Add hsaKmtRuntimeEnable and disable.
Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Change-Id: I083f9293948e975546a1b3c1334cb41499b9ab1f
The debugger and debug agent no longer use the Thunk API.
Remove all deprecated functions and keep commented
references for future KFD tests.
Update and the keep the version checks for future use
and hsaKmtRuntimeEnable/Disable.
Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Signed-off-by: Laurent Morichetti <laurent.morichetti@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Change-Id: Ia2f10d82f5ac36d0bd1bda233810f26e8a154d55
Update hsaKmtCreateQueue to initialize the new save area header with the
exception payload and event ID.
Signed-by-off: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Reviewed-by: Sean Keely <sean.keely@amd.com>
Change-Id: Icd38062dc982cb29b30644699014eeb0b3e26d00
Added Support for launch kernel functions to fill_api_db
Added support for hipMemcpyToSymbol in add_memcpy
Added support for hsa_amd_memory_pool_allocate to be counted as source of allocations
Change-Id: I456a242ca1bc0c1bd39ae687a455b02c588de466
__fmm_release is sometimes called with the aperture lock, and sometimes
without. Consistently call it with the aperture lock held and remove the
lock/unlock calls from this function.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Change-Id: I80dddc64cc0703e5eed8e9f1eb65b75a2c7ae2eb
Unlock mutex if MMIO mapping fails. This happens on all GFXv8 GPUs.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Change-Id: I1dee1cbddefd9185c24ea79377f49f8ae2c5ff57
If the devices aren't peer-accessible, we shouldn't try to run a test
that requires that the devices be peer-accessible. Thus, add a check in
MapVramToGPUNodesTest to check for peer accessibility before executing
the peer mappings.
Signed-off-by: Kent Russell <kent.russell@amd.com>
Change-Id: Ib79b141f8c1ac6d85f5ab49d62af62ec10b988b7
Test Thunk multiple threads register and deregister same userptr race
condition, to emulate application register same userptr to multiple
GPUs using multiple threads.
Use thread barrier to sync the threads, to start register userptr at
same time.
Change-Id: I6723dc39f75908026fa14a490e39e1fe49a13a1b
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Limit test buffer size to 3/4 total VRAM size, and max 1GB.
Signed-off-by: Aaron Liu <aaron.liu@amd.com>
Change-Id: I937e10b0a6bd8215e3865b50f22ce75b3982a6f7
Use HIP_API_ID_NONE to detect unsupported API instead of
HIP_API_ID_NUMBER which can grow with a new version of the API.
This HIP_API_ID_NONE enum has a fixed value of 0 so the
HIP_API_IDs really start at FIRST.
Change-Id: I760aa50ddf6fa6d46bf20555ad7d429335a53f97
When building the json data flow, from_us_list has (timestamp, stream_id, thread_id).
stream_id used to be interpreted as from_tid and tid as to_tid. But that's not correct.
stream_id is always a destination and tid is the initiator (source).
Change-Id: I2f5bb86a387b4003b17271c90bdf9de4b59a79bf
Aperture locking is too fine-grained, it has race between find userptr
and allocate userptr object.
Change _fmm_allocate_device and fmm_allocate_memory_object to not take
the aperture lock, the callers take it, this implements an atomic find
userptr or allocate a new one.
Change-Id: I6773404e22c1f4382a211c5a9817df23c5534a2a
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Certain special signals do not carry their updates via their signal
value. These signals are wrappers around special KFD events, of
which the only current instance informs about VM faults. We either
need to check each signal for this special event type or rely on
the checking done in hsa_amd_signal_wait_any. Since there will always
be a small number of these signals it doesn't make much since to
penalize the performance path with this check. Additionally we know
that the signal indicated by hsa_amd_signal_wait_any is satisfied so
don't need to recheck it's conditions.
Change-Id: I9fc6298300ad543d823ecd28ca8fab4ad26c23ef
Clang now warns about set but unused variables. It also now
recognizes -Wno-error=unused-but-set-variable so this patch moves
that option back to the general options list.
Change-Id: Id800e87eb688b9441b14380e2246ad586179f31a
The rdcd uses another thread to listen the GPU events. That thread
runs in a tight loop which consume 100% CPU.
The fix will add a sleep to yield CPU.
Bug: SWDEV-291576
Change-Id: I7996720aab4a80346d79b1c73ee532d2abcd93cc
This is causing PSDB/OSDB failures so disable it until investigation is
done
Signed-off-by: Kent Russell <kent.russell@amd.com>
Change-Id: I666cd45fdf8ae585486adc7cf43eacd1700704bb
Allows determining if the host can directly access HMM memory that
is physically resident in vram.
Change-Id: Ie452eedd0e27fe1b511afd416f5a1cd01b3d84e8
To test ACCESS_IN_PLACE GPU mapping update to system memory.
Change-Id: I5b990215f39692e829128d848125e1ae0d571e03
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
CoherentHostAccess flag member moved from HSA_MEMORYPROPERTY
to HSA_CAPABILITY struct. Now this is reported to the
topology as a capability of the device instead of a device
memory property.
Change-Id: I48e43e4b4a0635b711b62933734587facdfbf88b
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
Enables the fragment allocator to handle >2MB allocations, maintaining
good TLB alignment. Prior code contained a bug that caused the effective
API granule for vram allocations >2MB to be bumped to 2MB.
Also adjusts the block cache's block retention heuristic to not
count discarded blocks as in use. This will reduce block retention
when a significant amount of large blocks or IPC is in use.
Change-Id: I30bd85eb87951df822211f799d9cfe579ab109c6
Marker events inside hcc_ops_trace.txt are from barriers so they are not meant to be stored in ops_patch_data map.
Added support for hipMemset events which are a kind of memory copy.
Change-Id: I213fe959bcd35ff0371613ba5bffd95bc53e06b5
RDC can optimize by bulk fetching multiple metrics using a single
rocm_smi call. However, currently this is not completely supported in
all ASIC generations. By default disable this for now.
Set environment variable RDC_BULK_FETCH_ENABLED=TRUE to enable
RDC bulk fetch.
BUG: SWDEV-289316
Change-Id: Ibb55514f198356dccf5f47bb0fd2d53c17acb251
recordid cannot be just a counter. The code removed was doing
just that i.e. incrementing a counter. Recordid has to come
from recvals data structure. That code was left there since
a while when Evgeny and Rachida were trying to prototype this feature.
I am not sure why it was not spotted before.
Change-Id: Ia867066dcfca083fcd4111f2aefc2fec88c26314
move blacklisted test case from gfx902 iommuv2 to dgpu path.
Signed-off-by: Yifan Zhang <yifan1.zhang@amd.com>
Change-Id: I8b101226ca8dcd0c12c484f5f6ce12fe73a75bdc
Signed-off-by: Yifan Zhang <yifan1.zhang@amd.com>
(cherry picked from commit 9cf4377572321396225950b9a58beb549120c2a3)
1st issue was that one of the ostream ops failed to print the
content of the struct.
2nd issue: get_ptr_type was called with args being src/dest
pointers while it should be the agents pointers for src/dest.
3rd issue: memcopies map used (recordid, procid, is_async)
as a key but this is not enough as some copies share same key,
so I added begin/end timestamps as a way to distinguish between them.
Change-Id: I7c6e80e74e30ea572f21612aaf0cf7efec6e91e6
Under high async handler load signal retention and event sorting
become bottlenecks. This change processes more handlers in a
single pass to amortize wait_any overheads.
Change-Id: I8b276e102db647e3858e120547aa0c6fca85ab4c
it is to optimize memory allocation latency, which
changes alignment from 2MB to 1GB.
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
Signed-off-by: Eric Huang <jinhuieric.huang@amd.com>
Change-Id: I7818e9f13b17e2c0992e75b17f978dc03a018a57
The API interface defines how the caller will use the API. An
example also shows how the API can be used.
It also defines the RdcDiagnostic module which can load the
library dynamically and then dispatch diagnostic test to run.
Change-Id: I1e041aab86f7e19338860f5ba65262977f4ea9cb
Jay Cornwall
Sean Keely
Jonathan Kim
Ammar ELWazir
Christophe Paquot
Felix Kuehling
Kent Russell
Philip Yang
Aaron Liu
changzhu
rachida
Bill(Shuzhou) Liu
AMD
Divya Shikre
Alex Sierra
Yifan Zhang