drm/amdgpu: revert "Adjust removal control flow for smu v13_0_2"

Calling amdgpu_device_ip_resume_phase1() during shutdown leaves the
HW in an active state and is an unbalanced use of the IP callbacks.

Using the IP callbacks like this can lead to memory leaks, double
free and imbalanced reference counters.

Leaving the HW in an active state can lead to DMA accesses to memory now
freed by the driver.

Both is a complete no-go for driver unload so completely revert the
workaround for now.

This reverts commit f5c7e7797060255dbc8160734ccc5ad6183c5e04.

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>

drm/amdgpu: revert "Adjust removal control flow for smu v13_0_2"

Calling amdgpu_device_ip_resume_phase1() during shutdown leaves the
HW in an active state and is an unbalanced use of the IP callbacks.

Using the IP callbacks like this can lead to memory leaks, double
free and imbalanced reference counters.

Leaving the HW in an active state can lead to DMA accesses to memory now
freed by the driver.

Both is a complete no-go for driver unload so completely revert the
workaround for now.

This reverts commit f5c7e7797060255dbc8160734ccc5ad6183c5e04.

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>
Cc: stable@vger.kernel.org

drm/amdgpu : Add hive ras recovery check

If one of the devices in the hive detects a
fatal error, need to send ras recovery reset
message to PMFW of all devices in the hive.
For that add a flag in hive to indicate that
it's undergoing ras recovery

Signed-off-by: Asad Kamal <asad.kamal@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Rework xgmi_wafl_pcs ras sw_init

To align with other IP blocks.

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Stanley Yang <Stanley.Yang@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Adjust removal control flow for smu v13_0_2

Adjust removal control flow for smu v13_0_2:
During amdgpu uninstallation, when removing the first
device, the kernel needs to first send a mode1reset message
to all gpu devices. Otherwise, smu initialization will fail
the next time amdgpu is installed.

V2:
1. Update commit comments.
2. Remove the global variable amdgpu_device_remove_cnt
and add a variable to the structure amdgpu_hive_info.
3. Use hive to detect the first removed device instead of
a global variable.

V3:
1. Update commit comments.
2. Split a patch into multiple patches.
3. The current patch does:
a. Add a work mode of AMDGPU_RESET_FOR_DEVICE_REMOVE into
the existing gpu recover path, which make all devices
in hive list only have HW reset but no resume (except
the base IP).
b. Call AMDGPU_RESET_FOR_DEVICE_REMOVE and
AMDGPU_NEED_FULL_RESET mode of amdgpu_device_gpu_recover
in amdgpu_pci_remove when removing the first device in
hive list.
c. When removing the first device, the IP blocks keyword
function call sequence is as follows:
.suspend->mode1reset->.resume(basic ip)->.hw_fini->.early_fini->.sw_fini.
^ |
|-<----------<---------<----|
The first three sequences are because of a call to
amdgpu_device_gpu_recover. The three sequences will be
executed in a loop until all devices in the hive list
are iterated.
The sequences starting from .hw_fini only apply to the
first device. Since .suspend has been called before,
except the resumed phase1 basic ip blocks, all other ip
blocks .hw_fini of current device will do nothing.
d. When removing other devices, the calling sequences is the
same as legacy:
.hw_fini -> .early_fini -> .sw_fini.
Since .suspend has been called when removing the first device,
except the resumed phase1 basic ip blocks, all of other ip
blocks .hw_fini of current device will do nothing.

Signed-off-by: YiPeng Chai <YiPeng.Chai@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Add use_xgmi_p2p module parameter

This parameter controls xGMI p2p communication, which is enabled by
default. However, it can be disabled by setting it to 0. In case xGMI
p2p is disabled in a dGPU, PCIe p2p interface will be used instead.
This parameter is ignored in GPUs that do not support xGMI
p2p configuration.

Signed-off-by: Alex Sierra <alex.sierra@amd.com>
Acked-by: Luben Tuikov <luben.tuikov@amd.com>
Acked-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Rework reset domain to be refcounted.

The reset domain contains register access semaphor
now and so needs to be present as long as each device
in a hive needs it and so it cannot be binded to XGMI
hive life cycle.
Adress this by making reset domain refcounted and pointed
by each member of the hive and the hive itself.

v4:

Fix crash on boot witrh XGMI hive by adding type to reset_domain.
XGMI will only create a new reset_domain if prevoius was of single
device type meaning it's first boot. Otherwsie it will take a
refocunt to exsiting reset_domain from the amdgou device.

Add a wrapper around reset_domain->refcount get/put
and a wrapper around send to reset wq (Lijo)

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Link: https://www.spinics.net/lists/amd-gfx/msg74121.html

drm/amdgpu: Drop hive->in_reset

Since we serialize all resets no need to protect from concurrent
resets.

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Link: https://www.spinics.net/lists/amd-gfx/msg74115.html

drm/amdgpu: Introduce reset domain

Defined a reset_domain struct such that
all the entities that go through reset
together will be serialized one against
another. Do it for both single device and
XGMI hive cases.

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Suggested-by: Christian König <ckoenig.leichtzumerken@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Link: https://www.spinics.net/lists/amd-gfx/msg74111.html

drm/amdgpu: Modify xgmi block to fit for the unified ras block data and ops

1.Modify gmc block to fit for the unified ras block data and ops.
2.Change amdgpu_xgmi_ras_funcs to amdgpu_xgmi_ras, and the corresponding variable name remove _funcs suffix.
3.Remove the const flag of gmc ras variable so that gmc ras block can be able to be inserted into amdgpu device ras block link list.
4.Invoke amdgpu_ras_register_ras_block function to register gmc ras block into amdgpu device ras block link list.
5.Remove the redundant code about gmc in amdgpu_ras.c after using the unified ras block.

Signed-off-by: yipechai <YiPeng.Chai@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: John Clements <john.clements@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdkfd: report xgmi bandwidth between direct peers to the kfd

Report the min/max bandwidth in megabytes to the kfd for direct
xgmi connections only. Indirect peers will report 0 since
indirect route is unknown.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: move xgmi ras functions to xgmi_ras_funcs

xgmi ras is not managed by gpu driver when gpu is
connected to cpu through xgmi. move all xgmi ras
functions to xgmi_ras_funcs so gpu driver only
initializes xgmi ras functions when it manages
xgmi ras.

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Dennis Li <Dennis.Li@amd.com>
Reviewed-by: John Clements <John.Clements@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: refine create and release logic of hive info

Change to dynamically create and release hive info object,
which help driver support more hives in the future.

v2:
Change to save hive object pointer in adev, to avoid locking
xgmi_mutex every time when calling amdgpu_get_xgmi_hive.

v3:
1. Change type of hive object pointer in adev from void* to
amdgpu_hive_info*.
2. remove unnecessary variable initialization.

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>

drm/amdgpu: refine codes to avoid reentering GPU recovery

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>

drm/amdgpu: revert "fix system hang issue during GPU reset"

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 df9c8d1aa278c435c30a69b8f2418b4a52fcb929.

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>

drm/amdgpu: fix system hang issue during GPU reset

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>

drm/amdgpu: fix race between pstate and remote buffer map

Vega20 arbitrates pstate at hive level and not device level. Last peer to
remote buffer unmap could drop P-State while another process is still
remote buffer mapped.

With this fix, P-States still needs to be disabled for now as SMU bug
was discovered on synchronous P2P transfers. This should be fixed in the
next FW update.

Signed-off-by: Jonathan Kim <Jonathan.Kim@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: added xgmi ras error reset sequence

added mechanism to clear xgmi ras status inbetween error queries

Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: John Clements <john.clements@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add helper funcs to detect PCS error

Since from vega20, hardware supports run-time detect
and report XGMI/WAFL PCS ras error. Add helper functions
to walkthrough every type of ras error and report it if
any.

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: move get_xgmi_relative_phy_addr to amdgpu_xgmi.c

centralize all the xgmi related function to amdgpu_xgmi.c

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: move xgmi init/fini to xgmi_add/remove_device call (v2)

For sriov, psp ip block has to be initialized before
ih block for the dynamic register programming interface
that needed for vf ih ring buffer. On the other hand,
current psp ip block hw_init function will initialize
xgmi session which actaully depends on interrupt to
return session context. This results an empty xgmi ta
session id and later failures on all the xgmi ta cmd
invoked from vf. xgmi ta session initialization has to
be done after ih ip block hw_init call.

to unify xgmi session init/fini for both bare-metal
sriov virtualization use scenario, move xgmi ta init
to xgmi_add_device call, and accordingly terminate xgmi
ta session in xgmi_remove_device call.

The existing suspend/resume sequence will not be changed.

v2: squash in return fix from Nirmoy

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Frank Min <Frank.Min@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Add task barrier to XGMI hive.

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Le Ma <Le.Ma@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: move xgmi ras fini to xgmi block

it's more suitable to put xgmi ras fini in xgmi block

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: initialize ras structures for xgmi block (v2)

init ras common interface and fs node for xgmi block

v2: remove unnecessary physical node number check before
invoking amdgpu_xgmi_ras_late_init

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Implement get num of hops between two xgmi device

KFD need to provide the info for upper level to determine the data path

Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: XGMI pstate switch initial support

Driver vote low to high pstate switch whenever there is an outstanding
XGMI mapping request. Driver vote high to low pstate when all the
outstanding XGMI mapping is terminated.

Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: revert "XGMI pstate switch initial support"

This reverts commit 9b638f9751308ae3ae8f28e0c6e9decffd97f5f9.

Adding this to the mapping is complete nonsense and the whole
implementation looks racy. This patch wasn't thoughtfully reviewed
and should be reverted for now.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Liu, Shaoyun <Shaoyun.Liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: XGMI pstate switch initial support

Driver vote low to high pstate switch whenever there is an outstanding
XGMI mapping request. Driver vote high to low pstate when all the
outstanding XGMI mapping is terminated.

Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Add sysfs entries for xgmi hive v2.

For each device a file xgmi_device_id is created.
On the first device a subdirectory named xgmi_hive_info is created,
It contains a file named hive_id and symlinks named node 1-4 linking
to each device in the hive.

v2: Return error codes instead of '-1' and few misspellings.

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>

drm/amd/amdgpu: add missing mutex lock to amdgpu_get_xgmi_hive() (v3)

v2: Move locks around in other functions so that this
function can stand on its own. Also only hold the hive
specific lock for add/remove device instead of the driver
global lock so you can't add/remove devices in parallel from
one hive.

v3: add reset_lock

Acked-by: Shaoyun.liu < Shaoyun.liu@amd.com>
Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Handle xgmi device removal.

XGMI hive has some resources allocted on device init which
needs to be deallocated when the device is unregistered.

v2: Remove creation of dedicated wq for XGMI hive reset.
v3: Use the gmc.xgmi.supported flag

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Expose hive adev list and xgmi_mutex

It's needed for device reset of entire hive.

v3:
Add per hive lock to allow avoiding duplicate resets triggered by
multiple members of same hive.
Expose amdgpu_hive_info instead of adding getter functions.

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Refactor amdgpu_xgmi_add_device

This is prep work for updating each PSP FW in hive after
GPU reset.
Split into build topology SW state and update each PSP FW in the hive.
Save topology and count of XGMI devices for reuse.

v2: Create seperate header for XGMI.

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>