When hsa_amd_vmem_set_access is called, do not remove permissions for
unspecified agents. Also updating documentation in header to clarify
this.
Change-Id: I3bb4cf08ba399f85cc67b17fd13a4a40d862415f
Currently, KFDPerformanceTest.P2PBandWidthTest cannot work if there are
more than 16 KFD nodes in the system. This limit was put in to match the
number of SDMA queues supported on a single node.
This patch updates the test to make it run on systems with more than
16 KFD nodes.
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Change-Id: I561d0cdef664cae84fb9c13a801052e2001256e5
Socket server accept calls do not guarantee synchronous actions
post-accept. This can result in a race condition.
To resolve this, first limit the socket server's listen backlog to a
single connection. This will force competing clients to busy-retry
until timeout.
Second, make the DMABUF IPC file descriptor send-receive and import
calls into an atomic routine per connection.
By doing these fixes, not only to we resolve potential races but
we guarantee that any exporter process will create at most one
file descriptor that will only last for the duration of the import
transaction. This alleviates any concern on running into system
limits for the number of open file descriptors per process.
Change-Id: I6d8b14795a680d89a2707e082fa027d525792e05
Discarding blocks for reallocation on IPC export for better memory
performance trigger memory violations with DMA BUF exports so bypass
this for now as application performance drops haven't been observed
with the bypass.
The raw fragment should be passed to the DMA Buf export call as well
since offsets will be implicitly applied in the Thunk/KFD for
export/import calls.
Also, use the agent information directly from the pointer
information so that the export call doesn't have to scan memory to find
this. Pass the node ID in the handle so that the import call doesn't
have to make two thunk imports to fetch the node ID for GPU memory
imports.
Finally, allow the user to use DMA Buf IPC via
HSA_ENABLE_IPC_MODE_LEGACY=0 for developer testing as legacy mode will
be applied by default.
Change-Id: Ie8fe267f8768fa5df37126078406f7065f69ff4e
Fix some potentially unreleased memory, null value checks, files not
closed, and other such issues reported by codeql
Change-Id: Ia679aff97a773a642d8c8cbadeae30955554a62e
Signed-off-by: Kent Russell <kent.russell@amd.com>
In VM with 6vcpu, cpu schedule of
queue_delayed_work(system_freezable_wq) is lower than BM.
HSA_SMI_EVENT_QUEUE_RESTORE event from case HMMProfilingEvent/0 got
delayed execution and caused HMMProfilingEvent/1 fail.
The fix is only listen to HSA_SMI_EVENT_MIGRATE_START event and ignore
all other events.
Change-Id: I534e49b030bd4c534bc7a63eb431f4907659c8cd
Update commentary on HWS scheduler support bit for GPU debugging in
the HSA capabilities node properties field.
Change-Id: I59c519d74a528d5ecf5817ef94e75091314bd844
Fix data race by protecting events_page access with mutex in event create
Fix potential NULL dereference in hsaKmtWaitOnMultipleEvents_Ext
Fix unchecked return value in hsaKmtCreateEvent function
Change-Id: I434bef43666e5205a8b061259569c1d99a952752
We had skipped doing it for PAGE_SIZE, but it should be left as the
regular PAGE_SHIFT name, especially for users who are using different
headers. We want PAGE_SHIFT and PAGE_SIZE to be consistent with one
another, so set them both explicitly to the same value if either
of them is undefined
Change-Id: I121d81c48409dd77351b59a192d824e2419a2410
Signed-off-by: Kent Russell <kent.russell@amd.com>
When an entry is deleted from the array, it's set to nullptr
but not removed. Most other functions that
iterate over the array check if the entry is nullptr
but this loop in IterateExecutables did not.
Change-Id: I763b361eea59f6df201bb86ead0234e95f2cf79c
Return false if trying to free a NULL pointer (or invalid size)
internally in ROCr. This is to detect errors within ROCr when trying
to free NULL pointers. If a user of ROCr tries to free a NULL
pointer, this condition should be caught at the beginning of the
Runtime::FreeMemory(...) function and return HSA_STATUS_SUCCESS. This
matches the behavior of the free(...) or delete functions that
silently ignores calls when the passed a NULL pointer.
Change-Id: I84bc26928b35023e19cd9f214b42c6ee9508029c
Individual simple tests such as CPUAccessToGPUMemoryTest are taking
several hours on emulators as the total amount of VRAM keeps increasing.
Limit the pool sizes to 2GB, only on emulator.
Change-Id: I4b33e8549f89413da255731e6748f606ca64a663
Adds support for AllocateMemoryOnly inside XDNA driver.
Move the IsLocalMemory() check inside the KFD driver
since the XDNA driver can, and needs to, create handles
on system memory buffer objects.
Changed handle variable name from thunk_handle to user_mode_driver_handle,
which is more representative if we support non-GPU drivers.
Change-Id: I95db9d575afd1ab0ff2de74cea5175d9a12a721b
If BUILD_SHARED_LIBS doesn't get set at all, other projects importing
hsakmt may have an inconsistent state regarding SHARED vs STATIC.
Instead of setting the option and setting it as an option(), just detect
the variable and use it
Change-Id: I9d5a5fc6049ca5351f5e7c63d38ee9bfcb89bdad
Signed-off-by: Kent Russell <kent.russell@amd.com>
The fmm_node_[added|removed] functions were added in the initial FMM
support, but weren't used. Remove them now since no one's referencing
them
Change-Id: I1e46e57294a72012227b38f46c7099de0b9263be
Signed-off-by: Kent Russell <kent.russell@amd.com>
Adds support for the packet interface for interacting with
the Embedded Runtime (ERT) on AIE agents. The ERT is what
interprets command packets send to the AIE agent work
queues.
Change-Id: Id28fb98056b2c046354c446bdc9568d74385bea1
Adds support for initialzing the XDNA driver so that
a hardware context can be created for an AIE queue.
Right now this initializes the device heap in the driver,
gets the relevant tile parameters for the AIE agent,
and creates a hardware context that backs the AIE queue.
Change-Id: Ib90e1bc67a8637f6db3ff2bebe34677843796417
Ensure rocprofiler-register is linked and added to DEB and RPM package dependencies.
Github ticket - https://github.com/ROCm/ROCm/issues/3654
Change-Id: Iaaaca8bfa81ca33da147673ef1be798109b70aa5
Add a negative test to try to free the ring buffer of a queue and
confirm that a memory error is generated.
Change-Id: I4afd95c69c62f7c3e1138d5d6c4a5fd237631e43
GFX 9.4.x has better performance for CPU-GPU copies when using
engines in reverse order from other devices.
Change-Id: I1eaebf0e837bb7f44712f40d5115df618f6a73d7
If the KFD doesn't support targeting SDMA engines, ensure that ROCr
selects the correct downstream queue type by using an invalid engine.
Change-Id: Ia6848126f67f3d35ab37248633e8e0e6e2d77fff
To support fully-static library ROCm builds, ensure that all global
symbols are prefixed with something meaningful to avoid collisions with
other libraries
A script was made using" objdump -C -t" to get a list of symbols,
then checking if the global symbols have a meaningful prefix (for thunk:
hsakmt or kmt in various cases)
Change-Id: Ifd353f64a3344eb60d1f6c4e041aa20967b38a59
Signed-off-by: Kent Russell <kent.russell@amd.com>
- Use HSA_ALLOCATE_QUEUE_DEV_MEM=1 to create AQL queue in device
memory.
- Before writing AQL packet header to the queue use an SFENCE to ensure
that there is no reodering of the writes over PCIE
Change-Id: I5eacdc35108c4a1e245c75ae349b7495451aa60d
Fix for some places where the ISA buffers are not declared as
executable. Previous code in Thunk was blindly setting exec bit on all
memory allocations so this issue was masked.
Change-Id: Ic7a1169c69fb85ff9e8ea7bcc49a1845b37c08ff
The function can return NULL if it fails to create the backend, so check
for NULL before using it.
Change-Id: I4d6501bffd6dd0fc0d0f2224720f7d6dca1646f3
Signed-off-by: Kent Russell <kent.russell@amd.com>
Shweta Khatri
David Yat Sin
Mukul Joshi
Jonathan Kim
Kent Russell
Samuel Zhang
Longlong Yao
Young Hui
Yiannis Papadopoulos
James Xu
Sam Wu
Tony Gutierrez
Ramesh Errabolu
Ranjith Ramakrishnan
Chris Freehill
Wang, Yanyao
Saleel Kudchadker