124) +#define DRM_XE_GENL_MERT_ERROR_NONFATAL (125) +#define DRM_XE_GENL_SGGI_ERROR_FATAL (126) +#define DRM_XE_GENL_SGLI_ERROR_FATAL (127) +#define DRM_XE_GENL_SGCI_ERROR_FATAL (128) +#define DRM_XE_GENL_MERT_ERROR_FATAL (129) +#if defined(__cplusplus) }#endif Signature · 1 line
2.25.1
Re: [RFC v5 3/5] drm/xe/RAS: Expose the error counters
Quoted 49 lines · Rodrigo Vivi
Quoted 48 lines · Aravind Iddamsetty
We expose the various error counters supported on a hardware via genl subsytem through the registered commands to userspace. The
DRM_RAS_CMD_QUERYlists the error names with config id,DRM_RAD_CMD_READ_ONEreturns the counter value for the requested config id and theDRM_RAS_CMD_READ_ALLlists the counters for all errors along with their names and config ids.v2: Rebase
v3:
1. presentlyxe_list_errorsfills blank data for IGFX, prevent it by having an early check ofIS_DGFX(Michael J. Ruhl)
2. update errors from all sourcesv4: Check QUERY param, if its normal return not supported.
Reviewed-by Michael J. Ruhl <[email protected]>Signed-off-by Aravind Iddamsetty <[email protected]>--- drivers/gpu/drm/xe/xe_hw_error.c | 15 +- drivers/gpu/drm/xe/xe_netlink.c | 509 ++++++++++++++++++++++++++++++- include/uapi/drm/xe_drm.h | 85 ++++++ 3 files changed, 602 insertions(+), 7 deletions(-)drivers/gpu/drm/xe/xe_hw_error.c +8
-4@@ -531,16 +531,21 @@ static void xe_clear_all_soc_errors(struct xe_device *xe)while (hw_err <HARDWARE_ERROR_MAX ) {for (i =0 ; i <XE_SOC_NUM_IEH ; i++) -xe_mmio_write32 (>->tile->mmio,SOC_GSYSEVTCTL_REG (base, slave_base, i), +xe_mmio_write32 (>->tile->mmio, +SOC_GSYSEVTCTL_REG (base, slave_base, i), ~REG_BIT (hw_err)); -xe_mmio_write32 (>->tile->mmio,SOC_GLOBAL_ERR_STAT_MASTER_REG (base, hw_err), +xe_mmio_write32 (>->tile->mmio, +SOC_GLOBAL_ERR_STAT_MASTER_REG (base, hw_err),REG_GENMASK (31 ,0 )); -xe_mmio_write32 (>->tile->mmio,SOC_LOCAL_ERR_STAT_MASTER_REG (base, hw_err), +xe_mmio_write32 (>->tile->mmio, +SOC_LOCAL_ERR_STAT_MASTER_REG (base, hw_err),REG_GENMASK (31 ,0 )); - xe_mmio_write32(>->tile->mmio, SOC_GLOBAL_ERR_STAT_SLAVE_REG(slave_base, hw_err), + xe_mmio_write32(>->tile->mmio, + SOC_GLOBAL_ERR_STAT_SLAVE_REG(slave_base, hw_err),
Quoted 8 lines
drivers/gpu/drm/xe/xe_hw_error.c static void xe_clear_all_soc_errors(struct xe_device *xe) · line 549
Quoted 5 lines
drivers/gpu/drm/xe/xe_hw_error.c +2
-1REG_GENMASK(31, 0)); -xe_mmio_write32 (>->tile->mmio,SOC_LOCAL_ERR_STAT_SLAVE_REG (slave_base, hw_err), +xe_mmio_write32 (>->tile->mmio, +SOC_LOCAL_ERR_STAT_SLAVE_REG (slave_base, hw_err),REG_GENMASK (31 ,0 ));probably a fixup for the patch in the other series?
Which btw it was hard to understand the dependency. We should make this series indepentent of the other one.
agree this shouldn't have been part of this series, I might have messed it as part of rebasing.
Quoted 275 lines
Quoted 275 lines
drivers/gpu/drm/xe/xe_hw_error.c +0
-0hw_err++; }drivers/gpu/drm/xe/xe_netlink.c +262
-1@@ -6,16 +6,521 @@#include <net/genetlink.h> #include <uapi/drm/drm_netlink.h> +#include <drm/xe_drm.h> +#include "xe_device.h" -static int xe_genl_list_errors (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info) +#define MAX_ERROR_NAME 100 + +static const char *const xe_hw_error_events[] = { + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3_SNG ] ="correctable-l3-sng" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_GUC ] ="correctable-guc" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SAMPLER ] ="correctable-sampler" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SLM ] ="correctable-slm" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_IC ] ="correctable-eu-ic" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_GRF ] ="correctable-eu-grf" , + [DRM_XE_GENL_GT_ERROR_FATAL_ARR_BIST ] ="fatal-array-bist" , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_DOUB ] ="fatal-l3-double" , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_ECC_CHK ] ="fatal-l3-ecc-checker" , + [DRM_XE_GENL_GT_ERROR_FATAL_GUC ] ="fatal-guc" , + [DRM_XE_GENL_GT_ERROR_FATAL_IDI_PAR ] ="fatal-idi-parity" , + [DRM_XE_GENL_GT_ERROR_FATAL_SQIDI ] ="fatal-sqidi" , + [DRM_XE_GENL_GT_ERROR_FATAL_SAMPLER ] ="fatal-sampler" , + [DRM_XE_GENL_GT_ERROR_FATAL_SLM ] ="fatal-slm" , + [DRM_XE_GENL_GT_ERROR_FATAL_EU_IC ] ="fatal-eu-ic" , + [DRM_XE_GENL_GT_ERROR_FATAL_EU_GRF ] ="fatal-eu-grf" , + [DRM_XE_GENL_GT_ERROR_FATAL_FPU ] ="fatal-fpu" , + [DRM_XE_GENL_GT_ERROR_FATAL_TLB ] ="fatal-tlb" , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_FABRIC ] ="fatal-l3-fabric" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SUBSLICE ] ="correctable-subslice" , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3BANK ] ="correctable-l3bank" , + [DRM_XE_GENL_GT_ERROR_FATAL_SUBSLICE ] ="fatal-subslice" , + [DRM_XE_GENL_GT_ERROR_FATAL_L3BANK ] ="fatal-l3bank" , + [DRM_XE_GENL_SGUNIT_ERROR_CORRECTABLE ] ="sgunit-correctable" , + [DRM_XE_GENL_SGUNIT_ERROR_NONFATAL ] ="sgunit-nonfatal" , + [DRM_XE_GENL_SGUNIT_ERROR_FATAL ] ="sgunit-fatal" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMD ] ="soc-nonfatal-csc-psf-cmd-parity" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMP ] ="soc-nonfatal-csc-psf-unexpected-completion" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_REQ ] ="soc-nonfatal-csc-psf-unsupported-request" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_ANR_MDFI ] ="soc-nonfatal-anr-mdfi" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2T ] ="soc-nonfatal-mdfi-t2t" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2C ] ="soc-nonfatal-mdfi-t2c" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,0 )] ="soc-nonfatal-hbm-ss0-0" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,1 )] ="soc-nonfatal-hbm-ss0-1" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,2 )] ="soc-nonfatal-hbm-ss0-2" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,3 )] ="soc-nonfatal-hbm-ss0-3" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,4 )] ="soc-nonfatal-hbm-ss0-4" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,5 )] ="soc-nonfatal-hbm-ss0-5" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,6 )] ="soc-nonfatal-hbm-ss0-6" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,7 )] ="soc-nonfatal-hbm-ss0-7" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,8 )] ="soc-nonfatal-hbm-ss1-0" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,9 )] ="soc-nonfatal-hbm-ss1-1" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,10 )] ="soc-nonfatal-hbm-ss1-2" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,11 )] ="soc-nonfatal-hbm-ss1-3" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,12 )] ="soc-nonfatal-hbm-ss1-4" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,13 )] ="soc-nonfatal-hbm-ss1-5" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,14 )] ="soc-nonfatal-hbm-ss1-6" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,15 )] ="soc-nonfatal-hbm-ss1-7" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,0 )] ="soc-nonfatal-hbm-ss2-0" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,1 )] ="soc-nonfatal-hbm-ss2-1" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,2 )] ="soc-nonfatal-hbm-ss2-2" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,3 )] ="soc-nonfatal-hbm-ss2-3" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,4 )] ="soc-nonfatal-hbm-ss2-4" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,5 )] ="soc-nonfatal-hbm-ss2-5" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,6 )] ="soc-nonfatal-hbm-ss2-6" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,7 )] ="soc-nonfatal-hbm-ss2-7" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,8 )] ="soc-nonfatal-hbm-ss3-0" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,9 )] ="soc-nonfatal-hbm-ss3-1" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,10 )] ="soc-nonfatal-hbm-ss3-2" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,11 )] ="soc-nonfatal-hbm-ss3-3" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,12 )] ="soc-nonfatal-hbm-ss3-4" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,13 )] ="soc-nonfatal-hbm-ss3-5" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,14 )] ="soc-nonfatal-hbm-ss3-6" , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,15 )] ="soc-nonfatal-hbm-ss3-7" , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMD ] ="soc-fatal-csc-psf-cmd-parity" , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMP ] ="soc-fatal-csc-psf-unexpected-completion" , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_REQ ] ="soc-fatal-csc-psf-unsupported-request" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PUNIT ] ="soc-fatal-punit" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMD ] ="soc-fatal-pcie-psf-command-parity" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMP ] ="soc-fatal-pcie-psf-unexpected-completion" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_REQ ] ="soc-fatal-pcie-psf-unsupported-request" , + [DRM_XE_GENL_SOC_ERROR_FATAL_ANR_MDFI ] ="soc-fatal-anr-mdfi" , + [DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2T ] ="soc-fatal-mdfi-t2t" , + [DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2C ] ="soc-fatal-mdfi-t2c" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_AER ] ="soc-fatal-malformed-pcie-aer" , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_ERR ] ="soc-fatal-malformed-pcie-err" , + [DRM_XE_GENL_SOC_ERROR_FATAL_UR_COND ] ="soc-fatal-ur-condition-ieh" , + [DRM_XE_GENL_SOC_ERROR_FATAL_SERR_SRCS ] ="soc-fatal-from-serr-sources" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,0 )] ="soc-fatal-hbm-ss0-0" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,1 )] ="soc-fatal-hbm-ss0-1" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,2 )] ="soc-fatal-hbm-ss0-2" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,3 )] ="soc-fatal-hbm-ss0-3" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,4 )] ="soc-fatal-hbm-ss0-4" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,5 )] ="soc-fatal-hbm-ss0-5" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,6 )] ="soc-fatal-hbm-ss0-6" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,7 )] ="soc-fatal-hbm-ss0-7" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,8 )] ="soc-fatal-hbm-ss1-0" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,9 )] ="soc-fatal-hbm-ss1-1" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,10 )] ="soc-fatal-hbm-ss1-2" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,11 )] ="soc-fatal-hbm-ss1-3" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,12 )] ="soc-fatal-hbm-ss1-4" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,13 )] ="soc-fatal-hbm-ss1-5" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,14 )] ="soc-fatal-hbm-ss1-6" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,15 )] ="soc-fatal-hbm-ss1-7" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,0 )] ="soc-fatal-hbm-ss2-0" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,1 )] ="soc-fatal-hbm-ss2-1" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,2 )] ="soc-fatal-hbm-ss2-2" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,3 )] ="soc-fatal-hbm-ss2-3" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,4 )] ="soc-fatal-hbm-ss2-4" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,5 )] ="soc-fatal-hbm-ss2-5" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,6 )] ="soc-fatal-hbm-ss2-6" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,7 )] ="soc-fatal-hbm-ss2-7" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,8 )] ="soc-fatal-hbm-ss3-0" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,9 )] ="soc-fatal-hbm-ss3-1" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,10 )] ="soc-fatal-hbm-ss3-2" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,11 )] ="soc-fatal-hbm-ss3-3" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,12 )] ="soc-fatal-hbm-ss3-4" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,13 )] ="soc-fatal-hbm-ss3-5" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,14 )] ="soc-fatal-hbm-ss3-6" , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,15 )] ="soc-fatal-hbm-ss3-7" , + [DRM_XE_GENL_GSC_ERROR_CORRECTABLE_SRAM_ECC ] ="gsc-correctable-sram-ecc" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_SHUTDOWN ] ="gsc-nonfatal-mia-shutdown" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_INTERNAL ] ="gsc-nonfatal-mia-internal" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_SRAM_ECC ] ="gsc-nonfatal-sram-ecc" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_WDG_TIMEOUT ] ="gsc-nonfatal-wdg-timeout" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_ROM_PARITY ] ="gsc-nonfatal-rom-parity" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_UCODE_PARITY ] ="gsc-nonfatal-ucode-parity" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_VLT_GLITCH ] ="gsc-nonfatal-vlt-glitch" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_PULL ] ="gsc-nonfatal-fuse-pull" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_CRC_CHECK ] ="gsc-nonfatal-fuse-crc-check" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_SELF_MBIST ] ="gsc-nonfatal-self-mbist" , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_AON_RF_PARITY ] ="gsc-nonfatal-aon-parity" , + [DRM_XE_GENL_SGGI_ERROR_NONFATAL ] ="sggi-nonfatal-data-parity" , + [DRM_XE_GENL_SGLI_ERROR_NONFATAL ] ="sgli-nonfatal-data-parity" , + [DRM_XE_GENL_SGCI_ERROR_NONFATAL ] ="sgci-nonfatal-data-parity" , + [DRM_XE_GENL_MERT_ERROR_NONFATAL ] ="mert-nonfatal-data-parity" , + [DRM_XE_GENL_SGGI_ERROR_FATAL ] ="sggi-fatal-data-parity" , + [DRM_XE_GENL_SGLI_ERROR_FATAL ] ="sgli-fatal-data-parity" , + [DRM_XE_GENL_SGCI_ERROR_FATAL ] ="sgci-fatal-data-parity" , + [DRM_XE_GENL_MERT_ERROR_FATAL ] ="mert-nonfatal-data-parity" , +}; + +static const unsigned long xe_hw_error_map[] = { + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3_SNG ] =XE_HW_ERR_GT_CORR_L3_SNG , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_GUC ] =XE_HW_ERR_GT_CORR_GUC , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SAMPLER ] =XE_HW_ERR_GT_CORR_SAMPLER , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SLM ] =XE_HW_ERR_GT_CORR_SLM , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_IC ] =XE_HW_ERR_GT_CORR_EU_IC , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_GRF ] =XE_HW_ERR_GT_CORR_EU_GRF , + [DRM_XE_GENL_GT_ERROR_FATAL_ARR_BIST ] =XE_HW_ERR_GT_FATAL_ARR_BIST , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_DOUB ] =XE_HW_ERR_GT_FATAL_L3_DOUB , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_ECC_CHK ] =XE_HW_ERR_GT_FATAL_L3_ECC_CHK , + [DRM_XE_GENL_GT_ERROR_FATAL_GUC ] =XE_HW_ERR_GT_FATAL_GUC , + [DRM_XE_GENL_GT_ERROR_FATAL_IDI_PAR ] =XE_HW_ERR_GT_FATAL_IDI_PAR , + [DRM_XE_GENL_GT_ERROR_FATAL_SQIDI ] =XE_HW_ERR_GT_FATAL_SQIDI , + [DRM_XE_GENL_GT_ERROR_FATAL_SAMPLER ] =XE_HW_ERR_GT_FATAL_SAMPLER , + [DRM_XE_GENL_GT_ERROR_FATAL_SLM ] =XE_HW_ERR_GT_FATAL_SLM , + [DRM_XE_GENL_GT_ERROR_FATAL_EU_IC ] =XE_HW_ERR_GT_FATAL_EU_IC , + [DRM_XE_GENL_GT_ERROR_FATAL_EU_GRF ] =XE_HW_ERR_GT_FATAL_EU_GRF , + [DRM_XE_GENL_GT_ERROR_FATAL_FPU ] =XE_HW_ERR_GT_FATAL_FPU , + [DRM_XE_GENL_GT_ERROR_FATAL_TLB ] =XE_HW_ERR_GT_FATAL_TLB , + [DRM_XE_GENL_GT_ERROR_FATAL_L3_FABRIC ] =XE_HW_ERR_GT_FATAL_L3_FABRIC , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_SUBSLICE ] =XE_HW_ERR_GT_CORR_SUBSLICE , + [DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3BANK ] =XE_HW_ERR_GT_CORR_L3BANK , + [DRM_XE_GENL_GT_ERROR_FATAL_SUBSLICE ] =XE_HW_ERR_GT_FATAL_SUBSLICE , + [DRM_XE_GENL_GT_ERROR_FATAL_L3BANK ] =XE_HW_ERR_GT_FATAL_L3BANK , + [DRM_XE_GENL_SGUNIT_ERROR_CORRECTABLE ] =XE_HW_ERR_TILE_CORR_SGUNIT , + [DRM_XE_GENL_SGUNIT_ERROR_NONFATAL ] =XE_HW_ERR_TILE_NONFATAL_SGUNIT , + [DRM_XE_GENL_SGUNIT_ERROR_FATAL ] =XE_HW_ERR_TILE_FATAL_SGUNIT , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMD ] =XE_HW_ERR_SOC_NONFATAL_CSC_PSF_CMD , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMP ] =XE_HW_ERR_SOC_NONFATAL_CSC_PSF_CMP , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_REQ ] =XE_HW_ERR_SOC_NONFATAL_CSC_PSF_REQ , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_ANR_MDFI ] =XE_HW_ERR_SOC_NONFATAL_ANR_MDFI , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2T ] =XE_HW_ERR_SOC_NONFATAL_MDFI_T2T , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2C ] =XE_HW_ERR_SOC_NONFATAL_MDFI_T2C , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,0 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,1 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,2 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,3 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,4 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,5 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,6 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,7 )] =XE_HW_ERR_SOC_NONFATAL_HBM0_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,8 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,9 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,10 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,11 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,12 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,13 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,14 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (0 ,15 )] =XE_HW_ERR_SOC_NONFATAL_HBM1_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,0 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,1 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,2 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,3 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,4 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,5 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,6 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,7 )] =XE_HW_ERR_SOC_NONFATAL_HBM2_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,8 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,9 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,10 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,11 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,12 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,13 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,14 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (1 ,15 )] =XE_HW_ERR_SOC_NONFATAL_HBM3_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMD ] =XE_HW_ERR_SOC_FATAL_CSC_PSF_CMD , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMP ] =XE_HW_ERR_SOC_FATAL_CSC_PSF_CMP , + [DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_REQ ] =XE_HW_ERR_SOC_FATAL_CSC_PSF_REQ , + [DRM_XE_GENL_SOC_ERROR_FATAL_PUNIT ] =XE_HW_ERR_SOC_FATAL_PUNIT , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMD ] =XE_HW_ERR_SOC_FATAL_PCIE_PSF_CMD , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMP ] =XE_HW_ERR_SOC_FATAL_PCIE_PSF_CMP , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_REQ ] =XE_HW_ERR_SOC_FATAL_PCIE_PSF_REQ , + [DRM_XE_GENL_SOC_ERROR_FATAL_ANR_MDFI ] =XE_HW_ERR_SOC_FATAL_ANR_MDFI , + [DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2T ] =XE_HW_ERR_SOC_FATAL_MDFI_T2T , + [DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2C ] =XE_HW_ERR_SOC_FATAL_MDFI_T2C , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_AER ] =XE_HW_ERR_SOC_FATAL_PCIE_AER , + [DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_ERR ] =XE_HW_ERR_SOC_FATAL_PCIE_ERR , + [DRM_XE_GENL_SOC_ERROR_FATAL_UR_COND ] =XE_HW_ERR_SOC_FATAL_UR_COND , + [DRM_XE_GENL_SOC_ERROR_FATAL_SERR_SRCS ] =XE_HW_ERR_SOC_FATAL_SERR_SRCS , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,0 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,1 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,2 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,3 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,4 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,5 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,6 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,7 )] =XE_HW_ERR_SOC_FATAL_HBM0_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,8 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,9 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,10 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,11 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,12 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,13 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,14 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (0 ,15 )] =XE_HW_ERR_SOC_FATAL_HBM1_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,0 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,1 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,2 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,3 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,4 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,5 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,6 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,7 )] =XE_HW_ERR_SOC_FATAL_HBM2_CHNL7 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,8 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL0 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,9 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL1 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,10 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL2 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,11 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL3 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,12 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL4 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,13 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL5 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,14 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL6 , + [DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,15 )] =XE_HW_ERR_SOC_FATAL_HBM3_CHNL7 , + [DRM_XE_GENL_GSC_ERROR_CORRECTABLE_SRAM_ECC ] =XE_HW_ERR_GSC_CORR_SRAM , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_SHUTDOWN ] =XE_HW_ERR_GSC_NONFATAL_MIA_SHUTDOWN , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_INTERNAL ] =XE_HW_ERR_GSC_NONFATAL_MIA_INTERNAL , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_SRAM_ECC ] =XE_HW_ERR_GSC_NONFATAL_SRAM , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_WDG_TIMEOUT ] =XE_HW_ERR_GSC_NONFATAL_WDG , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_ROM_PARITY ] =XE_HW_ERR_GSC_NONFATAL_ROM_PARITY , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_UCODE_PARITY ] =XE_HW_ERR_GSC_NONFATAL_UCODE_PARITY , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_VLT_GLITCH ] =XE_HW_ERR_GSC_NONFATAL_VLT_GLITCH , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_PULL ] =XE_HW_ERR_GSC_NONFATAL_FUSE_PULL , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_CRC_CHECK ] =XE_HW_ERR_GSC_NONFATAL_FUSE_CRC , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_SELF_MBIST ] =XE_HW_ERR_GSC_NONFATAL_SELF_MBIST , + [DRM_XE_GENL_GSC_ERROR_NONFATAL_AON_RF_PARITY ] =XE_HW_ERR_GSC_NONFATAL_AON_RF_PARITY , + [DRM_XE_GENL_SGGI_ERROR_NONFATAL ] =XE_HW_ERR_TILE_NONFATAL_SGGI , + [DRM_XE_GENL_SGLI_ERROR_NONFATAL ] =XE_HW_ERR_TILE_NONFATAL_SGLI ,
Quoted 8 lines
drivers/gpu/drm/xe/xe_netlink.c line 279
Quoted 7 lines
drivers/gpu/drm/xe/xe_netlink.c +7
-0+ [DRM_XE_GENL_SGCI_ERROR_NONFATAL] = XE_HW_ERR_TILE_NONFATAL_SGCI, + [DRM_XE_GENL_MERT_ERROR_NONFATAL] = XE_HW_ERR_TILE_NONFATAL_MERT, + [DRM_XE_GENL_SGGI_ERROR_FATAL] = XE_HW_ERR_TILE_FATAL_SGGI, + [DRM_XE_GENL_SGLI_ERROR_FATAL] = XE_HW_ERR_TILE_FATAL_SGLI, + [DRM_XE_GENL_SGCI_ERROR_FATAL] = XE_HW_ERR_TILE_FATAL_SGCI, + [DRM_XE_GENL_MERT_ERROR_FATAL] = XE_HW_ERR_TILE_FATAL_MERT, +};probably deserves a separate header file?
is that to have better readability? as the header will ever be used only in this file will it have any advantage?
Quoted 242 lines
Quoted 242 lines
drivers/gpu/drm/xe/xe_netlink.c +234
-1+ +static unsigned int config_gt_id (const u64 config) +{ +return config >> __XE_GENL_GT_SHIFT; +} + +static u64 config_counter (const u64 config) +{ +return config & ~(~0ULL << __XE_GENL_GT_SHIFT); +} + +static bool is_gt_error (const u64 config) +{ +unsigned int error; + + error =config_counter (config); +if (error <=DRM_XE_GENL_GT_ERROR_FATAL_FPU ) +return true; + +return false; +} + +static bool is_gt_vector_error (const u64 config) +{ +unsigned int error; + + error =config_counter (config); +if (error >=DRM_XE_GENL_GT_ERROR_FATAL_TLB && + error <=DRM_XE_GENL_GT_ERROR_FATAL_L3BANK ) +return true; + +return false; +} + +static bool is_pvc_invalid_gt_errors (const u64 config) +{ +switch (config_counter (config)) { +case DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3_SNG : +case DRM_XE_GENL_GT_ERROR_CORRECTABLE_SAMPLER : +case DRM_XE_GENL_GT_ERROR_FATAL_ARR_BIST : +case DRM_XE_GENL_GT_ERROR_FATAL_L3_DOUB : +case DRM_XE_GENL_GT_ERROR_FATAL_L3_ECC_CHK : +case DRM_XE_GENL_GT_ERROR_FATAL_IDI_PAR : +case DRM_XE_GENL_GT_ERROR_FATAL_SQIDI : +case DRM_XE_GENL_GT_ERROR_FATAL_SAMPLER : +case DRM_XE_GENL_GT_ERROR_FATAL_EU_IC : +return true; +default : +return false; + } +} + +static bool is_gsc_hw_error (const u64 config) +{ +if (config_counter (config) >=DRM_XE_GENL_GSC_ERROR_CORRECTABLE_SRAM_ECC && +config_counter (config) <=DRM_XE_GENL_GSC_ERROR_NONFATAL_AON_RF_PARITY ) +return true; + +return false; +} + +static bool is_soc_error (const u64 config) { +if (config_counter (config) >=DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMD && +config_counter (config) <=DRM_XE_GENL_SOC_ERROR_FATAL_HBM (1 ,15 )) +return true; + +return false; +} + +static int +config_status (struct xe_device *xe,u64 config) +{ +unsigned int gt_id =config_gt_id (config); +struct xe_gt *gt =xe_device_get_gt (xe, gt_id); + +if (!IS_DGFX (xe)) +return -ENODEV ; + +if (gt->info.type ==XE_GT_TYPE_UNINITIALIZED ) +return -ENOENT ; + +/* GSC HW ERRORS are present on root tile of +* platform supporting MEMORY SPARING only +*/ +if (is_gsc_hw_error (config) && !(xe->info.platform ==XE_PVC && !gt_id)) +return -ENODEV ; + +/* GT vectors error are valid on Platforms supporting error vectors only */ +if (is_gt_vector_error (config) && xe->info.platform !=XE_PVC ) +return -ENODEV ; + +/* Skip gt errors not supported on pvc */ +if (is_pvc_invalid_gt_errors (config) && xe->info.platform ==XE_PVC ) +return -ENODEV ; + +/* FATAL FPU error is valid on PVC only */ +if (config_counter (config) ==DRM_XE_GENL_GT_ERROR_FATAL_FPU && + !(xe->info.platform ==XE_PVC )) +return -ENODEV ; + +if (is_soc_error (config) && !(xe->info.platform ==XE_PVC )) +return -ENODEV ; + +return (config_counter (config) >= +ARRAY_SIZE (xe_hw_error_map)) ? -ENOENT :0 ; +} + +static u64 get_counter_value (struct xe_device *xe,u64 config) +{ +const unsigned int gt_id =config_gt_id (config); +struct xe_gt *gt =xe_device_get_gt (xe, gt_id); +unsigned int id =config_counter (config); + +if (is_gt_error (config) ||is_gt_vector_error (config)) +return xa_to_value (xa_load (>->errors.hw_error, xe_hw_error_map[id])); + +return xa_to_value (xa_load (>->tile->errors.hw_error, xe_hw_error_map[id])); +} + +static int fill_error_details (struct xe_device *xe,struct genl_info *info,struct sk_buff *new_msg) +{ +struct nlattr *entry_attr; +bool counter = false; +struct xe_gt *gt; +int i, j; + +BUILD_BUG_ON (ARRAY_SIZE (xe_hw_error_events) != +ARRAY_SIZE (xe_hw_error_map)); + +if (info->genlhdr->cmd ==DRM_RAS_CMD_READ_ALL ) + counter = true; + + entry_attr =nla_nest_start (new_msg,DRM_RAS_ATTR_QUERY_REPLY ); +if (!entry_attr) +return -EMSGSIZE ; + +for_each_gt (gt, xe, j) { +char str[MAX_ERROR_NAME ]; +u64 val; + +for (i =0 ; i <ARRAY_SIZE (xe_hw_error_events); i++) { +u64 config =DRM_XE_HW_ERROR (j, i); + +if (config_status (xe, config)) +continue ; + +/* should this be cleared everytime */ +snprintf (str,sizeof (str),"error-gt%d-%s" , j, xe_hw_error_events[i]); + +if (nla_put_string (new_msg,DRM_RAS_ATTR_ERROR_NAME , str)) + goto err; +if (nla_put_u64_64bit (new_msg,DRM_RAS_ATTR_ERROR_ID , config,DRM_ATTR_PAD )) + goto err; +if (counter) { + val =get_counter_value (xe, config); +if (nla_put_u64_64bit (new_msg,DRM_RAS_ATTR_ERROR_VALUE , val, +DRM_ATTR_PAD )) + goto err; + } + } + } + +nla_nest_end (new_msg, entry_attr); +return 0 ; +err: +drm_dbg_driver (&xe->drm,"msg buff is small\n" ); +nla_nest_cancel (new_msg, entry_attr); +nlmsg_free (new_msg); + +return -EMSGSIZE ; +} + +static int xe_genl_list_errors (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info) +{ +struct xe_device *xe =to_xe_device (drm); +size_t msg_size =NLMSG_DEFAULT_SIZE ; +enum drm_cmd_request_type query_type; +struct sk_buff *new_msg; +int retries =2 ; +void *usrhdr; +int ret =0 ; + +if (!IS_DGFX (xe)) +return -ENODEV ; + +/* Support verbose only errors */ + query_type =nla_get_u8 (info->attrs[DRM_RAS_ATTR_QUERY ]); +if (query_type ==DRM_RAS_CMD_QUERY_NORMAL ) +return -EOPNOTSUPP ; + +do { + new_msg =drm_genl_alloc_msg (drm, info, msg_size, &usrhdr); +if (!new_msg) +return -ENOMEM ; + + ret =fill_error_details (xe, info, new_msg); +if (!ret) +break ; + + msg_size +=NLMSG_DEFAULT_SIZE ; + }while (retries--); + +if (!ret) + ret =drm_genl_reply (new_msg, info, usrhdr); + +return ret; }static int xe_genl_read_error (struct drm_device *drm,struct sk_buff *msg,struct genl_info *info) { -return 0 ; +struct xe_device *xe =to_xe_device (drm); +size_t msg_size =NLMSG_DEFAULT_SIZE ; +struct sk_buff *new_msg; +void *usrhdr; +int ret =0 ; +int retries =2 ; +u64 config, val; + +if (info->genlhdr->cmd ==DRM_RAS_CMD_READ_BLOCK ) +return -EOPNOTSUPP ; + + config =nla_get_u64 (info->attrs[DRM_RAS_ATTR_ERROR_ID ]); + ret =config_status (xe, config); +if (ret) +return ret; +do { + new_msg =drm_genl_alloc_msg (drm, info, msg_size, &usrhdr); +if (!new_msg) +return -ENOMEM ; + + val =get_counter_value (xe, config); +if (nla_put_u64_64bit (new_msg,DRM_RAS_ATTR_ERROR_VALUE , val,DRM_ATTR_PAD )) { + msg_size +=NLMSG_DEFAULT_SIZE ; +continue ; + } + +break ; + }while (retries--);
Quoted 8 lines
drivers/gpu/drm/xe/xe_netlink.c line 488
Quoted 5 lines
drivers/gpu/drm/xe/xe_netlink.c +4
-0+ + ret =drm_genl_reply (new_msg, info, usrhdr); + +return ret; }this .c without any public function and no .h was what draw my attention that there was something wrong with the registration process. We need to have init and finish per component.
I had designed these similar to drm_driver ops, even with the new registration model i believe they would still be static.
Is there a concern?
Quoted 20 lines
Quoted 8 lines
drivers/gpu/drm/xe/xe_netlink.c +0
-0/* driver callbacks to DRM netlink commands*/ include/uapi/drm/xe_drm.h +0
-0include/uapi/drm/xe_drm.h struct drm_xe_query_eu_stall { · line 1983
Quoted 10 lines
include/uapi/drm/xe_drm.h +7
-0@@ -1974,6 +1974,91 @@ struct drm_xe_query_eu_stall {__u64 sampling_rates[]; }; +/* +* Top bits of every counter are GT id. +*/ +#define __XE_GENL_GT_SHIFT (56) +/** +* DOC: XE GENL netlink event IDs +* TODO: Add more details yes, please it is hard to understand why here.
Quoted 3 lines
And also I have the feeling that this deserves to be together with the other definitions above all together in a separate header. perhaps even per platform?!
this is a UAPI the below IDs are used by drm_ras tool to read a particular error counter.
so felt these should be part of this header. But, why separate headers aren't UAPI
stuff be in this header instead.
Thanks,
Aravind.
Quoted 85 lines
Quoted 85 lines
include/uapi/drm/xe_drm.h +78
-0+*/ +#define DRM_XE_HW_ERROR (gt, id) \ + ((id) | ((__u64)(gt) << __XE_GENL_GT_SHIFT)) + +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3_SNG (0) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_GUC (1) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_SAMPLER (2) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_SLM (3) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_IC (4) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_EU_GRF (5) +#define DRM_XE_GENL_GT_ERROR_FATAL_ARR_BIST (6) +#define DRM_XE_GENL_GT_ERROR_FATAL_L3_DOUB (7) +#define DRM_XE_GENL_GT_ERROR_FATAL_L3_ECC_CHK (8) +#define DRM_XE_GENL_GT_ERROR_FATAL_GUC (9) +#define DRM_XE_GENL_GT_ERROR_FATAL_IDI_PAR (10) +#define DRM_XE_GENL_GT_ERROR_FATAL_SQIDI (11) +#define DRM_XE_GENL_GT_ERROR_FATAL_SAMPLER (12) +#define DRM_XE_GENL_GT_ERROR_FATAL_SLM (13) +#define DRM_XE_GENL_GT_ERROR_FATAL_EU_IC (14) +#define DRM_XE_GENL_GT_ERROR_FATAL_EU_GRF (15) +#define DRM_XE_GENL_GT_ERROR_FATAL_FPU (16) +#define DRM_XE_GENL_GT_ERROR_FATAL_TLB (17) +#define DRM_XE_GENL_GT_ERROR_FATAL_L3_FABRIC (18) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_SUBSLICE (19) +#define DRM_XE_GENL_GT_ERROR_CORRECTABLE_L3BANK (20) +#define DRM_XE_GENL_GT_ERROR_FATAL_SUBSLICE (21) +#define DRM_XE_GENL_GT_ERROR_FATAL_L3BANK (22) +#define DRM_XE_GENL_SGUNIT_ERROR_CORRECTABLE (23) +#define DRM_XE_GENL_SGUNIT_ERROR_NONFATAL (24) +#define DRM_XE_GENL_SGUNIT_ERROR_FATAL (25) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMD (26) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_CMP (27) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_CSC_PSF_REQ (28) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_ANR_MDFI (29) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2T (30) +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_MDFI_T2C (31) +#define DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMD (32) +#define DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_CMP (33) +#define DRM_XE_GENL_SOC_ERROR_FATAL_CSC_PSF_REQ (34) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PUNIT (35) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMD (36) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_CMP (37) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_PSF_REQ (38) +#define DRM_XE_GENL_SOC_ERROR_FATAL_ANR_MDFI (39) +#define DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2T (40) +#define DRM_XE_GENL_SOC_ERROR_FATAL_MDFI_T2C (41) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_AER (42) +#define DRM_XE_GENL_SOC_ERROR_FATAL_PCIE_ERR (43) +#define DRM_XE_GENL_SOC_ERROR_FATAL_UR_COND (44) +#define DRM_XE_GENL_SOC_ERROR_FATAL_SERR_SRCS (45) + +#define DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM (ss, n)\ + (DRM_XE_GENL_SOC_ERROR_FATAL_SERR_SRCS + 0x1 + (ss) * 0x10 + (n)) +#define DRM_XE_GENL_SOC_ERROR_FATAL_HBM (ss, n)\ + (DRM_XE_GENL_SOC_ERROR_NONFATAL_HBM(1, 15) + 0x1 + (ss) * 0x10 + (n)) + +/* 109 is the last ID used by SOC errors */ +#define DRM_XE_GENL_GSC_ERROR_CORRECTABLE_SRAM_ECC (110) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_SHUTDOWN (111) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_MIA_INTERNAL (112) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_SRAM_ECC (113) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_WDG_TIMEOUT (114) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_ROM_PARITY (115) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_UCODE_PARITY (116) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_VLT_GLITCH (117) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_PULL (118) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_FUSE_CRC_CHECK (119) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_SELF_MBIST (120) +#define DRM_XE_GENL_GSC_ERROR_NONFATAL_AON_RF_PARITY (121) +#define DRM_XE_GENL_SGGI_ERROR_NONFATAL (122) +#define DRM_XE_GENL_SGLI_ERROR_NONFATAL (123) +#define DRM_XE_GENL_SGCI_ERROR_NONFATAL (124) +#define DRM_XE_GENL_MERT_ERROR_NONFATAL (125) +#define DRM_XE_GENL_SGGI_ERROR_FATAL (126) +#define DRM_XE_GENL_SGLI_ERROR_FATAL (127) +#define DRM_XE_GENL_SGCI_ERROR_FATAL (128) +#define DRM_XE_GENL_MERT_ERROR_FATAL (129) +#if defined(__cplusplus) }#endif Signature · 1 line
2.25.1
RFC v5 4/5 drm/netlink: Define multicast groups
In reply to Aravind IddamsettyNetlink subsystem supports event notifications to userspace. we define two multicast groups for correctable and uncorrectable errors to which userspace can subscribe and be notified when any of those errors happen. The group names are local to the driver's genl netlink family.
--- drivers/gpu/drm/drm_netlink.c | 7 +++++++ include/drm/drm_netlink.h | 5 +++++ include/uapi/drm/drm_netlink.h | 4 ++++ 3 files changed, 16 insertions(+)
drivers/gpu/drm/drm_netlink.c +7 -0
@@ -15,6 +15,11 @@DEFINE_XARRAY (drm_dev_xarray); +static const struct genl_multicast_group drm_event_mcgrps[] = { + [DRM_GENL_MCAST_CORR_ERR ] = { .name =DRM_GENL_MCAST_GROUP_NAME_CORR_ERR , }, + [DRM_GENL_MCAST_UNCORR_ERR ] = { .name =DRM_GENL_MCAST_GROUP_NAME_UNCORR_ERR , }, +}; +/** * drm_genl_reply - response to a request * @msg: socket buffer
@@ -156,6 +161,8 @@ static void drm_genl_family_init(struct drm_device *dev) dev->drm_genl_family->ops = drm_genl_ops; dev->drm_genl_family->n_ops =ARRAY_SIZE (drm_genl_ops); dev->drm_genl_family->maxattr =DRM_ATTR_MAX ; + dev->drm_genl_family->mcgrps = drm_event_mcgrps; + dev->drm_genl_family->n_mcgrps =ARRAY_SIZE (drm_event_mcgrps); dev->drm_genl_family->module = dev->dev->driver->owner; }
include/drm/drm_netlink.h +5 -0
@@ -12,6 +12,11 @@ struct drm_device;struct genl_info ;struct sk_buff ; +enum mcgrps_events { +DRM_GENL_MCAST_CORR_ERR , +DRM_GENL_MCAST_UNCORR_ERR , +}; +struct driver_genl_ops { int (*doit)(struct drm_device *dev, struct sk_buff *skb,
include/uapi/drm/drm_netlink.h +4 -0
@@ -26,6 +26,8 @@#define _DRM_NETLINK_H_#define DRM_GENL_VERSION 1 +#define DRM_GENL_MCAST_GROUP_NAME_CORR_ERR "drm_corr_err" +#define DRM_GENL_MCAST_GROUP_NAME_UNCORR_ERR "drm_uncorr_err"#if defined(__cplusplus)extern "C" {
@@ -50,6 +52,8 @@ enum drm_genl_error_cmds {DRM_RAS_CMD_READ_BLOCK ,/** @DRM_RAS_CMD_READ_ALL: Command to get counters of all errors */ DRM_RAS_CMD_READ_ALL , +/** @DRM_RAS_CMD_ERROR_EVENT: Command sent as part of multicast event */ +DRM_RAS_CMD_ERROR_EVENT , __DRM_CMD_MAX, DRM_CMD_MAX = __DRM_CMD_MAX - 1,
Signature · 1 line
2.25.1
Rodrigo Vivi
maintainerRe: [RFC v5 4/5] drm/netlink: Define multicast groups
Quoted 18 lines · Aravind Iddamsetty
Netlink subsystem supports event notifications to userspace. we define two multicast groups for correctable and uncorrectable errors to which userspace can subscribe and be notified when any of those errors happen. The group names are local to the driver's genl netlink family.
Reviewed-by Michael J. Ruhl <[email protected]>Signed-off-by Aravind Iddamsetty <[email protected]>--- drivers/gpu/drm/drm_netlink.c | 7 +++++++ include/drm/drm_netlink.h | 5 +++++ include/uapi/drm/drm_netlink.h | 4 ++++ 3 files changed, 16 insertions(+)drivers/gpu/drm/drm_netlink.c +0
-0
drivers/gpu/drm/drm_netlink.c line 21
Quoted 7 lines
drivers/gpu/drm/drm_netlink.c +4
-0@@ -15,6 +15,11 @@DEFINE_XARRAY (drm_dev_xarray); +static const struct genl_multicast_group drm_event_mcgrps[] = { + [DRM_GENL_MCAST_CORR_ERR ] = { .name =DRM_GENL_MCAST_GROUP_NAME_CORR_ERR , }, + [DRM_GENL_MCAST_UNCORR_ERR ] = { .name =DRM_GENL_MCAST_GROUP_NAME_UNCORR_ERR , }, +};
this was the thing I thought for that 'monitor' but well, that can be ignored and we can indeed leave this per error component. but I also don't like it to be forced. It should be a driver definition and driver adoption.
Quoted 54 lines
drivers/gpu/drm/drm_netlink.c +3
-0+/** * drm_genl_reply - response to a request * @msg: socket buffer @@ -156,6 +161,8 @@ static void drm_genl_family_init(struct drm_device *dev) dev->drm_genl_family->ops = drm_genl_ops; dev->drm_genl_family->n_ops =ARRAY_SIZE (drm_genl_ops); dev->drm_genl_family->maxattr =DRM_ATTR_MAX ; + dev->drm_genl_family->mcgrps = drm_event_mcgrps; + dev->drm_genl_family->n_mcgrps =ARRAY_SIZE (drm_event_mcgrps); dev->drm_genl_family->module = dev->dev->driver->owner; }include/drm/drm_netlink.h +5
-0@@ -12,6 +12,11 @@ struct drm_device;struct genl_info ;struct sk_buff ; +enum mcgrps_events { +DRM_GENL_MCAST_CORR_ERR , +DRM_GENL_MCAST_UNCORR_ERR , +}; +struct driver_genl_ops { int (*doit)(struct drm_device *dev, struct sk_buff *skb,include/uapi/drm/drm_netlink.h +4
-0@@ -26,6 +26,8 @@#define _DRM_NETLINK_H_#define DRM_GENL_VERSION 1 +#define DRM_GENL_MCAST_GROUP_NAME_CORR_ERR "drm_corr_err" +#define DRM_GENL_MCAST_GROUP_NAME_UNCORR_ERR "drm_uncorr_err"#if defined(__cplusplus)extern "C" {@@ -50,6 +52,8 @@ enum drm_genl_error_cmds {DRM_RAS_CMD_READ_BLOCK ,/** @DRM_RAS_CMD_READ_ALL: Command to get counters of all errors */ DRM_RAS_CMD_READ_ALL , +/** @DRM_RAS_CMD_ERROR_EVENT: Command sent as part of multicast event */ +DRM_RAS_CMD_ERROR_EVENT , __DRM_CMD_MAX, DRM_CMD_MAX = __DRM_CMD_MAX - 1,Signature · 1 line
2.25.1
RFC v5 5/5 drm/xe/RAS: send multicast event on occurrence of an error
In reply to Aravind IddamsettyWhenever a correctable or an uncorrectable error happens an event is sent to the corresponding listeners of these groups.
v2: Rebase
v3: protect with CONFIG_NET define.
--- drivers/gpu/drm/xe/xe_hw_error.c | 41 ++++++++++++++++++++++++++++++++ 1 file changed, 41 insertions(+)
drivers/gpu/drm/xe/xe_hw_error.c +41 -0
@@ -2,6 +2,8 @@/* * Copyright © 2023 Intel Corporation */ +#include <net/genetlink.h> +#include <uapi/drm/drm_netlink.h> #include "xe_gt_printk.h" #include "xe_hw_error.h"
@@ -776,6 +778,43 @@ xe_soc_hw_error_handler(struct xe_tile *tile, const enum hardware_error hw_err) (HARDWARE_ERROR_MAX << 1) + 1); } +#ifdef CONFIG_NET +static void +generate_netlink_event (struct xe_device *xe,const enum hardware_error hw_err) +{ +struct sk_buff *msg; +void *hdr; + +if (!xe->drm.drm_genl_family) +return ; + + msg =nlmsg_new (NLMSG_DEFAULT_SIZE ,GFP_ATOMIC ); +if (!msg) { +drm_dbg_driver (&xe->drm,"couldn't allocate memory for error multicast event\n" ); +return ; + } + + hdr =genlmsg_put (msg,0 ,0 , xe->drm.drm_genl_family,0 ,DRM_RAS_CMD_ERROR_EVENT ); +if (!hdr) { +drm_dbg_driver (&xe->drm,"mutlicast msg buffer is small\n" ); +nlmsg_free (msg); +return ; + } + +genlmsg_end (msg, hdr); + +genlmsg_multicast (xe->drm.drm_genl_family, msg,0 , + hw_err ? +DRM_GENL_MCAST_UNCORR_ERR + :DRM_GENL_MCAST_CORR_ERR , +GFP_ATOMIC ); +} +#else +static void +generate_netlink_event (struct xe_device *xe,const enum hardware_error hw_err) +{} +#endif +static void xe_hw_error_source_handler (struct xe_tile *tile,const enum hardware_error hw_err) {
@@ -837,6 +876,8 @@ xe_hw_error_source_handler(struct xe_tile *tile, const enum hardware_error hw_er }xe_mmio_write32 (&tile->mmio,DEV_ERR_STAT_REG (hw_err), errsrc); + +generate_netlink_event (tile_to_xe (tile), hw_err); unlock:spin_unlock_irqrestore (&tile_to_xe (tile)->irq.lock, flags); }
Signature · 1 line
2.25.1
Rodrigo Vivi
maintainerRe: [RFC v5 5/5] drm/xe/RAS: send multicast event on occurrence of an error
Quoted 36 lines · Aravind Iddamsetty
Whenever a correctable or an uncorrectable error happens an event is sent to the corresponding listeners of these groups.
v2: Rebase
v3: protect withCONFIG_NETdefine.Reviewed-by Michael J. Ruhl <[email protected]> #v2Signed-off-by Aravind Iddamsetty <[email protected]>--- drivers/gpu/drm/xe/xe_hw_error.c | 41 ++++++++++++++++++++++++++++++++ 1 file changed, 41 insertions(+)drivers/gpu/drm/xe/xe_hw_error.c +9
-0@@ -2,6 +2,8 @@/* * Copyright © 2023 Intel Corporation */ +#include <net/genetlink.h> +#include <uapi/drm/drm_netlink.h> #include "xe_gt_printk.h" #include "xe_hw_error.h" @@ -776,6 +778,43 @@ xe_soc_hw_error_handler(struct xe_tile *tile, const enum hardware_error hw_err) (HARDWARE_ERROR_MAX << 1) + 1); } +#ifdef CONFIG_NET +static void +generate_netlink_event (struct xe_device *xe,const enum hardware_error hw_err) +{ +struct sk_buff *msg; +void *hdr; +
drivers/gpu/drm/xe/xe_hw_error.c xe_soc_hw_error_handler(struct xe_tile *tile, const enum hardware_error hw_err) · line 797
Quoted 10 lines
drivers/gpu/drm/xe/xe_hw_error.c +10
-0+if (!xe->drm.drm_genl_family) +return ; + + msg =nlmsg_new (NLMSG_DEFAULT_SIZE ,GFP_ATOMIC ); +if (!msg) { +drm_dbg_driver (&xe->drm,"couldn't allocate memory for error multicast event\n" ); +return ; + } + + hdr =genlmsg_put (msg,0 ,0 , xe->drm.drm_genl_family,0 ,DRM_RAS_CMD_ERROR_EVENT );
this is something that could be wrapped up in the drm_ras
Quoted 35 lines
drivers/gpu/drm/xe/xe_hw_error.c +22
-0+if (!hdr) { +drm_dbg_driver (&xe->drm,"mutlicast msg buffer is small\n" ); +nlmsg_free (msg); +return ; + } + +genlmsg_end (msg, hdr); + +genlmsg_multicast (xe->drm.drm_genl_family, msg,0 , + hw_err ? +DRM_GENL_MCAST_UNCORR_ERR + :DRM_GENL_MCAST_CORR_ERR , + GFP_ATOMIC); +} +#else +static void +generate_netlink_event (struct xe_device *xe,const enum hardware_error hw_err) +{} +#endif +static void xe_hw_error_source_handler (struct xe_tile *tile,const enum hardware_error hw_err) {@@ -837,6 +876,8 @@ xe_hw_error_source_handler(struct xe_tile *tile, const enum hardware_error hw_er }xe_mmio_write32 (&tile->mmio,DEV_ERR_STAT_REG (hw_err), errsrc); + +generate_netlink_event (tile_to_xe (tile), hw_err); unlock:spin_unlock_irqrestore (&tile_to_xe (tile)->irq.lock, flags); }Signature · 1 line
2.25.1
Re: [RFC v5 5/5] drm/xe/RAS: send multicast event on occurrence of an error
Quoted 40 lines · Rodrigo Vivi
Quoted 39 lines · Aravind Iddamsetty
Whenever a correctable or an uncorrectable error happens an event is sent to the corresponding listeners of these groups.
v2: Rebase
v3: protect withCONFIG_NETdefine.Reviewed-by Michael J. Ruhl <[email protected]> #v2Signed-off-by Aravind Iddamsetty <[email protected]>--- drivers/gpu/drm/xe/xe_hw_error.c | 41 ++++++++++++++++++++++++++++++++ 1 file changed, 41 insertions(+)drivers/gpu/drm/xe/xe_hw_error.c +12
-0@@ -2,6 +2,8 @@/* * Copyright © 2023 Intel Corporation */ +#include <net/genetlink.h> +#include <uapi/drm/drm_netlink.h> #include "xe_gt_printk.h" #include "xe_hw_error.h" @@ -776,6 +778,43 @@ xe_soc_hw_error_handler(struct xe_tile *tile, const enum hardware_error hw_err) (HARDWARE_ERROR_MAX << 1) + 1); } +#ifdef CONFIG_NET +static void +generate_netlink_event (struct xe_device *xe,const enum hardware_error hw_err) +{ +struct sk_buff *msg; +void *hdr; + +if (!xe->drm.drm_genl_family) +return ; +
Quoted 8 lines
drivers/gpu/drm/xe/xe_hw_error.c xe_soc_hw_error_handler(struct xe_tile *tile, const enum hardware_error hw_err) · line 797
Quoted 7 lines
drivers/gpu/drm/xe/xe_hw_error.c +7
-0+ msg =nlmsg_new (NLMSG_DEFAULT_SIZE ,GFP_ATOMIC ); +if (!msg) { +drm_dbg_driver (&xe->drm,"couldn't allocate memory for error multicast event\n" ); +return ; + } + + hdr =genlmsg_put (msg,0 ,0 , xe->drm.drm_genl_family,0 ,DRM_RAS_CMD_ERROR_EVENT );this is something that could be wrapped up in the
drm_ras
are you referring to entire generate_netlink_event function? because I thought driver might want to pass in custom info as part of event may be like the error ID that was reported by HW.
Thanks,
Aravind.
Quoted 36 lines
Quoted 36 lines
drivers/gpu/drm/xe/xe_hw_error.c +22
-0+if (!hdr) { +drm_dbg_driver (&xe->drm,"mutlicast msg buffer is small\n" ); +nlmsg_free (msg); +return ; + } + +genlmsg_end (msg, hdr); + +genlmsg_multicast (xe->drm.drm_genl_family, msg,0 , + hw_err ? +DRM_GENL_MCAST_UNCORR_ERR + :DRM_GENL_MCAST_CORR_ERR , + GFP_ATOMIC); +} +#else +static void +generate_netlink_event (struct xe_device *xe,const enum hardware_error hw_err) +{} +#endif +static void xe_hw_error_source_handler (struct xe_tile *tile,const enum hardware_error hw_err) {@@ -837,6 +876,8 @@ xe_hw_error_source_handler(struct xe_tile *tile, const enum hardware_error hw_er }xe_mmio_write32 (&tile->mmio,DEV_ERR_STAT_REG (hw_err), errsrc); + +generate_netlink_event (tile_to_xe (tile), hw_err); unlock:spin_unlock_irqrestore (&tile_to_xe (tile)->irq.lock, flags); }Signature · 1 line
2.25.1
Quoted 11 lines · Aravind Iddamsetty
Our hardware supports
RAS(Reliability, Availability, Serviceability)by reporting the errors to the host, which the KMD processes and exposes a set of error counters which can be used by observability tools to take corrective actions or repairs. Traditionally there were being exposed via PMU (for relative counters) and sysfs interface (for absolute value) in our internal branch. But, due to the limitations in this approach to use two interfaces and also not able to have an event based reporting or configurability, an alternative approach to try netlink was suggested by community for drm subsystem wide UAPI for RAS and telemetry as discussed in [2].
Quoted 4 lines
This [2] is the inspiration to this series. It uses the generic netlink(genl) family subsystem and exposes a set of commands that can be used by every drm driver, the framework provides a means to have custom commands too.
It seems this series was originally conceived in 2023. In the meantime, tooling has been introduced to auto-generate all the netlink boilerplate code from a YAML description in Documentation/netlink/specs/. I *think* using it is mandatory for all newly introduced Netlink protocols.
Basically you create the uapi and kernel header files plus kernel source like this:
tools/net/ynl/pyynl/ynl_gen_c.py --spec Documentation/netlink/specs/drm.yaml \
--mode uapi --headertools/net/ynl/pyynl/ynl_gen_c.py --spec Documentation/netlink/specs/drm.yaml \
--mode kernel --headertools/net/ynl/pyynl/ynl_gen_c.py --spec Documentation/netlink/specs/drm.yaml \
--mode kernel --source
And then you add both the YAML file as well as the generated files to the commit. The reason you have to do that is because Python is optional for building the kernel per Documentation/process/changes.rst, so the files cannot be generated at compile time. It is possible though to regenerate them with tools/net/ynl/ynl-regen.sh whenever the YAML file is changed.
ynl_gen_c.py is capable of auto-generating code for user space applications as well. And there's tools/net/ynl/pyynl/cli.py to listen to events or send requests without having to write any code.
Thanks,
Lukas
Quoted 12 lines · Lukas Wunner
Quoted 11 lines · Aravind Iddamsetty
Our hardware supports
RAS(Reliability, Availability, Serviceability)by reporting the errors to the host, which the KMD processes and exposes a set of error counters which can be used by observability tools to take corrective actions or repairs. Traditionally there were being exposed via PMU (for relative counters) and sysfs interface (for absolute value) in our internal branch. But, due to the limitations in this approach to use two interfaces and also not able to have an event based reporting or configurability, an alternative approach to try netlink was suggested by community for drm subsystem wide UAPI for RAS and telemetry as discussed in [2].
Quoted 8 lines
Quoted 4 lines
This [2] is the inspiration to this series. It uses the generic netlink(genl) family subsystem and exposes a set of commands that can be used by every drm driver, the framework provides a means to have custom commands too.
It seems this series was originally conceived in 2023. In the meantime, tooling has been introduced to auto-generate all the netlink boilerplate code from a YAML description in
Documentation/netlink/specs/. I *think* using it is mandatory for all newly introduced Netlink protocols.
Thanks Lukas letting me know this. Will do the necessary.
Regards,
Aravind.
Quoted 25 lines
Basically you create the uapi and kernel header files plus kernel source like this:
tools/net/ynl/pyynl/ynl_gen_c.py--specDocumentation/netlink/specs/drm.yaml\
--mode uapi --headertools/net/ynl/pyynl/ynl_gen_c.py--specDocumentation/netlink/specs/drm.yaml\
--mode kernel --headertools/net/ynl/pyynl/ynl_gen_c.py--specDocumentation/netlink/specs/drm.yaml\
--mode kernel --sourceAnd then you add both the YAML file as well as the generated files to the commit. The reason you have to do that is because Python is optional for building the kernel per
Documentation/process/changes.rst, so the files cannot be generated at compile time. It is possible though to regenerate them withtools/net/ynl/ynl-regen.shwhenever the YAML file is changed.
ynl_gen_c.pyis capable of auto-generating code for user space applications as well. And there'stools/net/ynl/pyynl/cli.pyto listen to events or send requests without having to write any code.Thanks,
Lukas
Rodrigo Vivi
maintainerQuoted 38 lines · Aravind Iddamsetty
Revisiting this patch series to address pending feedback and help move the discussion towards a conclusion. This revision includes updates based on previous comments[1] and aims to clarify outstanding concerns. Specifically added command to facility reporting errors from IP blocks to support AMDGPU driver model of RAS.
[1]: https://lore.kernel.org/all/[email protected]/I sincerely appreciate everyones patience and thoughtful reviews so far, and I hope this refreshed series facilitates the final evaluation and acceptance.
Please feel free to share any further suggestions or questions.
Thank you for your continued consideration.
----------------------------------------------------------------------Our hardware supports
RAS(Reliability, Availability, Serviceability)by reporting the errors to the host, which the KMD processes and exposes a set of error counters which can be used by observability tools to take corrective actions or repairs. Traditionally there were being exposed via PMU (for relative counters) and sysfs interface (for absolute value) in our internal branch. But, due to the limitations in this approach to use two interfaces and also not able to have an event based reporting or configurability, an alternative approach to try netlink was suggested by community for drm subsystem wide UAPI for RAS and telemetry as discussed in [2].This [2] is the inspiration to this series. It uses the generic netlink(genl) family subsystem and exposes a set of commands that can be used by every drm driver, the framework provides a means to have custom commands too. Each drm driver instance in this example xe driver instance registers a family and operations to the genl subsystem through which it enumerates and reports the error counters. An event based notification is also supported to which userpace can subscribe to and be notified when any error occurs and read the error counter this avoids continuous polling on error counter. This can also be extended to threshold based notification.
I'm bringing some thoughts below and I'd like to get inputs from folks involved in the original discussions here please.
Any thought is welcomed so we can move faster towards a real GPU standard RAS solution.
Quoted 7 lines
This series is on top [3] series which introduces error counting infra in Xe driver.
[3]: https://lore.kernel.org/all/[email protected]/V5:
Add support to read error corresponding to an IP BLOCK
I honestly don't believe that this version solves all the concerns raised by AMD folks in the previous reviews. It is true that this is bringing ways of reading errors per IP block, but if I understood them correctly, they would like better (and separate) ways to declare and handle the errors coming from different IP block, rather than simply reading/querying for them filtered out.
So, I have som grouping ideas below.
Quoted 4 lines
v4:
1. Rebase
2. renamedrm_genl_sendtodrm_genl_reply
But before going to the ideas below I'd like to also raise the naming issue that I see with this proposal.
I was recently running some experiments to devlink with this and similar cases. I don't believe that devlink is a good fit for our drm-ras. It is way too much centric on network devices and any addition there to our GPU RAS would be a heavy lift. But, there are some good things from there that we could perhaps get inspiration from.
Starting from the name. devlink is the name of the tool and the name of the framework. It uses netlink on the back, but totally abstracting that. Here in this version we can see:drm_ras: the tooldrm_netlink: the abstraction
drm_genl_*: the wrapper?
So, I believe that as devlink we should have a single name for everything and avoid wrappers but providing the real module registration, with groups, and functions. Entirely abstracting the netlink and focusing on the RAS functionalities.
I'm terrible with naming, but playing a bit with AI for some suggestions, I'd say that my favorites are:
drmras - no '_' like most of the tools, but not only for the tool, but also for the files and functions.
drmlink - more link, but less ras :/
grill - GPU RAS Interface Link Layer
For the rest of the examples below I'm going with grill, but let me know your preferences.
Quoted 17 lines
3. catch error from
xa_storeand handle appropriately
4. presentlyxe_list_errorsfills blank data for IGFX, prevent it by having an early check ofIS_DGFX(Michael J. Ruhl)v3:
1. Rebase on latest RAS series for XE
2. dropDRIVER_NETLINKflag and use thedriver_genl_opsstructure to register to netlink subsystemv2: define common interfaces to genl netlink subsystem that all drm drivers can leverage.
Below is an example tool
drm_raswhich demonstrates the use of the
supported commands. The tool will be sent to ML with the subject
"[RFC i-g-t v3 0/1] A tool to demonstrate use of netlink sockets to read RAS error counters"
https://lore.kernel.org/all/[email protected]/
Quoted 4 lines
read single error counter:
$ ./drm_ras READ_ONE --device=drm:/dev/dri/card1 --error_id=0x0000000000000005 counter value 0
no need for --device, that should be mandatory argument. And we could accept BDF or card identification
$ grill list 00:03:00.0 - card0 - xe
$ grill 00:03:00.0 list # Querying available modules. monitor - global erros - gt erros - soc
Yes, my idea is that driver should be able to register modules and group per module
GRILL would be designed to accommodate multiple kinds of RAS modules, each module, with groups, categories and operations.
Modules: monitor, error, flash?!, etc?!
Groups: Global or per IP block depending on the HW underneath
Categories: Sub-groups like correctable-error vs uncorrectable-error for instance if/where
it makes sense.
Operations: Monitor: set-threshold / listen (listen is just a tool operation, but every monitor
needs to provide events over netlink)
Error: read, clear, logs
$ grill 00:03:00.0 error global counter list # List all available counters in this gpu
$ grill 00:03:00.0 error global counter show soc_fatal_hbm2_chnl0 # Show a specific counter.
$ grill 00:03:00.0 error global log # Print all the stashed CPER logs (stash can be hw/fw/sw or a combination - in AMD case it is a dump of their debugfs ring)
So, I'm sure the next question is what if the log is global, but the counters are not? Well, perhaps we should have different Modules for error-counter split from error-logging ?!
So yes, my thoughts still have some opens, but I'd like to hear your thoughts and opinions on the overall idea here.
Thanks in advance,
Rodrigo.
Quoted 325 lines
read all error counters:
$ ./drm_ras READ_ALL --device=drm:/dev/dri/card1 name config-id countererror-gt0-correctable-guc 0x0000000000000001 0 error-gt0-correctable-slm 0x0000000000000003 0 error-gt0-correctable-eu-ic 0x0000000000000004 0 error-gt0-correctable-eu-grf 0x0000000000000005 0 error-gt0-fatal-guc 0x0000000000000009 0 error-gt0-fatal-slm 0x000000000000000d 0 error-gt0-fatal-eu-grf 0x000000000000000f 0 error-gt0-fatal-fpu 0x0000000000000010 0 error-gt0-fatal-tlb 0x0000000000000011 0 error-gt0-fatal-l3-fabric 0x0000000000000012 0 error-gt0-correctable-subslice 0x0000000000000013 0 error-gt0-correctable-l3bank 0x0000000000000014 0 error-gt0-fatal-subslice 0x0000000000000015 0 error-gt0-fatal-l3bank 0x0000000000000016 0 error-gt0-sgunit-correctable 0x0000000000000017 0 error-gt0-sgunit-nonfatal 0x0000000000000018 0 error-gt0-sgunit-fatal 0x0000000000000019 0 error-gt0-soc-fatal-psf-csc-0 0x000000000000001a 0 error-gt0-soc-fatal-psf-csc-1 0x000000000000001b 0 error-gt0-soc-fatal-psf-csc-2 0x000000000000001c 0 error-gt0-soc-fatal-punit 0x000000000000001d 0 error-gt0-soc-fatal-psf-0 0x000000000000001e 0 error-gt0-soc-fatal-psf-1 0x000000000000001f 0 error-gt0-soc-fatal-psf-2 0x0000000000000020 0 error-gt0-soc-fatal-cd0 0x0000000000000021 0 error-gt0-soc-fatal-cd0-mdfi 0x0000000000000022 0 error-gt0-soc-fatal-mdfi-east 0x0000000000000023 0 error-gt0-soc-fatal-mdfi-south 0x0000000000000024 0 error-gt0-soc-fatal-hbm-ss0-0 0x0000000000000025 0 error-gt0-soc-fatal-hbm-ss0-1 0x0000000000000026 0 error-gt0-soc-fatal-hbm-ss0-2 0x0000000000000027 0 error-gt0-soc-fatal-hbm-ss0-3 0x0000000000000028 0 error-gt0-soc-fatal-hbm-ss0-4 0x0000000000000029 0 error-gt0-soc-fatal-hbm-ss0-5 0x000000000000002a 0 error-gt0-soc-fatal-hbm-ss0-6 0x000000000000002b 0 error-gt0-soc-fatal-hbm-ss0-7 0x000000000000002c 0 error-gt0-soc-fatal-hbm-ss1-0 0x000000000000002d 0 error-gt0-soc-fatal-hbm-ss1-1 0x000000000000002e 0 error-gt0-soc-fatal-hbm-ss1-2 0x000000000000002f 0 error-gt0-soc-fatal-hbm-ss1-3 0x0000000000000030 0 error-gt0-soc-fatal-hbm-ss1-4 0x0000000000000031 0 error-gt0-soc-fatal-hbm-ss1-5 0x0000000000000032 0 error-gt0-soc-fatal-hbm-ss1-6 0x0000000000000033 0 error-gt0-soc-fatal-hbm-ss1-7 0x0000000000000034 0 error-gt0-soc-fatal-hbm-ss2-0 0x0000000000000035 0 error-gt0-soc-fatal-hbm-ss2-1 0x0000000000000036 0 error-gt0-soc-fatal-hbm-ss2-2 0x0000000000000037 0 error-gt0-soc-fatal-hbm-ss2-3 0x0000000000000038 0 error-gt0-soc-fatal-hbm-ss2-4 0x0000000000000039 0 error-gt0-soc-fatal-hbm-ss2-5 0x000000000000003a 0 error-gt0-soc-fatal-hbm-ss2-6 0x000000000000003b 0 error-gt0-soc-fatal-hbm-ss2-7 0x000000000000003c 0 error-gt0-soc-fatal-hbm-ss3-0 0x000000000000003d 0 error-gt0-soc-fatal-hbm-ss3-1 0x000000000000003e 0 error-gt0-soc-fatal-hbm-ss3-2 0x000000000000003f 0 error-gt0-soc-fatal-hbm-ss3-3 0x0000000000000040 0 error-gt0-soc-fatal-hbm-ss3-4 0x0000000000000041 0 error-gt0-soc-fatal-hbm-ss3-5 0x0000000000000042 0 error-gt0-soc-fatal-hbm-ss3-6 0x0000000000000043 0 error-gt0-soc-fatal-hbm-ss3-7 0x0000000000000044 0 error-gt0-gsc-correctable-sram-ecc 0x0000000000000045 0 error-gt0-gsc-nonfatal-mia-shutdown 0x0000000000000046 0 error-gt0-gsc-nonfatal-mia-int 0x0000000000000047 0 error-gt0-gsc-nonfatal-sram-ecc 0x0000000000000048 0 error-gt0-gsc-nonfatal-wdg-timeout 0x0000000000000049 0 error-gt0-gsc-nonfatal-rom-parity 0x000000000000004a 0 error-gt0-gsc-nonfatal-ucode-parity 0x000000000000004b 0 error-gt0-gsc-nonfatal-glitch-det 0x000000000000004c 0 error-gt0-gsc-nonfatal-fuse-pull 0x000000000000004d 0 error-gt0-gsc-nonfatal-fuse-crc-check 0x000000000000004e 0 error-gt0-gsc-nonfatal-selfmbist 0x000000000000004f 0 error-gt0-gsc-nonfatal-aon-parity 0x0000000000000050 0 error-gt1-correctable-guc 0x1000000000000001 0 error-gt1-correctable-slm 0x1000000000000003 0 error-gt1-correctable-eu-ic 0x1000000000000004 0 error-gt1-correctable-eu-grf 0x1000000000000005 0 error-gt1-fatal-guc 0x1000000000000009 0 error-gt1-fatal-slm 0x100000000000000d 0 error-gt1-fatal-eu-grf 0x100000000000000f 0 error-gt1-fatal-fpu 0x1000000000000010 0 error-gt1-fatal-tlb 0x1000000000000011 0 error-gt1-fatal-l3-fabric 0x1000000000000012 0 error-gt1-correctable-subslice 0x1000000000000013 0 error-gt1-correctable-l3bank 0x1000000000000014 0 error-gt1-fatal-subslice 0x1000000000000015 0 error-gt1-fatal-l3bank 0x1000000000000016 0 error-gt1-sgunit-correctable 0x1000000000000017 0 error-gt1-sgunit-nonfatal 0x1000000000000018 0 error-gt1-sgunit-fatal 0x1000000000000019 0 error-gt1-soc-fatal-psf-csc-0 0x100000000000001a 0 error-gt1-soc-fatal-psf-csc-1 0x100000000000001b 0 error-gt1-soc-fatal-psf-csc-2 0x100000000000001c 0 error-gt1-soc-fatal-punit 0x100000000000001d 0 error-gt1-soc-fatal-psf-0 0x100000000000001e 0 error-gt1-soc-fatal-psf-1 0x100000000000001f 0 error-gt1-soc-fatal-psf-2 0x1000000000000020 0 error-gt1-soc-fatal-cd0 0x1000000000000021 0 error-gt1-soc-fatal-cd0-mdfi 0x1000000000000022 0 error-gt1-soc-fatal-mdfi-east 0x1000000000000023 0 error-gt1-soc-fatal-mdfi-south 0x1000000000000024 0 error-gt1-soc-fatal-hbm-ss0-0 0x1000000000000025 0 error-gt1-soc-fatal-hbm-ss0-1 0x1000000000000026 0 error-gt1-soc-fatal-hbm-ss0-2 0x1000000000000027 0 error-gt1-soc-fatal-hbm-ss0-3 0x1000000000000028 0 error-gt1-soc-fatal-hbm-ss0-4 0x1000000000000029 0 error-gt1-soc-fatal-hbm-ss0-5 0x100000000000002a 0 error-gt1-soc-fatal-hbm-ss0-6 0x100000000000002b 0 error-gt1-soc-fatal-hbm-ss0-7 0x100000000000002c 0 error-gt1-soc-fatal-hbm-ss1-0 0x100000000000002d 0 error-gt1-soc-fatal-hbm-ss1-1 0x100000000000002e 0 error-gt1-soc-fatal-hbm-ss1-2 0x100000000000002f 0 error-gt1-soc-fatal-hbm-ss1-3 0x1000000000000030 0 error-gt1-soc-fatal-hbm-ss1-4 0x1000000000000031 0 error-gt1-soc-fatal-hbm-ss1-5 0x1000000000000032 0 error-gt1-soc-fatal-hbm-ss1-6 0x1000000000000033 0 error-gt1-soc-fatal-hbm-ss1-7 0x1000000000000034 0 error-gt1-soc-fatal-hbm-ss2-0 0x1000000000000035 0 error-gt1-soc-fatal-hbm-ss2-1 0x1000000000000036 0 error-gt1-soc-fatal-hbm-ss2-2 0x1000000000000037 0 error-gt1-soc-fatal-hbm-ss2-3 0x1000000000000038 0 error-gt1-soc-fatal-hbm-ss2-4 0x1000000000000039 0 error-gt1-soc-fatal-hbm-ss2-5 0x100000000000003a 0 error-gt1-soc-fatal-hbm-ss2-6 0x100000000000003b 0 error-gt1-soc-fatal-hbm-ss2-7 0x100000000000003c 0 error-gt1-soc-fatal-hbm-ss3-0 0x100000000000003d 0 error-gt1-soc-fatal-hbm-ss3-1 0x100000000000003e 0 error-gt1-soc-fatal-hbm-ss3-2 0x100000000000003f 0 error-gt1-soc-fatal-hbm-ss3-3 0x1000000000000040 0 error-gt1-soc-fatal-hbm-ss3-4 0x1000000000000041 0 error-gt1-soc-fatal-hbm-ss3-5 0x1000000000000042 0 error-gt1-soc-fatal-hbm-ss3-6 0x1000000000000043 0 error-gt1-soc-fatal-hbm-ss3-7 0x1000000000000044 0wait on a error event:
$ ./drm_ras WAIT_ON_EVENT --device=drm:/dev/dri/card1 waiting for error event error event received counter value 0list all errors:
$ ./drm_ras LIST_ERRORS --device=drm:/dev/dri/card1 name config-iderror-gt0-correctable-guc 0x0000000000000001 error-gt0-correctable-slm 0x0000000000000003 error-gt0-correctable-eu-ic 0x0000000000000004 error-gt0-correctable-eu-grf 0x0000000000000005 error-gt0-fatal-guc 0x0000000000000009 error-gt0-fatal-slm 0x000000000000000d error-gt0-fatal-eu-grf 0x000000000000000f error-gt0-fatal-fpu 0x0000000000000010 error-gt0-fatal-tlb 0x0000000000000011 error-gt0-fatal-l3-fabric 0x0000000000000012 error-gt0-correctable-subslice 0x0000000000000013 error-gt0-correctable-l3bank 0x0000000000000014 error-gt0-fatal-subslice 0x0000000000000015 error-gt0-fatal-l3bank 0x0000000000000016 error-gt0-sgunit-correctable 0x0000000000000017 error-gt0-sgunit-nonfatal 0x0000000000000018 error-gt0-sgunit-fatal 0x0000000000000019 error-gt0-soc-fatal-psf-csc-0 0x000000000000001a error-gt0-soc-fatal-psf-csc-1 0x000000000000001b error-gt0-soc-fatal-psf-csc-2 0x000000000000001c error-gt0-soc-fatal-punit 0x000000000000001d error-gt0-soc-fatal-psf-0 0x000000000000001e error-gt0-soc-fatal-psf-1 0x000000000000001f error-gt0-soc-fatal-psf-2 0x0000000000000020 error-gt0-soc-fatal-cd0 0x0000000000000021 error-gt0-soc-fatal-cd0-mdfi 0x0000000000000022 error-gt0-soc-fatal-mdfi-east 0x0000000000000023 error-gt0-soc-fatal-mdfi-south 0x0000000000000024 error-gt0-soc-fatal-hbm-ss0-0 0x0000000000000025 error-gt0-soc-fatal-hbm-ss0-1 0x0000000000000026 error-gt0-soc-fatal-hbm-ss0-2 0x0000000000000027 error-gt0-soc-fatal-hbm-ss0-3 0x0000000000000028 error-gt0-soc-fatal-hbm-ss0-4 0x0000000000000029 error-gt0-soc-fatal-hbm-ss0-5 0x000000000000002a error-gt0-soc-fatal-hbm-ss0-6 0x000000000000002b error-gt0-soc-fatal-hbm-ss0-7 0x000000000000002c error-gt0-soc-fatal-hbm-ss1-0 0x000000000000002d error-gt0-soc-fatal-hbm-ss1-1 0x000000000000002e error-gt0-soc-fatal-hbm-ss1-2 0x000000000000002f error-gt0-soc-fatal-hbm-ss1-3 0x0000000000000030 error-gt0-soc-fatal-hbm-ss1-4 0x0000000000000031 error-gt0-soc-fatal-hbm-ss1-5 0x0000000000000032 error-gt0-soc-fatal-hbm-ss1-6 0x0000000000000033 error-gt0-soc-fatal-hbm-ss1-7 0x0000000000000034 error-gt0-soc-fatal-hbm-ss2-0 0x0000000000000035 error-gt0-soc-fatal-hbm-ss2-1 0x0000000000000036 error-gt0-soc-fatal-hbm-ss2-2 0x0000000000000037 error-gt0-soc-fatal-hbm-ss2-3 0x0000000000000038 error-gt0-soc-fatal-hbm-ss2-4 0x0000000000000039 error-gt0-soc-fatal-hbm-ss2-5 0x000000000000003a error-gt0-soc-fatal-hbm-ss2-6 0x000000000000003b error-gt0-soc-fatal-hbm-ss2-7 0x000000000000003c error-gt0-soc-fatal-hbm-ss3-0 0x000000000000003d error-gt0-soc-fatal-hbm-ss3-1 0x000000000000003e error-gt0-soc-fatal-hbm-ss3-2 0x000000000000003f error-gt0-soc-fatal-hbm-ss3-3 0x0000000000000040 error-gt0-soc-fatal-hbm-ss3-4 0x0000000000000041 error-gt0-soc-fatal-hbm-ss3-5 0x0000000000000042 error-gt0-soc-fatal-hbm-ss3-6 0x0000000000000043 error-gt0-soc-fatal-hbm-ss3-7 0x0000000000000044 error-gt0-gsc-correctable-sram-ecc 0x0000000000000045 error-gt0-gsc-nonfatal-mia-shutdown 0x0000000000000046 error-gt0-gsc-nonfatal-mia-int 0x0000000000000047 error-gt0-gsc-nonfatal-sram-ecc 0x0000000000000048 error-gt0-gsc-nonfatal-wdg-timeout 0x0000000000000049 error-gt0-gsc-nonfatal-rom-parity 0x000000000000004a error-gt0-gsc-nonfatal-ucode-parity 0x000000000000004b error-gt0-gsc-nonfatal-glitch-det 0x000000000000004c error-gt0-gsc-nonfatal-fuse-pull 0x000000000000004d error-gt0-gsc-nonfatal-fuse-crc-check 0x000000000000004e error-gt0-gsc-nonfatal-selfmbist 0x000000000000004f error-gt0-gsc-nonfatal-aon-parity 0x0000000000000050 error-gt1-correctable-guc 0x1000000000000001 error-gt1-correctable-slm 0x1000000000000003 error-gt1-correctable-eu-ic 0x1000000000000004 error-gt1-correctable-eu-grf 0x1000000000000005 error-gt1-fatal-guc 0x1000000000000009 error-gt1-fatal-slm 0x100000000000000d error-gt1-fatal-eu-grf 0x100000000000000f error-gt1-fatal-fpu 0x1000000000000010 error-gt1-fatal-tlb 0x1000000000000011 error-gt1-fatal-l3-fabric 0x1000000000000012 error-gt1-correctable-subslice 0x1000000000000013 error-gt1-correctable-l3bank 0x1000000000000014 error-gt1-fatal-subslice 0x1000000000000015 error-gt1-fatal-l3bank 0x1000000000000016 error-gt1-sgunit-correctable 0x1000000000000017 error-gt1-sgunit-nonfatal 0x1000000000000018 error-gt1-sgunit-fatal 0x1000000000000019 error-gt1-soc-fatal-psf-csc-0 0x100000000000001a error-gt1-soc-fatal-psf-csc-1 0x100000000000001b error-gt1-soc-fatal-psf-csc-2 0x100000000000001c error-gt1-soc-fatal-punit 0x100000000000001d error-gt1-soc-fatal-psf-0 0x100000000000001e error-gt1-soc-fatal-psf-1 0x100000000000001f error-gt1-soc-fatal-psf-2 0x1000000000000020 error-gt1-soc-fatal-cd0 0x1000000000000021 error-gt1-soc-fatal-cd0-mdfi 0x1000000000000022 error-gt1-soc-fatal-mdfi-east 0x1000000000000023 error-gt1-soc-fatal-mdfi-south 0x1000000000000024 error-gt1-soc-fatal-hbm-ss0-0 0x1000000000000025 error-gt1-soc-fatal-hbm-ss0-1 0x1000000000000026 error-gt1-soc-fatal-hbm-ss0-2 0x1000000000000027 error-gt1-soc-fatal-hbm-ss0-3 0x1000000000000028 error-gt1-soc-fatal-hbm-ss0-4 0x1000000000000029 error-gt1-soc-fatal-hbm-ss0-5 0x100000000000002a error-gt1-soc-fatal-hbm-ss0-6 0x100000000000002b error-gt1-soc-fatal-hbm-ss0-7 0x100000000000002c error-gt1-soc-fatal-hbm-ss1-0 0x100000000000002d error-gt1-soc-fatal-hbm-ss1-1 0x100000000000002e error-gt1-soc-fatal-hbm-ss1-2 0x100000000000002f error-gt1-soc-fatal-hbm-ss1-3 0x1000000000000030 error-gt1-soc-fatal-hbm-ss1-4 0x1000000000000031 error-gt1-soc-fatal-hbm-ss1-5 0x1000000000000032 error-gt1-soc-fatal-hbm-ss1-6 0x1000000000000033 error-gt1-soc-fatal-hbm-ss1-7 0x1000000000000034 error-gt1-soc-fatal-hbm-ss2-0 0x1000000000000035 error-gt1-soc-fatal-hbm-ss2-1 0x1000000000000036 error-gt1-soc-fatal-hbm-ss2-2 0x1000000000000037 error-gt1-soc-fatal-hbm-ss2-3 0x1000000000000038 error-gt1-soc-fatal-hbm-ss2-4 0x1000000000000039 error-gt1-soc-fatal-hbm-ss2-5 0x100000000000003a error-gt1-soc-fatal-hbm-ss2-6 0x100000000000003b error-gt1-soc-fatal-hbm-ss2-7 0x100000000000003c error-gt1-soc-fatal-hbm-ss3-0 0x100000000000003d error-gt1-soc-fatal-hbm-ss3-1 0x100000000000003e error-gt1-soc-fatal-hbm-ss3-2 0x100000000000003f error-gt1-soc-fatal-hbm-ss3-3 0x1000000000000040 error-gt1-soc-fatal-hbm-ss3-4 0x1000000000000041 error-gt1-soc-fatal-hbm-ss3-5 0x1000000000000042 error-gt1-soc-fatal-hbm-ss3-6 0x1000000000000043 error-gt1-soc-fatal-hbm-ss3-7 0x1000000000000044Cc Alex Deucher <[email protected]>Cc David Airlie <[email protected]>Cc Simona Vetter <[email protected]>Cc Joonas Lahtinen <[email protected]>Cc Rodrigo Vivi <[email protected]>Cc Hawking Zhang <[email protected]>Cc Lijo Lazar <[email protected]>Cc Ruhl, Michael J <[email protected]>Cc Riana Tauro <[email protected]>Cc Anshuman Gupta <[email protected]>Aravind Iddamsetty (5):
drm/netlink: Add netlink infrastructure drm/xe/RAS: Register netlink capability drm/xe/RAS: Expose the error counters drm/netlink: Define multicast groups
drm/xe/RAS: send multicast event on occurrence of an errordrivers/gpu/drm/Makefile | 1 + drivers/gpu/drm/drm_drv.c | 7 + drivers/gpu/drm/drm_netlink.c | 219 +++++++++++ drivers/gpu/drm/xe/Makefile | 2 + drivers/gpu/drm/xe/xe_device.c | 6 + drivers/gpu/drm/xe/xe_device_types.h | 1 + drivers/gpu/drm/xe/xe_hw_error.c | 56 ++- drivers/gpu/drm/xe/xe_netlink.c | 531 +++++++++++++++++++++++++++ include/drm/drm_device.h | 10 + include/drm/drm_drv.h | 7 + include/drm/drm_netlink.h | 46 +++ include/uapi/drm/drm_netlink.h | 105 ++++++ include/uapi/drm/xe_drm.h | 85 +++++ 13 files changed, 1071 insertions(+), 5 deletions(-) create mode 100644 drivers/gpu/drm/drm_netlink.c create mode 100644 drivers/gpu/drm/xe/xe_netlink.c create mode 100644 include/drm/drm_netlink.h create mode 100644 include/uapi/drm/drm_netlink.hSignature · 1 line
2.25.1
Rodrigo Vivi
maintainerQuoted 126 lines · Rodrigo Vivi
Quoted 38 lines · Aravind Iddamsetty
Revisiting this patch series to address pending feedback and help move the discussion towards a conclusion. This revision includes updates based on previous comments[1] and aims to clarify outstanding concerns. Specifically added command to facility reporting errors from IP blocks to support AMDGPU driver model of RAS.
[1]: https://lore.kernel.org/all/[email protected]/I sincerely appreciate everyones patience and thoughtful reviews so far, and I hope this refreshed series facilitates the final evaluation and acceptance.
Please feel free to share any further suggestions or questions.
Thank you for your continued consideration.
----------------------------------------------------------------------Our hardware supports
RAS(Reliability, Availability, Serviceability)by reporting the errors to the host, which the KMD processes and exposes a set of error counters which can be used by observability tools to take corrective actions or repairs. Traditionally there were being exposed via PMU (for relative counters) and sysfs interface (for absolute value) in our internal branch. But, due to the limitations in this approach to use two interfaces and also not able to have an event based reporting or configurability, an alternative approach to try netlink was suggested by community for drm subsystem wide UAPI for RAS and telemetry as discussed in [2].This [2] is the inspiration to this series. It uses the generic netlink(genl) family subsystem and exposes a set of commands that can be used by every drm driver, the framework provides a means to have custom commands too. Each drm driver instance in this example xe driver instance registers a family and operations to the genl subsystem through which it enumerates and reports the error counters. An event based notification is also supported to which userpace can subscribe to and be notified when any error occurs and read the error counter this avoids continuous polling on error counter. This can also be extended to threshold based notification.
I'm bringing some thoughts below and I'd like to get inputs from folks involved in the original discussions here please.
Any thought is welcomed so we can move faster towards a real GPU standard RAS solution.Quoted 7 lines
This series is on top [3] series which introduces error counting infra in Xe driver.
[3]: https://lore.kernel.org/all/[email protected]/V5:
Add support to read error corresponding to an IP BLOCKI honestly don't believe that this version solves all the concerns raised by AMD folks in the previous reviews. It is true that this is bringing ways of reading errors per IP block, but if I understood them correctly, they would like better (and separate) ways to declare and handle the errors coming from different IP block, rather than simply reading/querying for them filtered out.
So, I have som grouping ideas below.
Quoted 4 lines
v4:
1. Rebase
2. renamedrm_genl_sendtodrm_genl_replyBut before going to the ideas below I'd like to also raise the naming issue that I see with this proposal.
I was recently running some experiments to devlink with this and similar cases. I don't believe that devlink is a good fit for our drm-ras. It is way too much centric on network devices and any addition there to our GPU RAS would be a heavy lift. But, there are some good things from there that we could perhaps get inspiration from.
Starting from the name. devlink is the name of the tool and the name of the framework. It uses netlink on the back, but totally abstracting that. Here in this version we can see:
drm_ras: the tooldrm_netlink: the abstraction
drm_genl_*: the wrapper?So, I believe that as devlink we should have a single name for everything and avoid wrappers but providing the real module registration, with groups, and functions. Entirely abstracting the netlink and focusing on the RAS functionalities.
I'm terrible with naming, but playing a bit with AI for some suggestions, I'd say that my favorites are:
drmras - no '_' like most of the tools, but not only for the tool, but also for the files and functions.
drmlink - more link, but less ras :/
grill - GPU RAS Interface Link LayerFor the rest of the examples below I'm going with grill, but let me know your preferences.
Quoted 17 lines
3. catch error from
xa_storeand handle appropriately
4. presentlyxe_list_errorsfills blank data for IGFX, prevent it by having an early check ofIS_DGFX(Michael J. Ruhl)v3:
1. Rebase on latest RAS series for XE
2. dropDRIVER_NETLINKflag and use thedriver_genl_opsstructure to register to netlink subsystemv2: define common interfaces to genl netlink subsystem that all drm drivers can leverage.
Below is an example tool
drm_raswhich demonstrates the use of the
supported commands. The tool will be sent to ML with the subject
"[RFC i-g-t v3 0/1] A tool to demonstrate use of netlink sockets to read RAS error counters"
https://lore.kernel.org/all/[email protected]/Quoted 4 lines
read single error counter:
$ ./drm_ras READ_ONE --device=drm:/dev/dri/card1 --error_id=0x0000000000000005 counter value 0no need for --device, that should be mandatory argument. And we could accept BDF or card identification
$ grill list 00:03:00.0 - card0 - xe
Quoted 6 lines
$ grill 00:03:00.0 list # Querying available modules. monitor - global erros - gt erros - socYes, my idea is that driver should be able to register modules and group per module
Please allow me to emphasize here that the group registration is just to make this extensible and also to accommodate the AMD case, but not change the original essence of the goal which is to create the drm-ras solution.
Quoted 3 lines
GRILL would be designed to accommodate multiple kinds of RAS modules, each module, with groups, categories and operations.
also let me take back on the naming here.
Please let's go with the obvious drm_ras.
Perhaps drm_ras for the code here and drmras (one-word) for the IGT tool.
Quoted 2 lines
Modules: monitor, error, flash?!, etc?!
Here, please let me also tune it down a bit. The overall goal continue to be the creation of our drm-ras framework using netlink to report error counters and events.
Any addition is a good-to-have, but shouldn't delay the main goal. Also, any addition should be carefully reviewed individually in the future. My only wish here at this moment is to think from the very beginning in something that is expansible.
Thanks,
Rodrigo.
Quoted 354 lines
Groups: Global or per IP block depending on the HW underneath
Categories: Sub-groups like correctable-error vs uncorrectable-error for instance if/where
it makes sense.
Operations: Monitor: set-threshold / listen (listen is just a tool operation, but every monitor
needs to provide events over netlink)
Error: read, clear, logs$ grill 00:03:00.0 error global counter list # List all available counters in this gpu$ grill 00:03:00.0 error global counter show soc_fatal_hbm2_chnl0 # Show a specific counter.$ grill 00:03:00.0 error global log # Print all the stashed CPER logs (stash can be hw/fw/sw or a combination - in AMD case it is a dump of their debugfs ring)So, I'm sure the next question is what if the log is global, but the counters are not? Well, perhaps we should have different Modules for error-counter split from error-logging ?!
So yes, my thoughts still have some opens, but I'd like to hear your thoughts and opinions on the overall idea here.
Thanks in advance,
Rodrigo.Quoted 325 lines
read all error counters:
$ ./drm_ras READ_ALL --device=drm:/dev/dri/card1 name config-id countererror-gt0-correctable-guc 0x0000000000000001 0 error-gt0-correctable-slm 0x0000000000000003 0 error-gt0-correctable-eu-ic 0x0000000000000004 0 error-gt0-correctable-eu-grf 0x0000000000000005 0 error-gt0-fatal-guc 0x0000000000000009 0 error-gt0-fatal-slm 0x000000000000000d 0 error-gt0-fatal-eu-grf 0x000000000000000f 0 error-gt0-fatal-fpu 0x0000000000000010 0 error-gt0-fatal-tlb 0x0000000000000011 0 error-gt0-fatal-l3-fabric 0x0000000000000012 0 error-gt0-correctable-subslice 0x0000000000000013 0 error-gt0-correctable-l3bank 0x0000000000000014 0 error-gt0-fatal-subslice 0x0000000000000015 0 error-gt0-fatal-l3bank 0x0000000000000016 0 error-gt0-sgunit-correctable 0x0000000000000017 0 error-gt0-sgunit-nonfatal 0x0000000000000018 0 error-gt0-sgunit-fatal 0x0000000000000019 0 error-gt0-soc-fatal-psf-csc-0 0x000000000000001a 0 error-gt0-soc-fatal-psf-csc-1 0x000000000000001b 0 error-gt0-soc-fatal-psf-csc-2 0x000000000000001c 0 error-gt0-soc-fatal-punit 0x000000000000001d 0 error-gt0-soc-fatal-psf-0 0x000000000000001e 0 error-gt0-soc-fatal-psf-1 0x000000000000001f 0 error-gt0-soc-fatal-psf-2 0x0000000000000020 0 error-gt0-soc-fatal-cd0 0x0000000000000021 0 error-gt0-soc-fatal-cd0-mdfi 0x0000000000000022 0 error-gt0-soc-fatal-mdfi-east 0x0000000000000023 0 error-gt0-soc-fatal-mdfi-south 0x0000000000000024 0 error-gt0-soc-fatal-hbm-ss0-0 0x0000000000000025 0 error-gt0-soc-fatal-hbm-ss0-1 0x0000000000000026 0 error-gt0-soc-fatal-hbm-ss0-2 0x0000000000000027 0 error-gt0-soc-fatal-hbm-ss0-3 0x0000000000000028 0 error-gt0-soc-fatal-hbm-ss0-4 0x0000000000000029 0 error-gt0-soc-fatal-hbm-ss0-5 0x000000000000002a 0 error-gt0-soc-fatal-hbm-ss0-6 0x000000000000002b 0 error-gt0-soc-fatal-hbm-ss0-7 0x000000000000002c 0 error-gt0-soc-fatal-hbm-ss1-0 0x000000000000002d 0 error-gt0-soc-fatal-hbm-ss1-1 0x000000000000002e 0 error-gt0-soc-fatal-hbm-ss1-2 0x000000000000002f 0 error-gt0-soc-fatal-hbm-ss1-3 0x0000000000000030 0 error-gt0-soc-fatal-hbm-ss1-4 0x0000000000000031 0 error-gt0-soc-fatal-hbm-ss1-5 0x0000000000000032 0 error-gt0-soc-fatal-hbm-ss1-6 0x0000000000000033 0 error-gt0-soc-fatal-hbm-ss1-7 0x0000000000000034 0 error-gt0-soc-fatal-hbm-ss2-0 0x0000000000000035 0 error-gt0-soc-fatal-hbm-ss2-1 0x0000000000000036 0 error-gt0-soc-fatal-hbm-ss2-2 0x0000000000000037 0 error-gt0-soc-fatal-hbm-ss2-3 0x0000000000000038 0 error-gt0-soc-fatal-hbm-ss2-4 0x0000000000000039 0 error-gt0-soc-fatal-hbm-ss2-5 0x000000000000003a 0 error-gt0-soc-fatal-hbm-ss2-6 0x000000000000003b 0 error-gt0-soc-fatal-hbm-ss2-7 0x000000000000003c 0 error-gt0-soc-fatal-hbm-ss3-0 0x000000000000003d 0 error-gt0-soc-fatal-hbm-ss3-1 0x000000000000003e 0 error-gt0-soc-fatal-hbm-ss3-2 0x000000000000003f 0 error-gt0-soc-fatal-hbm-ss3-3 0x0000000000000040 0 error-gt0-soc-fatal-hbm-ss3-4 0x0000000000000041 0 error-gt0-soc-fatal-hbm-ss3-5 0x0000000000000042 0 error-gt0-soc-fatal-hbm-ss3-6 0x0000000000000043 0 error-gt0-soc-fatal-hbm-ss3-7 0x0000000000000044 0 error-gt0-gsc-correctable-sram-ecc 0x0000000000000045 0 error-gt0-gsc-nonfatal-mia-shutdown 0x0000000000000046 0 error-gt0-gsc-nonfatal-mia-int 0x0000000000000047 0 error-gt0-gsc-nonfatal-sram-ecc 0x0000000000000048 0 error-gt0-gsc-nonfatal-wdg-timeout 0x0000000000000049 0 error-gt0-gsc-nonfatal-rom-parity 0x000000000000004a 0 error-gt0-gsc-nonfatal-ucode-parity 0x000000000000004b 0 error-gt0-gsc-nonfatal-glitch-det 0x000000000000004c 0 error-gt0-gsc-nonfatal-fuse-pull 0x000000000000004d 0 error-gt0-gsc-nonfatal-fuse-crc-check 0x000000000000004e 0 error-gt0-gsc-nonfatal-selfmbist 0x000000000000004f 0 error-gt0-gsc-nonfatal-aon-parity 0x0000000000000050 0 error-gt1-correctable-guc 0x1000000000000001 0 error-gt1-correctable-slm 0x1000000000000003 0 error-gt1-correctable-eu-ic 0x1000000000000004 0 error-gt1-correctable-eu-grf 0x1000000000000005 0 error-gt1-fatal-guc 0x1000000000000009 0 error-gt1-fatal-slm 0x100000000000000d 0 error-gt1-fatal-eu-grf 0x100000000000000f 0 error-gt1-fatal-fpu 0x1000000000000010 0 error-gt1-fatal-tlb 0x1000000000000011 0 error-gt1-fatal-l3-fabric 0x1000000000000012 0 error-gt1-correctable-subslice 0x1000000000000013 0 error-gt1-correctable-l3bank 0x1000000000000014 0 error-gt1-fatal-subslice 0x1000000000000015 0 error-gt1-fatal-l3bank 0x1000000000000016 0 error-gt1-sgunit-correctable 0x1000000000000017 0 error-gt1-sgunit-nonfatal 0x1000000000000018 0 error-gt1-sgunit-fatal 0x1000000000000019 0 error-gt1-soc-fatal-psf-csc-0 0x100000000000001a 0 error-gt1-soc-fatal-psf-csc-1 0x100000000000001b 0 error-gt1-soc-fatal-psf-csc-2 0x100000000000001c 0 error-gt1-soc-fatal-punit 0x100000000000001d 0 error-gt1-soc-fatal-psf-0 0x100000000000001e 0 error-gt1-soc-fatal-psf-1 0x100000000000001f 0 error-gt1-soc-fatal-psf-2 0x1000000000000020 0 error-gt1-soc-fatal-cd0 0x1000000000000021 0 error-gt1-soc-fatal-cd0-mdfi 0x1000000000000022 0 error-gt1-soc-fatal-mdfi-east 0x1000000000000023 0 error-gt1-soc-fatal-mdfi-south 0x1000000000000024 0 error-gt1-soc-fatal-hbm-ss0-0 0x1000000000000025 0 error-gt1-soc-fatal-hbm-ss0-1 0x1000000000000026 0 error-gt1-soc-fatal-hbm-ss0-2 0x1000000000000027 0 error-gt1-soc-fatal-hbm-ss0-3 0x1000000000000028 0 error-gt1-soc-fatal-hbm-ss0-4 0x1000000000000029 0 error-gt1-soc-fatal-hbm-ss0-5 0x100000000000002a 0 error-gt1-soc-fatal-hbm-ss0-6 0x100000000000002b 0 error-gt1-soc-fatal-hbm-ss0-7 0x100000000000002c 0 error-gt1-soc-fatal-hbm-ss1-0 0x100000000000002d 0 error-gt1-soc-fatal-hbm-ss1-1 0x100000000000002e 0 error-gt1-soc-fatal-hbm-ss1-2 0x100000000000002f 0 error-gt1-soc-fatal-hbm-ss1-3 0x1000000000000030 0 error-gt1-soc-fatal-hbm-ss1-4 0x1000000000000031 0 error-gt1-soc-fatal-hbm-ss1-5 0x1000000000000032 0 error-gt1-soc-fatal-hbm-ss1-6 0x1000000000000033 0 error-gt1-soc-fatal-hbm-ss1-7 0x1000000000000034 0 error-gt1-soc-fatal-hbm-ss2-0 0x1000000000000035 0 error-gt1-soc-fatal-hbm-ss2-1 0x1000000000000036 0 error-gt1-soc-fatal-hbm-ss2-2 0x1000000000000037 0 error-gt1-soc-fatal-hbm-ss2-3 0x1000000000000038 0 error-gt1-soc-fatal-hbm-ss2-4 0x1000000000000039 0 error-gt1-soc-fatal-hbm-ss2-5 0x100000000000003a 0 error-gt1-soc-fatal-hbm-ss2-6 0x100000000000003b 0 error-gt1-soc-fatal-hbm-ss2-7 0x100000000000003c 0 error-gt1-soc-fatal-hbm-ss3-0 0x100000000000003d 0 error-gt1-soc-fatal-hbm-ss3-1 0x100000000000003e 0 error-gt1-soc-fatal-hbm-ss3-2 0x100000000000003f 0 error-gt1-soc-fatal-hbm-ss3-3 0x1000000000000040 0 error-gt1-soc-fatal-hbm-ss3-4 0x1000000000000041 0 error-gt1-soc-fatal-hbm-ss3-5 0x1000000000000042 0 error-gt1-soc-fatal-hbm-ss3-6 0x1000000000000043 0 error-gt1-soc-fatal-hbm-ss3-7 0x1000000000000044 0wait on a error event:
$ ./drm_ras WAIT_ON_EVENT --device=drm:/dev/dri/card1 waiting for error event error event received counter value 0list all errors:
$ ./drm_ras LIST_ERRORS --device=drm:/dev/dri/card1 name config-iderror-gt0-correctable-guc 0x0000000000000001 error-gt0-correctable-slm 0x0000000000000003 error-gt0-correctable-eu-ic 0x0000000000000004 error-gt0-correctable-eu-grf 0x0000000000000005 error-gt0-fatal-guc 0x0000000000000009 error-gt0-fatal-slm 0x000000000000000d error-gt0-fatal-eu-grf 0x000000000000000f error-gt0-fatal-fpu 0x0000000000000010 error-gt0-fatal-tlb 0x0000000000000011 error-gt0-fatal-l3-fabric 0x0000000000000012 error-gt0-correctable-subslice 0x0000000000000013 error-gt0-correctable-l3bank 0x0000000000000014 error-gt0-fatal-subslice 0x0000000000000015 error-gt0-fatal-l3bank 0x0000000000000016 error-gt0-sgunit-correctable 0x0000000000000017 error-gt0-sgunit-nonfatal 0x0000000000000018 error-gt0-sgunit-fatal 0x0000000000000019 error-gt0-soc-fatal-psf-csc-0 0x000000000000001a error-gt0-soc-fatal-psf-csc-1 0x000000000000001b error-gt0-soc-fatal-psf-csc-2 0x000000000000001c error-gt0-soc-fatal-punit 0x000000000000001d error-gt0-soc-fatal-psf-0 0x000000000000001e error-gt0-soc-fatal-psf-1 0x000000000000001f error-gt0-soc-fatal-psf-2 0x0000000000000020 error-gt0-soc-fatal-cd0 0x0000000000000021 error-gt0-soc-fatal-cd0-mdfi 0x0000000000000022 error-gt0-soc-fatal-mdfi-east 0x0000000000000023 error-gt0-soc-fatal-mdfi-south 0x0000000000000024 error-gt0-soc-fatal-hbm-ss0-0 0x0000000000000025 error-gt0-soc-fatal-hbm-ss0-1 0x0000000000000026 error-gt0-soc-fatal-hbm-ss0-2 0x0000000000000027 error-gt0-soc-fatal-hbm-ss0-3 0x0000000000000028 error-gt0-soc-fatal-hbm-ss0-4 0x0000000000000029 error-gt0-soc-fatal-hbm-ss0-5 0x000000000000002a error-gt0-soc-fatal-hbm-ss0-6 0x000000000000002b error-gt0-soc-fatal-hbm-ss0-7 0x000000000000002c error-gt0-soc-fatal-hbm-ss1-0 0x000000000000002d error-gt0-soc-fatal-hbm-ss1-1 0x000000000000002e error-gt0-soc-fatal-hbm-ss1-2 0x000000000000002f error-gt0-soc-fatal-hbm-ss1-3 0x0000000000000030 error-gt0-soc-fatal-hbm-ss1-4 0x0000000000000031 error-gt0-soc-fatal-hbm-ss1-5 0x0000000000000032 error-gt0-soc-fatal-hbm-ss1-6 0x0000000000000033 error-gt0-soc-fatal-hbm-ss1-7 0x0000000000000034 error-gt0-soc-fatal-hbm-ss2-0 0x0000000000000035 error-gt0-soc-fatal-hbm-ss2-1 0x0000000000000036 error-gt0-soc-fatal-hbm-ss2-2 0x0000000000000037 error-gt0-soc-fatal-hbm-ss2-3 0x0000000000000038 error-gt0-soc-fatal-hbm-ss2-4 0x0000000000000039 error-gt0-soc-fatal-hbm-ss2-5 0x000000000000003a error-gt0-soc-fatal-hbm-ss2-6 0x000000000000003b error-gt0-soc-fatal-hbm-ss2-7 0x000000000000003c error-gt0-soc-fatal-hbm-ss3-0 0x000000000000003d error-gt0-soc-fatal-hbm-ss3-1 0x000000000000003e error-gt0-soc-fatal-hbm-ss3-2 0x000000000000003f error-gt0-soc-fatal-hbm-ss3-3 0x0000000000000040 error-gt0-soc-fatal-hbm-ss3-4 0x0000000000000041 error-gt0-soc-fatal-hbm-ss3-5 0x0000000000000042 error-gt0-soc-fatal-hbm-ss3-6 0x0000000000000043 error-gt0-soc-fatal-hbm-ss3-7 0x0000000000000044 error-gt0-gsc-correctable-sram-ecc 0x0000000000000045 error-gt0-gsc-nonfatal-mia-shutdown 0x0000000000000046 error-gt0-gsc-nonfatal-mia-int 0x0000000000000047 error-gt0-gsc-nonfatal-sram-ecc 0x0000000000000048 error-gt0-gsc-nonfatal-wdg-timeout 0x0000000000000049 error-gt0-gsc-nonfatal-rom-parity 0x000000000000004a error-gt0-gsc-nonfatal-ucode-parity 0x000000000000004b error-gt0-gsc-nonfatal-glitch-det 0x000000000000004c error-gt0-gsc-nonfatal-fuse-pull 0x000000000000004d error-gt0-gsc-nonfatal-fuse-crc-check 0x000000000000004e error-gt0-gsc-nonfatal-selfmbist 0x000000000000004f error-gt0-gsc-nonfatal-aon-parity 0x0000000000000050 error-gt1-correctable-guc 0x1000000000000001 error-gt1-correctable-slm 0x1000000000000003 error-gt1-correctable-eu-ic 0x1000000000000004 error-gt1-correctable-eu-grf 0x1000000000000005 error-gt1-fatal-guc 0x1000000000000009 error-gt1-fatal-slm 0x100000000000000d error-gt1-fatal-eu-grf 0x100000000000000f error-gt1-fatal-fpu 0x1000000000000010 error-gt1-fatal-tlb 0x1000000000000011 error-gt1-fatal-l3-fabric 0x1000000000000012 error-gt1-correctable-subslice 0x1000000000000013 error-gt1-correctable-l3bank 0x1000000000000014 error-gt1-fatal-subslice 0x1000000000000015 error-gt1-fatal-l3bank 0x1000000000000016 error-gt1-sgunit-correctable 0x1000000000000017 error-gt1-sgunit-nonfatal 0x1000000000000018 error-gt1-sgunit-fatal 0x1000000000000019 error-gt1-soc-fatal-psf-csc-0 0x100000000000001a error-gt1-soc-fatal-psf-csc-1 0x100000000000001b error-gt1-soc-fatal-psf-csc-2 0x100000000000001c error-gt1-soc-fatal-punit 0x100000000000001d error-gt1-soc-fatal-psf-0 0x100000000000001e error-gt1-soc-fatal-psf-1 0x100000000000001f error-gt1-soc-fatal-psf-2 0x1000000000000020 error-gt1-soc-fatal-cd0 0x1000000000000021 error-gt1-soc-fatal-cd0-mdfi 0x1000000000000022 error-gt1-soc-fatal-mdfi-east 0x1000000000000023 error-gt1-soc-fatal-mdfi-south 0x1000000000000024 error-gt1-soc-fatal-hbm-ss0-0 0x1000000000000025 error-gt1-soc-fatal-hbm-ss0-1 0x1000000000000026 error-gt1-soc-fatal-hbm-ss0-2 0x1000000000000027 error-gt1-soc-fatal-hbm-ss0-3 0x1000000000000028 error-gt1-soc-fatal-hbm-ss0-4 0x1000000000000029 error-gt1-soc-fatal-hbm-ss0-5 0x100000000000002a error-gt1-soc-fatal-hbm-ss0-6 0x100000000000002b error-gt1-soc-fatal-hbm-ss0-7 0x100000000000002c error-gt1-soc-fatal-hbm-ss1-0 0x100000000000002d error-gt1-soc-fatal-hbm-ss1-1 0x100000000000002e error-gt1-soc-fatal-hbm-ss1-2 0x100000000000002f error-gt1-soc-fatal-hbm-ss1-3 0x1000000000000030 error-gt1-soc-fatal-hbm-ss1-4 0x1000000000000031 error-gt1-soc-fatal-hbm-ss1-5 0x1000000000000032 error-gt1-soc-fatal-hbm-ss1-6 0x1000000000000033 error-gt1-soc-fatal-hbm-ss1-7 0x1000000000000034 error-gt1-soc-fatal-hbm-ss2-0 0x1000000000000035 error-gt1-soc-fatal-hbm-ss2-1 0x1000000000000036 error-gt1-soc-fatal-hbm-ss2-2 0x1000000000000037 error-gt1-soc-fatal-hbm-ss2-3 0x1000000000000038 error-gt1-soc-fatal-hbm-ss2-4 0x1000000000000039 error-gt1-soc-fatal-hbm-ss2-5 0x100000000000003a error-gt1-soc-fatal-hbm-ss2-6 0x100000000000003b error-gt1-soc-fatal-hbm-ss2-7 0x100000000000003c error-gt1-soc-fatal-hbm-ss3-0 0x100000000000003d error-gt1-soc-fatal-hbm-ss3-1 0x100000000000003e error-gt1-soc-fatal-hbm-ss3-2 0x100000000000003f error-gt1-soc-fatal-hbm-ss3-3 0x1000000000000040 error-gt1-soc-fatal-hbm-ss3-4 0x1000000000000041 error-gt1-soc-fatal-hbm-ss3-5 0x1000000000000042 error-gt1-soc-fatal-hbm-ss3-6 0x1000000000000043 error-gt1-soc-fatal-hbm-ss3-7 0x1000000000000044Cc Alex Deucher <[email protected]>Cc David Airlie <[email protected]>Cc Simona Vetter <[email protected]>Cc Joonas Lahtinen <[email protected]>Cc Rodrigo Vivi <[email protected]>Cc Hawking Zhang <[email protected]>Cc Lijo Lazar <[email protected]>Cc Ruhl, Michael J <[email protected]>Cc Riana Tauro <[email protected]>Cc Anshuman Gupta <[email protected]>Aravind Iddamsetty (5):
drm/netlink: Add netlink infrastructure drm/xe/RAS: Register netlink capability drm/xe/RAS: Expose the error counters drm/netlink: Define multicast groups
drm/xe/RAS: send multicast event on occurrence of an errordrivers/gpu/drm/Makefile | 1 + drivers/gpu/drm/drm_drv.c | 7 + drivers/gpu/drm/drm_netlink.c | 219 +++++++++++ drivers/gpu/drm/xe/Makefile | 2 + drivers/gpu/drm/xe/xe_device.c | 6 + drivers/gpu/drm/xe/xe_device_types.h | 1 + drivers/gpu/drm/xe/xe_hw_error.c | 56 ++- drivers/gpu/drm/xe/xe_netlink.c | 531 +++++++++++++++++++++++++++ include/drm/drm_device.h | 10 + include/drm/drm_drv.h | 7 + include/drm/drm_netlink.h | 46 +++ include/uapi/drm/drm_netlink.h | 105 ++++++ include/uapi/drm/xe_drm.h | 85 +++++ 13 files changed, 1071 insertions(+), 5 deletions(-) create mode 100644 drivers/gpu/drm/drm_netlink.c create mode 100644 drivers/gpu/drm/xe/xe_netlink.c create mode 100644 include/drm/drm_netlink.h create mode 100644 include/uapi/drm/drm_netlink.hSignature · 1 line
2.25.1
Quoted 126 lines · Rodrigo Vivi
Quoted 125 lines · Rodrigo Vivi
Quoted 38 lines · Aravind Iddamsetty
Revisiting this patch series to address pending feedback and help move the discussion towards a conclusion. This revision includes updates based on previous comments[1] and aims to clarify outstanding concerns. Specifically added command to facility reporting errors from IP blocks to support AMDGPU driver model of RAS.
[1]: https://lore.kernel.org/all/[email protected]/I sincerely appreciate everyones patience and thoughtful reviews so far, and I hope this refreshed series facilitates the final evaluation and acceptance.
Please feel free to share any further suggestions or questions.
Thank you for your continued consideration.
----------------------------------------------------------------------Our hardware supports
RAS(Reliability, Availability, Serviceability)by reporting the errors to the host, which the KMD processes and exposes a set of error counters which can be used by observability tools to take corrective actions or repairs. Traditionally there were being exposed via PMU (for relative counters) and sysfs interface (for absolute value) in our internal branch. But, due to the limitations in this approach to use two interfaces and also not able to have an event based reporting or configurability, an alternative approach to try netlink was suggested by community for drm subsystem wide UAPI for RAS and telemetry as discussed in [2].This [2] is the inspiration to this series. It uses the generic netlink(genl) family subsystem and exposes a set of commands that can be used by every drm driver, the framework provides a means to have custom commands too. Each drm driver instance in this example xe driver instance registers a family and operations to the genl subsystem through which it enumerates and reports the error counters. An event based notification is also supported to which userpace can subscribe to and be notified when any error occurs and read the error counter this avoids continuous polling on error counter. This can also be extended to threshold based notification.
I'm bringing some thoughts below and I'd like to get inputs from folks involved in the original discussions here please.
Any thought is welcomed so we can move faster towards a real GPU standard RAS solution.Quoted 6 lines
This series is on top [3] series which introduces error counting infra in Xe driver.
[3]: https://lore.kernel.org/all/[email protected]/V5:
Add support to read error corresponding to an IP BLOCKI honestly don't believe that this version solves all the concerns raised by AMD folks in the previous reviews. It is true that this is bringing ways of reading errors per IP block, but if I understood them correctly, they would like better (and separate) ways to declare and handle the errors coming from different IP block, rather than simply reading/querying for them filtered out.
So, I have som grouping ideas below.
Quoted 3 lines
v4:
1. Rebase
2. renamedrm_genl_sendtodrm_genl_replyBut before going to the ideas below I'd like to also raise the naming issue that I see with this proposal.
I was recently running some experiments to devlink with this and similar cases. I don't believe that devlink is a good fit for our drm-ras. It is way too much centric on network devices and any addition there to our GPU RAS would be a heavy lift. But, there are some good things from there that we could perhaps get inspiration from.
Starting from the name. devlink is the name of the tool and the name of the framework. It uses netlink on the back, but totally abstracting that. Here in this version we can see:
drm_ras: the tooldrm_netlink: the abstraction
drm_genl_*: the wrapper?So, I believe that as devlink we should have a single name for everything and avoid wrappers but providing the real module registration, with groups, and functions. Entirely abstracting the netlink and focusing on the RAS functionalities.
I'm terrible with naming, but playing a bit with AI for some suggestions, I'd say that my favorites are:
drmras - no '_' like most of the tools, but not only for the tool, but also for the files and functions.
drmlink - more link, but less ras :/
grill - GPU RAS Interface Link LayerFor the rest of the examples below I'm going with grill, but let me know your preferences.
Quoted 17 lines
3. catch error from
xa_storeand handle appropriately
4. presentlyxe_list_errorsfills blank data for IGFX, prevent it by having an early check ofIS_DGFX(Michael J. Ruhl)v3:
1. Rebase on latest RAS series for XE
2. dropDRIVER_NETLINKflag and use thedriver_genl_opsstructure to register to netlink subsystemv2: define common interfaces to genl netlink subsystem that all drm drivers can leverage.
Below is an example tool
drm_raswhich demonstrates the use of the
supported commands. The tool will be sent to ML with the subject
"[RFC i-g-t v3 0/1] A tool to demonstrate use of netlink sockets to read RAS error counters"
https://lore.kernel.org/all/[email protected]/Quoted 4 lines
read single error counter:
$ ./drm_ras READ_ONE --device=drm:/dev/dri/card1 --error_id=0x0000000000000005 counter value 0no need for --device, that should be mandatory argument. And we could accept BDF or card identification
$ grill list 00:03:00.0 - card0 - xe$ grill 00:03:00.0 list # Querying available modules. monitor - global erros - gt erros - soc
Quoted 5 lines
Quoted 1 line
Yes, my idea is that driver should be able to register modules and group per module
Please allow me to emphasize here that the group registration is just to make this extensible and also to accommodate the AMD case, but not change the original essence of the goal which is to create the drm-ras solution.
I'm not sure if i correctly understood the group reservation you are referring too.
Like i mentioned earlier the AMD's request is not register a group of IP as a separate
netlink family but rather have the ability to present the errors at IP block level
with the commands this series proposes.
IIUC the modules have different functionalities,I believe having them as a separate
netlink family might be an overkill, I believe we can support them as commands,
looking at devlink too flash update, health are all commands they are not registered as separate family.
Quoted 10 lines
Quoted 3 lines
GRILL would be designed to accommodate multiple kinds of RAS modules, each module, with groups, categories and operations.
also let me take back on the naming here.
Please let's go with the obvious
drm_ras.Perhaps
drm_rasfor the code here and drmras (one-word) for the IGT tool.Quoted 1 line
Modules: monitor, error, flash?!, etc?!
Quoted 8 lines
Here, please let me also tune it down a bit. The overall goal continue to be the creation of our drm-ras framework using netlink to report error counters and events.
Any addition is a good-to-have, but shouldn't delay the main goal. Also, any addition should be carefully reviewed individually in the future. My only wish here at this moment is to think from the very beginning in something that is expansible.
agree, but as you mentioned let's focus on error reporting and get it going as
we are designing it we can plan to make it extensible to any future requirements.
Thanks,
Aravind.
Quoted 357 lines
Thanks,
Rodrigo.Quoted 353 lines
Groups: Global or per IP block depending on the HW underneath
Categories: Sub-groups like correctable-error vs uncorrectable-error for instance if/where
it makes sense.
Operations: Monitor: set-threshold / listen (listen is just a tool operation, but every monitor
needs to provide events over netlink)
Error: read, clear, logs$ grill 00:03:00.0 error global counter list # List all available counters in this gpu$ grill 00:03:00.0 error global counter show soc_fatal_hbm2_chnl0 # Show a specific counter.$ grill 00:03:00.0 error global log # Print all the stashed CPER logs (stash can be hw/fw/sw or a combination - in AMD case it is a dump of their debugfs ring)So, I'm sure the next question is what if the log is global, but the counters are not? Well, perhaps we should have different Modules for error-counter split from error-logging ?!
So yes, my thoughts still have some opens, but I'd like to hear your thoughts and opinions on the overall idea here.
Thanks in advance,
Rodrigo.Quoted 324 lines
read all error counters:
$ ./drm_ras READ_ALL --device=drm:/dev/dri/card1 name config-id countererror-gt0-correctable-guc 0x0000000000000001 0 error-gt0-correctable-slm 0x0000000000000003 0 error-gt0-correctable-eu-ic 0x0000000000000004 0 error-gt0-correctable-eu-grf 0x0000000000000005 0 error-gt0-fatal-guc 0x0000000000000009 0 error-gt0-fatal-slm 0x000000000000000d 0 error-gt0-fatal-eu-grf 0x000000000000000f 0 error-gt0-fatal-fpu 0x0000000000000010 0 error-gt0-fatal-tlb 0x0000000000000011 0 error-gt0-fatal-l3-fabric 0x0000000000000012 0 error-gt0-correctable-subslice 0x0000000000000013 0 error-gt0-correctable-l3bank 0x0000000000000014 0 error-gt0-fatal-subslice 0x0000000000000015 0 error-gt0-fatal-l3bank 0x0000000000000016 0 error-gt0-sgunit-correctable 0x0000000000000017 0 error-gt0-sgunit-nonfatal 0x0000000000000018 0 error-gt0-sgunit-fatal 0x0000000000000019 0 error-gt0-soc-fatal-psf-csc-0 0x000000000000001a 0 error-gt0-soc-fatal-psf-csc-1 0x000000000000001b 0 error-gt0-soc-fatal-psf-csc-2 0x000000000000001c 0 error-gt0-soc-fatal-punit 0x000000000000001d 0 error-gt0-soc-fatal-psf-0 0x000000000000001e 0 error-gt0-soc-fatal-psf-1 0x000000000000001f 0 error-gt0-soc-fatal-psf-2 0x0000000000000020 0 error-gt0-soc-fatal-cd0 0x0000000000000021 0 error-gt0-soc-fatal-cd0-mdfi 0x0000000000000022 0 error-gt0-soc-fatal-mdfi-east 0x0000000000000023 0 error-gt0-soc-fatal-mdfi-south 0x0000000000000024 0 error-gt0-soc-fatal-hbm-ss0-0 0x0000000000000025 0 error-gt0-soc-fatal-hbm-ss0-1 0x0000000000000026 0 error-gt0-soc-fatal-hbm-ss0-2 0x0000000000000027 0 error-gt0-soc-fatal-hbm-ss0-3 0x0000000000000028 0 error-gt0-soc-fatal-hbm-ss0-4 0x0000000000000029 0 error-gt0-soc-fatal-hbm-ss0-5 0x000000000000002a 0 error-gt0-soc-fatal-hbm-ss0-6 0x000000000000002b 0 error-gt0-soc-fatal-hbm-ss0-7 0x000000000000002c 0 error-gt0-soc-fatal-hbm-ss1-0 0x000000000000002d 0 error-gt0-soc-fatal-hbm-ss1-1 0x000000000000002e 0 error-gt0-soc-fatal-hbm-ss1-2 0x000000000000002f 0 error-gt0-soc-fatal-hbm-ss1-3 0x0000000000000030 0 error-gt0-soc-fatal-hbm-ss1-4 0x0000000000000031 0 error-gt0-soc-fatal-hbm-ss1-5 0x0000000000000032 0 error-gt0-soc-fatal-hbm-ss1-6 0x0000000000000033 0 error-gt0-soc-fatal-hbm-ss1-7 0x0000000000000034 0 error-gt0-soc-fatal-hbm-ss2-0 0x0000000000000035 0 error-gt0-soc-fatal-hbm-ss2-1 0x0000000000000036 0 error-gt0-soc-fatal-hbm-ss2-2 0x0000000000000037 0 error-gt0-soc-fatal-hbm-ss2-3 0x0000000000000038 0 error-gt0-soc-fatal-hbm-ss2-4 0x0000000000000039 0 error-gt0-soc-fatal-hbm-ss2-5 0x000000000000003a 0 error-gt0-soc-fatal-hbm-ss2-6 0x000000000000003b 0 error-gt0-soc-fatal-hbm-ss2-7 0x000000000000003c 0 error-gt0-soc-fatal-hbm-ss3-0 0x000000000000003d 0 error-gt0-soc-fatal-hbm-ss3-1 0x000000000000003e 0 error-gt0-soc-fatal-hbm-ss3-2 0x000000000000003f 0 error-gt0-soc-fatal-hbm-ss3-3 0x0000000000000040 0 error-gt0-soc-fatal-hbm-ss3-4 0x0000000000000041 0 error-gt0-soc-fatal-hbm-ss3-5 0x0000000000000042 0 error-gt0-soc-fatal-hbm-ss3-6 0x0000000000000043 0 error-gt0-soc-fatal-hbm-ss3-7 0x0000000000000044 0 error-gt0-gsc-correctable-sram-ecc 0x0000000000000045 0 error-gt0-gsc-nonfatal-mia-shutdown 0x0000000000000046 0 error-gt0-gsc-nonfatal-mia-int 0x0000000000000047 0 error-gt0-gsc-nonfatal-sram-ecc 0x0000000000000048 0 error-gt0-gsc-nonfatal-wdg-timeout 0x0000000000000049 0 error-gt0-gsc-nonfatal-rom-parity 0x000000000000004a 0 error-gt0-gsc-nonfatal-ucode-parity 0x000000000000004b 0 error-gt0-gsc-nonfatal-glitch-det 0x000000000000004c 0 error-gt0-gsc-nonfatal-fuse-pull 0x000000000000004d 0 error-gt0-gsc-nonfatal-fuse-crc-check 0x000000000000004e 0 error-gt0-gsc-nonfatal-selfmbist 0x000000000000004f 0 error-gt0-gsc-nonfatal-aon-parity 0x0000000000000050 0 error-gt1-correctable-guc 0x1000000000000001 0 error-gt1-correctable-slm 0x1000000000000003 0 error-gt1-correctable-eu-ic 0x1000000000000004 0 error-gt1-correctable-eu-grf 0x1000000000000005 0 error-gt1-fatal-guc 0x1000000000000009 0 error-gt1-fatal-slm 0x100000000000000d 0 error-gt1-fatal-eu-grf 0x100000000000000f 0 error-gt1-fatal-fpu 0x1000000000000010 0 error-gt1-fatal-tlb 0x1000000000000011 0 error-gt1-fatal-l3-fabric 0x1000000000000012 0 error-gt1-correctable-subslice 0x1000000000000013 0 error-gt1-correctable-l3bank 0x1000000000000014 0 error-gt1-fatal-subslice 0x1000000000000015 0 error-gt1-fatal-l3bank 0x1000000000000016 0 error-gt1-sgunit-correctable 0x1000000000000017 0 error-gt1-sgunit-nonfatal 0x1000000000000018 0 error-gt1-sgunit-fatal 0x1000000000000019 0 error-gt1-soc-fatal-psf-csc-0 0x100000000000001a 0 error-gt1-soc-fatal-psf-csc-1 0x100000000000001b 0 error-gt1-soc-fatal-psf-csc-2 0x100000000000001c 0 error-gt1-soc-fatal-punit 0x100000000000001d 0 error-gt1-soc-fatal-psf-0 0x100000000000001e 0 error-gt1-soc-fatal-psf-1 0x100000000000001f 0 error-gt1-soc-fatal-psf-2 0x1000000000000020 0 error-gt1-soc-fatal-cd0 0x1000000000000021 0 error-gt1-soc-fatal-cd0-mdfi 0x1000000000000022 0 error-gt1-soc-fatal-mdfi-east 0x1000000000000023 0 error-gt1-soc-fatal-mdfi-south 0x1000000000000024 0 error-gt1-soc-fatal-hbm-ss0-0 0x1000000000000025 0 error-gt1-soc-fatal-hbm-ss0-1 0x1000000000000026 0 error-gt1-soc-fatal-hbm-ss0-2 0x1000000000000027 0 error-gt1-soc-fatal-hbm-ss0-3 0x1000000000000028 0 error-gt1-soc-fatal-hbm-ss0-4 0x1000000000000029 0 error-gt1-soc-fatal-hbm-ss0-5 0x100000000000002a 0 error-gt1-soc-fatal-hbm-ss0-6 0x100000000000002b 0 error-gt1-soc-fatal-hbm-ss0-7 0x100000000000002c 0 error-gt1-soc-fatal-hbm-ss1-0 0x100000000000002d 0 error-gt1-soc-fatal-hbm-ss1-1 0x100000000000002e 0 error-gt1-soc-fatal-hbm-ss1-2 0x100000000000002f 0 error-gt1-soc-fatal-hbm-ss1-3 0x1000000000000030 0 error-gt1-soc-fatal-hbm-ss1-4 0x1000000000000031 0 error-gt1-soc-fatal-hbm-ss1-5 0x1000000000000032 0 error-gt1-soc-fatal-hbm-ss1-6 0x1000000000000033 0 error-gt1-soc-fatal-hbm-ss1-7 0x1000000000000034 0 error-gt1-soc-fatal-hbm-ss2-0 0x1000000000000035 0 error-gt1-soc-fatal-hbm-ss2-1 0x1000000000000036 0 error-gt1-soc-fatal-hbm-ss2-2 0x1000000000000037 0 error-gt1-soc-fatal-hbm-ss2-3 0x1000000000000038 0 error-gt1-soc-fatal-hbm-ss2-4 0x1000000000000039 0 error-gt1-soc-fatal-hbm-ss2-5 0x100000000000003a 0 error-gt1-soc-fatal-hbm-ss2-6 0x100000000000003b 0 error-gt1-soc-fatal-hbm-ss2-7 0x100000000000003c 0 error-gt1-soc-fatal-hbm-ss3-0 0x100000000000003d 0 error-gt1-soc-fatal-hbm-ss3-1 0x100000000000003e 0 error-gt1-soc-fatal-hbm-ss3-2 0x100000000000003f 0 error-gt1-soc-fatal-hbm-ss3-3 0x1000000000000040 0 error-gt1-soc-fatal-hbm-ss3-4 0x1000000000000041 0 error-gt1-soc-fatal-hbm-ss3-5 0x1000000000000042 0 error-gt1-soc-fatal-hbm-ss3-6 0x1000000000000043 0 error-gt1-soc-fatal-hbm-ss3-7 0x1000000000000044 0wait on a error event:
$ ./drm_ras WAIT_ON_EVENT --device=drm:/dev/dri/card1 waiting for error event error event received counter value 0list all errors:
$ ./drm_ras LIST_ERRORS --device=drm:/dev/dri/card1 name config-iderror-gt0-correctable-guc 0x0000000000000001 error-gt0-correctable-slm 0x0000000000000003 error-gt0-correctable-eu-ic 0x0000000000000004 error-gt0-correctable-eu-grf 0x0000000000000005 error-gt0-fatal-guc 0x0000000000000009 error-gt0-fatal-slm 0x000000000000000d error-gt0-fatal-eu-grf 0x000000000000000f error-gt0-fatal-fpu 0x0000000000000010 error-gt0-fatal-tlb 0x0000000000000011 error-gt0-fatal-l3-fabric 0x0000000000000012 error-gt0-correctable-subslice 0x0000000000000013 error-gt0-correctable-l3bank 0x0000000000000014 error-gt0-fatal-subslice 0x0000000000000015 error-gt0-fatal-l3bank 0x0000000000000016 error-gt0-sgunit-correctable 0x0000000000000017 error-gt0-sgunit-nonfatal 0x0000000000000018 error-gt0-sgunit-fatal 0x0000000000000019 error-gt0-soc-fatal-psf-csc-0 0x000000000000001a error-gt0-soc-fatal-psf-csc-1 0x000000000000001b error-gt0-soc-fatal-psf-csc-2 0x000000000000001c error-gt0-soc-fatal-punit 0x000000000000001d error-gt0-soc-fatal-psf-0 0x000000000000001e error-gt0-soc-fatal-psf-1 0x000000000000001f error-gt0-soc-fatal-psf-2 0x0000000000000020 error-gt0-soc-fatal-cd0 0x0000000000000021 error-gt0-soc-fatal-cd0-mdfi 0x0000000000000022 error-gt0-soc-fatal-mdfi-east 0x0000000000000023 error-gt0-soc-fatal-mdfi-south 0x0000000000000024 error-gt0-soc-fatal-hbm-ss0-0 0x0000000000000025 error-gt0-soc-fatal-hbm-ss0-1 0x0000000000000026 error-gt0-soc-fatal-hbm-ss0-2 0x0000000000000027 error-gt0-soc-fatal-hbm-ss0-3 0x0000000000000028 error-gt0-soc-fatal-hbm-ss0-4 0x0000000000000029 error-gt0-soc-fatal-hbm-ss0-5 0x000000000000002a error-gt0-soc-fatal-hbm-ss0-6 0x000000000000002b error-gt0-soc-fatal-hbm-ss0-7 0x000000000000002c error-gt0-soc-fatal-hbm-ss1-0 0x000000000000002d error-gt0-soc-fatal-hbm-ss1-1 0x000000000000002e error-gt0-soc-fatal-hbm-ss1-2 0x000000000000002f error-gt0-soc-fatal-hbm-ss1-3 0x0000000000000030 error-gt0-soc-fatal-hbm-ss1-4 0x0000000000000031 error-gt0-soc-fatal-hbm-ss1-5 0x0000000000000032 error-gt0-soc-fatal-hbm-ss1-6 0x0000000000000033 error-gt0-soc-fatal-hbm-ss1-7 0x0000000000000034 error-gt0-soc-fatal-hbm-ss2-0 0x0000000000000035 error-gt0-soc-fatal-hbm-ss2-1 0x0000000000000036 error-gt0-soc-fatal-hbm-ss2-2 0x0000000000000037 error-gt0-soc-fatal-hbm-ss2-3 0x0000000000000038 error-gt0-soc-fatal-hbm-ss2-4 0x0000000000000039 error-gt0-soc-fatal-hbm-ss2-5 0x000000000000003a error-gt0-soc-fatal-hbm-ss2-6 0x000000000000003b error-gt0-soc-fatal-hbm-ss2-7 0x000000000000003c error-gt0-soc-fatal-hbm-ss3-0 0x000000000000003d error-gt0-soc-fatal-hbm-ss3-1 0x000000000000003e error-gt0-soc-fatal-hbm-ss3-2 0x000000000000003f error-gt0-soc-fatal-hbm-ss3-3 0x0000000000000040 error-gt0-soc-fatal-hbm-ss3-4 0x0000000000000041 error-gt0-soc-fatal-hbm-ss3-5 0x0000000000000042 error-gt0-soc-fatal-hbm-ss3-6 0x0000000000000043 error-gt0-soc-fatal-hbm-ss3-7 0x0000000000000044 error-gt0-gsc-correctable-sram-ecc 0x0000000000000045 error-gt0-gsc-nonfatal-mia-shutdown 0x0000000000000046 error-gt0-gsc-nonfatal-mia-int 0x0000000000000047 error-gt0-gsc-nonfatal-sram-ecc 0x0000000000000048 error-gt0-gsc-nonfatal-wdg-timeout 0x0000000000000049 error-gt0-gsc-nonfatal-rom-parity 0x000000000000004a error-gt0-gsc-nonfatal-ucode-parity 0x000000000000004b error-gt0-gsc-nonfatal-glitch-det 0x000000000000004c error-gt0-gsc-nonfatal-fuse-pull 0x000000000000004d error-gt0-gsc-nonfatal-fuse-crc-check 0x000000000000004e error-gt0-gsc-nonfatal-selfmbist 0x000000000000004f error-gt0-gsc-nonfatal-aon-parity 0x0000000000000050 error-gt1-correctable-guc 0x1000000000000001 error-gt1-correctable-slm 0x1000000000000003 error-gt1-correctable-eu-ic 0x1000000000000004 error-gt1-correctable-eu-grf 0x1000000000000005 error-gt1-fatal-guc 0x1000000000000009 error-gt1-fatal-slm 0x100000000000000d error-gt1-fatal-eu-grf 0x100000000000000f error-gt1-fatal-fpu 0x1000000000000010 error-gt1-fatal-tlb 0x1000000000000011 error-gt1-fatal-l3-fabric 0x1000000000000012 error-gt1-correctable-subslice 0x1000000000000013 error-gt1-correctable-l3bank 0x1000000000000014 error-gt1-fatal-subslice 0x1000000000000015 error-gt1-fatal-l3bank 0x1000000000000016 error-gt1-sgunit-correctable 0x1000000000000017 error-gt1-sgunit-nonfatal 0x1000000000000018 error-gt1-sgunit-fatal 0x1000000000000019 error-gt1-soc-fatal-psf-csc-0 0x100000000000001a error-gt1-soc-fatal-psf-csc-1 0x100000000000001b error-gt1-soc-fatal-psf-csc-2 0x100000000000001c error-gt1-soc-fatal-punit 0x100000000000001d error-gt1-soc-fatal-psf-0 0x100000000000001e error-gt1-soc-fatal-psf-1 0x100000000000001f error-gt1-soc-fatal-psf-2 0x1000000000000020 error-gt1-soc-fatal-cd0 0x1000000000000021 error-gt1-soc-fatal-cd0-mdfi 0x1000000000000022 error-gt1-soc-fatal-mdfi-east 0x1000000000000023 error-gt1-soc-fatal-mdfi-south 0x1000000000000024 error-gt1-soc-fatal-hbm-ss0-0 0x1000000000000025 error-gt1-soc-fatal-hbm-ss0-1 0x1000000000000026 error-gt1-soc-fatal-hbm-ss0-2 0x1000000000000027 error-gt1-soc-fatal-hbm-ss0-3 0x1000000000000028 error-gt1-soc-fatal-hbm-ss0-4 0x1000000000000029 error-gt1-soc-fatal-hbm-ss0-5 0x100000000000002a error-gt1-soc-fatal-hbm-ss0-6 0x100000000000002b error-gt1-soc-fatal-hbm-ss0-7 0x100000000000002c error-gt1-soc-fatal-hbm-ss1-0 0x100000000000002d error-gt1-soc-fatal-hbm-ss1-1 0x100000000000002e error-gt1-soc-fatal-hbm-ss1-2 0x100000000000002f error-gt1-soc-fatal-hbm-ss1-3 0x1000000000000030 error-gt1-soc-fatal-hbm-ss1-4 0x1000000000000031 error-gt1-soc-fatal-hbm-ss1-5 0x1000000000000032 error-gt1-soc-fatal-hbm-ss1-6 0x1000000000000033 error-gt1-soc-fatal-hbm-ss1-7 0x1000000000000034 error-gt1-soc-fatal-hbm-ss2-0 0x1000000000000035 error-gt1-soc-fatal-hbm-ss2-1 0x1000000000000036 error-gt1-soc-fatal-hbm-ss2-2 0x1000000000000037 error-gt1-soc-fatal-hbm-ss2-3 0x1000000000000038 error-gt1-soc-fatal-hbm-ss2-4 0x1000000000000039 error-gt1-soc-fatal-hbm-ss2-5 0x100000000000003a error-gt1-soc-fatal-hbm-ss2-6 0x100000000000003b error-gt1-soc-fatal-hbm-ss2-7 0x100000000000003c error-gt1-soc-fatal-hbm-ss3-0 0x100000000000003d error-gt1-soc-fatal-hbm-ss3-1 0x100000000000003e error-gt1-soc-fatal-hbm-ss3-2 0x100000000000003f error-gt1-soc-fatal-hbm-ss3-3 0x1000000000000040 error-gt1-soc-fatal-hbm-ss3-4 0x1000000000000041 error-gt1-soc-fatal-hbm-ss3-5 0x1000000000000042 error-gt1-soc-fatal-hbm-ss3-6 0x1000000000000043 error-gt1-soc-fatal-hbm-ss3-7 0x1000000000000044Cc Alex Deucher <[email protected]>Cc David Airlie <[email protected]>Cc Simona Vetter <[email protected]>Cc Joonas Lahtinen <[email protected]>Cc Rodrigo Vivi <[email protected]>Cc Hawking Zhang <[email protected]>Cc Lijo Lazar <[email protected]>Cc Ruhl, Michael J <[email protected]>Cc Riana Tauro <[email protected]>Cc Anshuman Gupta <[email protected]>Aravind Iddamsetty (5):
drm/netlink: Add netlink infrastructure drm/xe/RAS: Register netlink capability drm/xe/RAS: Expose the error counters drm/netlink: Define multicast groups
drm/xe/RAS: send multicast event on occurrence of an errordrivers/gpu/drm/Makefile | 1 + drivers/gpu/drm/drm_drv.c | 7 + drivers/gpu/drm/drm_netlink.c | 219 +++++++++++ drivers/gpu/drm/xe/Makefile | 2 + drivers/gpu/drm/xe/xe_device.c | 6 + drivers/gpu/drm/xe/xe_device_types.h | 1 + drivers/gpu/drm/xe/xe_hw_error.c | 56 ++- drivers/gpu/drm/xe/xe_netlink.c | 531 +++++++++++++++++++++++++++ include/drm/drm_device.h | 10 + include/drm/drm_drv.h | 7 + include/drm/drm_netlink.h | 46 +++ include/uapi/drm/drm_netlink.h | 105 ++++++ include/uapi/drm/xe_drm.h | 85 +++++ 13 files changed, 1071 insertions(+), 5 deletions(-) create mode 100644 drivers/gpu/drm/drm_netlink.c create mode 100644 drivers/gpu/drm/xe/xe_netlink.c create mode 100644 include/drm/drm_netlink.h create mode 100644 include/uapi/drm/drm_netlink.hSignature · 1 line
2.25.1
Quoted 51 lines · Rodrigo Vivi
Quoted 38 lines · Aravind Iddamsetty
Revisiting this patch series to address pending feedback and help move the discussion towards a conclusion. This revision includes updates based on previous comments[1] and aims to clarify outstanding concerns. Specifically added command to facility reporting errors from IP blocks to support AMDGPU driver model of RAS.
[1]: https://lore.kernel.org/all/[email protected]/I sincerely appreciate everyones patience and thoughtful reviews so far, and I hope this refreshed series facilitates the final evaluation and acceptance.
Please feel free to share any further suggestions or questions.
Thank you for your continued consideration.
----------------------------------------------------------------------Our hardware supports
RAS(Reliability, Availability, Serviceability)by reporting the errors to the host, which the KMD processes and exposes a set of error counters which can be used by observability tools to take corrective actions or repairs. Traditionally there were being exposed via PMU (for relative counters) and sysfs interface (for absolute value) in our internal branch. But, due to the limitations in this approach to use two interfaces and also not able to have an event based reporting or configurability, an alternative approach to try netlink was suggested by community for drm subsystem wide UAPI for RAS and telemetry as discussed in [2].This [2] is the inspiration to this series. It uses the generic netlink(genl) family subsystem and exposes a set of commands that can be used by every drm driver, the framework provides a means to have custom commands too. Each drm driver instance in this example xe driver instance registers a family and operations to the genl subsystem through which it enumerates and reports the error counters. An event based notification is also supported to which userpace can subscribe to and be notified when any error occurs and read the error counter this avoids continuous polling on error counter. This can also be extended to threshold based notification.
I'm bringing some thoughts below and I'd like to get inputs from folks involved in the original discussions here please.
Any thought is welcomed so we can move faster towards a real GPU standard RAS solution.Quoted 6 lines
This series is on top [3] series which introduces error counting infra in Xe driver.
[3]: https://lore.kernel.org/all/[email protected]/V5:
Add support to read error corresponding to an IP BLOCK
Quoted 7 lines
I honestly don't believe that this version solves all the concerns raised by AMD folks in the previous reviews. It is true that this is bringing ways of reading errors per IP block, but if I understood them correctly, they would like better (and separate) ways to declare and handle the errors coming from different IP block, rather than simply reading/querying for them filtered out.
So, I have som grouping ideas below.
As per the comment from Lijo,
https://lore.kernel.org/all/[email protected]/
the errors are grouped per bitmask, they are not expecting a separation at netlink level.
<31:24> = Block Id
<23:16> subblock id
<15:8> - interested instance
<7:0> - error_type
The interface should support errors per IP block and instance, which the current series support via DRM_RAS_CMD_READ_BLOCK.
when driver receives the command DRM_RAS_CMD_READ_BLOCK it is supposed to decipher the bits based on the above bitsmask.
The query command is expected to list the blocks and instances available, the counters of which will be read via DRM_RAS_CMD_READ_BLOCK.
Quoted 20 lines
Quoted 4 lines
v4:
1. Rebase
2. renamedrm_genl_sendtodrm_genl_replyBut before going to the ideas below I'd like to also raise the naming issue that I see with this proposal.
I was recently running some experiments to devlink with this and similar cases. I don't believe that devlink is a good fit for our drm-ras. It is way too much centric on network devices and any addition there to our GPU RAS would be a heavy lift. But, there are some good things from there that we could perhaps get inspiration from.
Starting from the name. devlink is the name of the tool and the name of the framework. It uses netlink on the back, but totally abstracting that. Here in this version we can see:
drm_ras: the tooldrm_netlink: the abstraction
drm_genl_*: the wrapper?
Quoted 4 lines
So, I believe that as devlink we should have a single name for everything and avoid wrappers but providing the real module registration, with groups, and functions. Entirely abstracting the netlink and focusing on the RAS functionalities.
sounds interesting and I feel it looks clean too. But that does mean we completely handle
the netlink framework inside the drm layer and not at the driver and expose callback ops to
drm drivers.
Thanks,
Aravind.