The rsmi_dev_temp_metric_get() can also support the HBM
temperatures which is retrieved from gpu_metrics.
Change-Id: I96b979296e90cf881523627b41b1a02849676416
Previously, RSMI assumed that the event counter values returned
from perf were only new events. But in fact, when we read the
counter values, they are running totals. To account for this, we
now record the value we read and take the difference between the
current value and the previously recorded value.
Change-Id: I1e04b514e89c7c4d4719889f2dae3a1283864e7f
Previously, during the rsmi_init discovery process, the existence
of an hwmon# directory was used to distinguish between gpus nodes
and non-gpu nodes. This isn't reliable in some scenarios. Instead,
the existence of the vbios_version file is used as an
indicator that the node is indeed a gpu.
Change-Id: Icfbe5c42ed0970077b05f25c3d209308a31bec85
The purpose of this patch is to implement GPU reset functionality
in the LIB, and to call it from the rocm_smi python CLI.
Signed-off-by: Ori Messinger <Ori.Messinger@amd.com>
Change-Id: Iaf525f7016f8354a7fd93af0209ca2e97ef4fd56
If we fail to find an expected temperature or voltage label
file, previously we were attempting to re-add a mapping of file
index to sensor types. Attempting to insert a map item that is already
present has no effect, so there should be no functional change.
This was a remnant of old code that should have been deleted.
Change-Id: Ie6f8a62f619a1ae58756e0fd891532434518cf78
The environment variable RSMI_DEBUG_INFINITE_LOOP is introduced
to facilitate debugging RSMI in user applications. When this
env. variable is non-zero, an infinite loop will be entered in
rsmi_init(). At this point, a debugger can be attached and RSMI
can be debugger. This only applies to debug builds.
Change-Id: I23f6dd730fc965764295070de053314a1cc5b6aa
Some rsmi apps fail without much explanation when
rsmi_init() fails. This patch hopes to provide some clues to
the reason for the failure.
Change-Id: Id51308dc327b9871d537dd3e709b677db4ef10bc
Previously, when a process holding a shared mutex was killed,
the next time an RSMI application was started, it would not be
able to obtain the mutex--the application would have to exit.
This fix uses pthread_mutexattr_setrobust() to detect this
situation and act accordingingly.
Also, add some missing, needed mutexes and move mutexes
closer to where the protect resource is used.
Change-Id: Icfdc3a246f4cfa3fd008e3f13472199abd76fd35
A special mode of operation to achieve minimal performance variation by letting
the user have the ability to provide the desired frequency to be set as the soft limit.
The user can control the entry and exit to the mode via rocm-smi a mechanism to
enter / exit performance determinism mode as below.
Enter performance determinism mode:
- hold a lock
- write performance_determinism to power_dpm_force_performance_level
- write input clk_freq to pp_dpm_sclk
- release lock
Exit performance determinism_mode:
- hold a lock
- write auto to power_dpm_force_performance_level
- release lock
Signed-off-by: Divya Shikre <DivyaUday.Shikre@amd.com>
Change-Id: Ia24e27954cdf1c4337ffc83d8948fbdfaf4552d2
WKS and server cards have an FRU with product information, so try to use
that for product name and product SKU if it exists.
Signed-off-by: Kent Russell <kent.russell@amd.com>
Change-Id: I40bbd3bf62f4cb02e96015ed1630112691cacbc3
VRAM and SDMA usage can be 64-bit long numbers. Use stoull()
instead of stoi() to convert the VRAM and SDMA usage strings to
numbers.
Change-Id: Ifadbada9f33320fc67666036ce8439823c1d1fb7
Event bitmask in KFD SMI event is now replaced with event index in
the SMI event message. Sending a event bitmask, which was a 64-bit
field with only 1 bit set, was quite wasteful of memory and also
potentially limiting to 64 events. Instead the kernel would send
event index in the SMI event message. As a result, update the
KFD SMI event handling to expect the event index in the message.
Change-Id: I3e74620788d3c1f7c0bdaa69e9d9ab3d1aba2c92
There may not be label files for some sensors on older
devices. We need to make sure there is a valid dummy
mapping in these cases.
Change-Id: Id6a8b71e554552be84a0e42a477070b504151e7f
Also:
* fix doxygen manual generation that was altered during
OAM refactor
* quiet some compile warnings.
Change-Id: I548a3cf00eb887bea3dbf58e362ca6dfe90bde28
Some systems have kfd sysfs properties entries that
are unreadable--for example, when a multi-gpu system is
dividing the gpus among containers, each container may
only be able to access certain gpus.
Previously, all kfd topology node properties entries were
assumed to be valid. Now, we check for readability before
declaring them "valid".
Fixes SWDEV-240169
Also:
* remove an assertion that would happen when no pcie
device identifier files are found on the system.
* fix cpplint issues
Change-Id: I74321b685159dd2628c890b33c39ad82988cb9dd
This corrects issues that arose after OAM reorganization.
It should address SWDEV-243294.
Also, fix some compile warnings that show up on RHEL.
Change-Id: Id14d444905da35cd7346bcfbcd82b6d0572708c4
A message is output in debug builds when pthread_unlock
returns an error. However, in most cases, it should return
EPERM. In fact, if it doesn't return EPERM, it is an
indication of a problem. This commit adjusts accordingly.
Change-Id: Ia5cad89aa6e68e79c1291ea21adffb0fa68f2300
Also, fix TestMutualExclusion and TestEvtNofifReadWrite.
Previously, some of the normal SetUp function was not
being done for this test. In some cases, no DRM
devices are being found on the test machine. Skip
those.
Change-Id: Iaa5a257841eb459aa57491ae9680c34a60d5ac2b
Also:
* When destroying a counter, make sure to stop the counter first
* In the test, do not stop (disable) the counter before
reading it.
* Clean up some whitespace in other tests
* Re-add manual pdf file
Change-Id: I0786ef3a994ca568299c77e44f092af8943ac33d
When pattern matching file names to determine API support, in
some environments std::regex will throw. This change is meant
to handle this more gracefully.
Change-Id: If1ccfe5bdd71ec4d08663c80692024488072e11b
This fixes a seg fault that would happen in release builds when
there are no KFD nodes on a system, which occurs when there are
not AMD gpus present in the system. This use case occurs
for higher application code that is meant to be gpu agnostic.
Change-Id: If374930bc2e62f9898f337349cde3ebb16091ff0
* Added a new test to verify mutual exclusion of access to device
resources
* Added some missing acquiring of mutexes to some RSMI calls, as
well as try-catch blocks.
Change-Id: I87aac009878a0b2d1f975e1d5b794d887bb23ff9
* Don't make different shared memory mutexes for different users
* Don't delete (unlink) the shared mutex file if the mutex
initialization fails. This may mess up other processes that
are using it. Instead, print a message on how to resolve the
situation, and then throw an error.
Note, this situation comes up when debug builds (usually)
either assert() or otherwise end execution without a proper
clean up.
* Remove cpplint from shared_mutex code
Change-Id: I5f8ca6150cac5c2405fb97007516da345093f966
Given a device index, return the corresponding NUMA node for the
device.
Also, add NUMA node tests to Sys Info Read test.
Change-Id: I0df4937470e6362e6737ccea568d4b3e5890c91a
This solution takes into account that some hwmons use
label files to map sensor types. The previous solution
did not take this into account.
Change-Id: I1d6204573cefa8197b2cfe0ffb412b545df3d80a
Divya Shikre
Bill(Shuzhou) Liu
Chris Freehill
Ori Messinger
Ramesh Errabolu
Kent Russell
Mukul Joshi
Mike Li