[Why]
When host trigger a whole gpu reset, guest will keep
waiting till host finish reset. But there's a work
queue in guest exchanging data between vf&pf which need
to access frame buffer. During whole gpu reset, frame
buffer is not accessable, and this causes the call trace.
[How]
After vf get reset notification from pf, stop data exchange.
Signed-off-by: Jingwen Chen <Jingwen.Chen2@amd.com>
Signed-off-by: Jack Zhang <Jack.Zhang1@amd.com>
Reviewed-by: Monk Liu <monk.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In Virtualization case, when one VF is sending too many
FLR requests, hypervisor would stop responding to this
VF's request for a long period of time. This is called
event guard. During this period of cooling time, guest
driver should wait instead of doing other things. After
this period of time, guest driver would resume reset
process and return to normal.
Currently, guest driver would wait 12 seconds and return fail
if it doesn't get response from host.
Solution: extend this waiting time in guest driver and poll
response periodically. Poll happens every 6 seconds and it will
last for 60 seconds.
v2: change the max repetition times from number to macro.
Signed-off-by: Jiange Zhao <Jiange.Zhao@amd.com>
Acked-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In function flr_work, we should do gpu recovery when no job
is running. Fix the logic by inverting it.
v2: modify the description
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Liu ChengZhe <ChengZhe.Liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
clients don't need reset-lock for synchronization when no
GPU recovery.
v2:
change to return the return value of down_read_killable.
v3:
if GPU recovery begin, VF ignore FLR notification.
Reviewed-by: Monk Liu <monk.liu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Dennis Li <Dennis.Li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
if other threads have holden the reset lock, recovery will
fail to try_lock. Therefore we introduce atomic hive->in_reset
and adev->in_gpu_reset, to avoid reentering GPU recovery.
v2:
drop "? true : false" in the definition of amdgpu_in_reset
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Dennis Li <Dennis.Li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Only for no job running test case need to do recover in
flr notification.
For having job in mirror list, then let guest driver to
hit job timeout, and then do recover.
Signed-off-by: jqdeng <Emily.Deng@amd.com>
Acked-by: Nirmoy Das <nirmoy.das@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The whole approach wasn't thought through till the end.
We already had a reset lock like this in the past and it caused the same problems like this one.
Completely revert the patch for now and add individual trylock protection to the hardware access functions as necessary.
This reverts commit df9c8d1aa2.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
when GPU hang, driver has multi-paths to enter amdgpu_device_gpu_recover,
the atomic adev->in_gpu_reset and hive->in_reset are used to avoid
re-entering GPU recovery.
During GPU reset and resume, it is unsafe that other threads access GPU,
which maybe cause GPU reset failed. Therefore the new rw_semaphore
adev->reset_sem is introduced, which protect GPU from being accessed by
external threads during recovery.
v2:
1. add rwlock for some ioctls, debugfs and file-close function.
2. change to use dqm->is_resetting and dqm_lock for protection in kfd
driver.
3. remove try_lock and change adev->in_gpu_reset as atomic, to avoid
re-enter GPU recovery for the same GPU hang.
v3:
1. change back to use adev->reset_sem to protect kfd callback
functions, because dqm_lock couldn't protect all codes, for example:
free_mqd must be called outside of dqm_lock;
[ 1230.176199] Hardware name: Supermicro SYS-7049GP-TRT/X11DPG-QT, BIOS 3.1 05/23/2019
[ 1230.177221] Call Trace:
[ 1230.178249] dump_stack+0x98/0xd5
[ 1230.179443] amdgpu_virt_kiq_reg_write_reg_wait+0x181/0x190 [amdgpu]
[ 1230.180673] gmc_v9_0_flush_gpu_tlb+0xcc/0x310 [amdgpu]
[ 1230.181882] amdgpu_gart_unbind+0xa9/0xe0 [amdgpu]
[ 1230.183098] amdgpu_ttm_backend_unbind+0x46/0x180 [amdgpu]
[ 1230.184239] ? ttm_bo_put+0x171/0x5f0 [ttm]
[ 1230.185394] ttm_tt_unbind+0x21/0x40 [ttm]
[ 1230.186558] ttm_tt_destroy.part.12+0x12/0x60 [ttm]
[ 1230.187707] ttm_tt_destroy+0x13/0x20 [ttm]
[ 1230.188832] ttm_bo_cleanup_memtype_use+0x36/0x80 [ttm]
[ 1230.189979] ttm_bo_put+0x1be/0x5f0 [ttm]
[ 1230.191230] amdgpu_bo_unref+0x1e/0x30 [amdgpu]
[ 1230.192522] amdgpu_amdkfd_free_gtt_mem+0xaf/0x140 [amdgpu]
[ 1230.193833] free_mqd+0x25/0x40 [amdgpu]
[ 1230.195143] destroy_queue_cpsch+0x1a7/0x270 [amdgpu]
[ 1230.196475] pqm_destroy_queue+0x105/0x260 [amdgpu]
[ 1230.197819] kfd_ioctl_destroy_queue+0x37/0x70 [amdgpu]
[ 1230.199154] kfd_ioctl+0x277/0x500 [amdgpu]
[ 1230.200458] ? kfd_ioctl_get_clock_counters+0x60/0x60 [amdgpu]
[ 1230.201656] ? tomoyo_file_ioctl+0x19/0x20
[ 1230.202831] ksys_ioctl+0x98/0xb0
[ 1230.204004] __x64_sys_ioctl+0x1a/0x20
[ 1230.205174] do_syscall_64+0x5f/0x250
[ 1230.206339] entry_SYSCALL_64_after_hwframe+0x49/0xbe
2. remove try_lock and introduce atomic hive->in_reset, to avoid
re-enter GPU recovery.
v4:
1. remove an unnecessary whitespace change in kfd_chardev.c
2. remove comment codes in amdgpu_device.c
3. add more detailed comment in commit message
4. define a wrap function amdgpu_in_reset
v5:
1. Fix some style issues.
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Suggested-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Suggested-by: Christian König <christian.koenig@amd.com>
Suggested-by: Felix Kuehling <Felix.Kuehling@amd.com>
Suggested-by: Lijo Lazar <Lijo.Lazar@amd.com>
Suggested-by: Luben Tukov <luben.tuikov@amd.com>
Signed-off-by: Dennis Li <Dennis.Li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fixes coccicheck warning:
drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c:253:2-20: WARNING: Assignment of 0/1 to bool variable
drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c:265:2-20: WARNING: Assignment of 0/1 to bool variable
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: zhengbin <zhengbin13@huawei.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Originally, due to the restriction from PSP and SMU, VF has
to send message to hypervisor driver to handle powerplay
change which is complicated and redundant. Currently, SMU
and PSP can support VF to directly handle powerplay
change by itself. Therefore, the old code about the handshake
between VF and PF to handle powerplay will be removed and VF
will use new the registers below to handshake with SMU.
mmMP1_SMN_C2PMSG_101: register to handle SMU message
mmMP1_SMN_C2PMSG_102: register to handle SMU parameter
mmMP1_SMN_C2PMSG_103: register to handle SMU response
v2: remove module parameter pp_one_vf
v3: fix the parens
v4: forbid vf to change smu feature
v5: use hwmon_attributes_visible to skip sepicified hwmon atrribute
v6: change skip condition at vega10_copy_table_to_smc
Signed-off-by: Yintian Tao <yttao@amd.com>
Acked-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
we can simplify all those unnecessary function under
SRIOV for vega10 since:
1) PSP L1 policy is by force enabled in SRIOV
2) original logic always set all flags which make itself
a dummy step
besides,
1) the ih_doorbell_range set should also be skipped
for VEGA10 SRIOV.
2) the gfx_common registers should also be skipped
for VEGA10 SRIOV.
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Emily Deng <Emily.Deng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Under Vega10 SR-IOV VF, L1 register access mode should be enabled by
default as the non-security VF will no longer be supported.
Signed-off-by: Trigger Huang <Trigger.Huang@amd.com>
Reviewed-by: Emily Deng <Emily.Deng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Set different register access mode according to the features
provided by firmware
Signed-off-by: Trigger Huang <Trigger.Huang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Every ring type can have its own timeout setting.
- V2: update lockup_timeout parameter format and cosmetic fixes
- V3: invalidate 0 and negative values
- V4: update lockup_timeout parameter format
Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
SR-IOV host side will send IDH_QUERY_ALIVE to guest VM to check
if this guest VM is still alive (not destroyed). The only thing
guest KMD need to do is to send ACK back to host.
Signed-off-by: Trigger Huang <Trigger.Huang@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Under vega10 virtualuzation, smu ip block will not be added.
Therefore, we need add pp clk query and force dpm level function
at amdgpu_virt_ops to support the feature.
v2: add get_pp_clk existence check and use kzalloc to allocate buf
v3: return -ENOMEM for allocation failure and correct the coding style
Signed-off-by: Yintian Tao <yttao@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
sriov's gpu_recover inside xgpu_ai_mailbox_flr_work would cause duplicate recover in TDR.
TDR's gpu_recover would be triggered by amdgpu_job_timedout,
that could avoid vk-cts failure by unexpected recover.
Signed-off-by: Wentao Lou <Wentao.Lou@amd.com>
Acked-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The pfvf exchange need be in exclusive mode. And add pfvf exchange in gpu
reset.
Signed-off-by: Emily Deng <Emily.Deng@amd.com>
Reviewed-By: Xiangliang Yu <Xiangliang.Yu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Check if we should call the function instead of providing the forced
flag.
v2: rebase on KFD changes (Alex)
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
After host os reset gpu reset, need to set flag in_gpu_reset to
zero.
Signed-off-by: Emily Deng <Emily.Deng@amd.com>
Reviewed-by: Monk Liu <monk.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Trivial fix to spelling mistake in pr_err error message text
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This is preparation for sharing client ID definitions
between amdgpu and amdkfd
Signed-off-by: Oak Zeng <Oak.Zeng@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
this patch actually refactor mailbox implmentations, and
all below changes are needed together to fix all those mailbox
handshake issues exposured by heavey TDR test.
1)refactor all mailbox functions based on byte accessing for mb_control
reason is to avoid touching non-related bits when writing trn/rcv part of
mailbox_control, this way some incorrect INTR sent to hypervisor
side could be avoided, and it fixes couple handshake bug.
2)trans_msg function re-impled: put a invalid
logic before transmitting message to make sure the ACK bit is in
a clear status, otherwise there is chance that ACK asserted already
before transmitting message and lead to fake ACK polling.
(hypervisor side have some tricks to workaround ACK bit being corrupted
by VF FLR which hase an side effects that may make guest side ACK bit
asserted wrongly), and clear TRANS_MSG words after message transferred.
3)for mailbox_flr_work, it is also re-worked: it takes the mutex lock
first if invoked, to block gpu recover's participate too early while
hypervisor side is doing VF FLR. (hypervisor sends FLR_NOTIFY to guest
before doing VF FLR and sentds FLR_COMPLETE after VF FLR done, and
the FLR_NOTIFY will trigger interrupt to guest which lead to
mailbox_flr_work being invoked)
This can avoid the issue that mailbox trans msg being cleared by its VF FLR.
4)for mailbox_rcv_irq IRQ routine, it should only peek msg and schedule
mailbox_flr_work, instead of ACK to hypervisor itself, because FLR_NOTIFY
msg sent from hypervisor side doesn't need VF's ACK (this is because
VF's ACK would lead to hypervisor clear its trans_valid/msg, and this
would cause handshake bug if trans_valid/msg is cleared not due to
correct VF ACK but from a wrong VF ACK like this "FLR_NOTIFY" one)
This fixed handshake bug that sometimes GUEST always couldn't receive
"READY_TO_ACCESS_GPU" msg from hypervisor.
5)seperate polling time limite accordingly:
POLL ACK cost no more than 500ms
POLL MSG cost no more than 12000ms
POLL FLR finish cost no more than 500ms
6) we still need to set adev into in_gpu_reset mode after we received
FLR_NOTIFY from host side, this can prevent innocent app wrongly succesed
to open amdgpu dri device.
FLR_NOFITY is received due to an IDLE hang detected from hypervisor side
which indicating GPU is already die in this VF.
v2:
use MACRO as the offset of mailbox_control register
don't test if NOTIFY_CMPL event in rcv_msg since it won't
recieve that message anymore
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Pixel Ding <Pixel.Ding@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
With introduction of amdgpu_gpu_recovery we don't need any more
to rely on amdgpu_lockup_timeout == 0 for disabling GPU reset.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add new parameter to control GPU recovery procedure.
v2:
Add auto logic where reset is disabled for bare metal and enabled
for SR-IOV.
Allow forced reset from debugfs.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Remove the header where it's not used.
Acked-by: Christian Konig <christian.koenig@amd.com>
Signed-off-by: Shaoyun Liu <Shaoyun.Liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Remove asic_reg/vega10 folder.
Signed-off-by: Feifei Xu <Feifei.Xu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cleanup asic_reg/vega10/NBIO folder.
Signed-off-by: Feifei Xu <Feifei.Xu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
1,new imple names amdgpu_gpu_recover which gives more hint
on what it does compared with gpu_reset
2,gpu_recover unify bare-metal and SR-IOV, only the asic reset
part is implemented differently
3,gpu_recover will increase hang job karma and mark its entity/context
as guilty if exceeds limit
V2:
4,in scheduler main routine the job from guilty context will be immedialy
fake signaled after it poped from queue and its fence be set with
"-ECANCELED" error
5,in scheduler recovery routine all jobs from the guilty entity would be
dropped
6,in run_job() routine the real IB submission would be skipped if @skip parameter
equales true or there was VRAM lost occured.
V3:
7,replace deprecated gpu reset, use new gpu recover
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Monk Liu <monk.liu@amd.com>
Signed-off-by: pding <Pixel.Ding@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
SR-IOV need to exchange some data between PF&VF through shared VRAM
PF will copy some necessary firmware and information to the shared
VRAM. It also requires some information from VF. PF will send a
key through mailbox2 to help guest calculate checksum so that it can
verify whether the data is correct.
So check the data on the specified offset of the shared VRAM, if the
checksum is right, read values from it and write some VF information
next to the data from PF.
Signed-off-by: Horace Chen <horace.chen@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This feature works for SRIOV enviroment. For non-SRIOV enviroment, the
trans_error function does nothing.
The error information includes error_code (16bit), error_flags(16bit)
and error_data(64bit). Since there are not many errors, we keep the
errors in an array and transfer all errors to Host before amdgpu
initialization function (amdgpu_device_init) exit.
Signed-off-by: Gavin Wan <Gavin.Wan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
that way we can know which job cause hang and
can do per sched reset/recovery instead of all
sched.
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Change message to debug level as VI does.
Signed-off-by: Xiangliang Yu <Xiangliang.Yu@amd.com>
Reviewed-by: Monk Liu <Monk.Liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
if sriov gpu reset is invoked by job timeout, it is run
in a global work-queue which is very slow and better not call
msleep ortherwise it takes long time to get back CPU.
so make below changes:
1: Change msleep 1 to mdelay 5
2: Ignore the ack fail from pf after time out,
because VF FLR will clear ack, sometime VF FLR is done
prior to the beginning of poll_ack so we can ignore this ack
TODO:
Put job_timedout (and the following gpu reset) in a driver thread,
instead of the global work_struct.
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Xiangliang Yu <Xiangliang.Yu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
they are lack in the bringup stage, we need them for GPU reset
feature.
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Implement mailbox protocol for AI so that guest vf can communicate
with GPU hypervisor.
Signed-off-by: Xiangliang Yu <Xiangliang.Yu@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Monk Liu <Monk.Liu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Jack Zhang