To avoid checking if fence exists multipled times, changed fence
handling to depend only on the fence status field:
Busy, which means CS still did not completed :
Add its QID so multi CS wait on its completion.
Finished, which means CS completed and fence exists:
Raise its completion bit if it finished mcs handling and
update if necessary the earliest timestamp.
Gone, which means CS already completed and fence deleted:
Update multi CS data to ignore timestamp and raise its
completion bit.
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
No need to check the return value if the following action is the same for
both cases. In addition, now that hl_ctx_free() doesn't print if the
context is not released, its name can be misleading as the context might
stay alive after it is executed with no indication for that.
Hence we can discard it and simply put the refcount.
Signed-off-by: Omer Shpigelman <oshpigelman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Remove the flag that determines whether to take a timestamp once the
interrupt arrives.
Instead, always take the timestamp once per interrupt.
This is a must for the user-space to measure its graph operations
to evaluate the graph computation time.
Signed-off-by: Yuri Nudelman <ynudelman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
When adding a new node to the hpriv list, the driver should
initialize its fields before adding the new node.
Otherwise, there may be some small chance of another thread traversing
that list and accessing the new node's fields without them being
initialized.
Signed-off-by: Moti Haimovski <mhaimovski@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Make the frequency set/get functionality common to all ASICs.
This makes more code reusable when adding support for newer ASICs.
Signed-off-by: Rajaravi Krishna Katta <rkatta@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Fix the following build/link error by adding a dependency on the CRC32
routines:
ld: drivers/misc/habanalabs/common/firmware_if.o: in function `hl_fw_dynamic_request_descriptor':
firmware_if.c:(.text.unlikely+0xc89): undefined reference to `crc32_le'
Fixes: 8a43c83fec ("habanalabs: load boot fit to device")
Signed-off-by: Vegard Nossum <vegard.nossum@oracle.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Implement the calls to the dma-buf kernel api to create a dma-buf
object backed by FD.
We block the option to mmap the DMA-BUF object because we don't support
DIRECT_IO and implicit P2P. We only implement support for explicit P2P
through importing the FD of the DMA-BUF.
In the export phase, we provide to the DMA-BUF object an array of pages
that represent the device's memory area. During the map callback,
we convert the array of pages into an SGT. We split/merge the pages
according to the dma max segment size of the importer.
To get the DMA address of the PCI bar, we use the dma_map_resources()
kernel API, because our device memory is not backed by page struct
and this API doesn't need page struct to map the physical address to
a DMA address.
We set the orig_nents member of the SGT to be 0, to indicate to other
drivers that we don't support CPU mappings.
Note that in Habanalabs's ASICs, the device memory is pinned and
immutable. Therefore, there is no need for dynamic mappings and pinning
callbacks.
Also note that in GAUDI we don't have an MMU towards the device memory
and the user works on physical addresses. Therefore, the user doesn't
pass through the kernel driver to allocate memory there. As a result,
only for GAUDI we receive from the user a device memory physical address
(instead of a handle) and a size.
We check the p2p distance using pci_p2pdma_distance_many() and refusing
to map dmabuf in case the distance doesn't allow p2p.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Gal Pressman <galpress@amazon.com>
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
When polling fences for multi CS, it is possible that fence is
no longer exists (its corresponding CS completed and the fence was
deleted) but we still accessing its parameters, causing NULL pointer
dereference.
Fixed by checking if fence exits before accessing its parameters.
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Race condition occurs when CS fence completes and multi CS did not
completed yet, while waiting for multi CS ends and returns indication
to user that the CS completed. Next wait for multi CS may be triggered
by previous multi CS completion without any current CS completed,
causing an error.
Example scenario :
1. User do multi CS wait for CSs 1 and 2 on master QID 0
2. CS 1 and 2 reached the "cs release" code. The thread of CS 1
completed both the CS and multi CS handling but the completion
thread of CS 2 completed the CS but still did not executed
complete_multi_cs (note that in CS completion the sequence is to
first do complete all for the CS and then another complete all to
signal the multi_cs)
3. User received indication that CS 1 and 2 completed (since we check
the CS fence and both indicated as completed) and immediately waits
on CS 3 and 4, also on master QID 0.
4. Completion thread of CS2 executed complete_multi_cs before
completion of CS 3 and 4 and so will trigger the multi CS wait of
CSs 3 and 4 as they wait on master QID 0.
This will trigger multi CS completion although none of its
current CS has been completed.
Fixed by adding multi CS complete handling indication for each CS.
CS will be marked to the user as completed only if its fence completed
and multi CS handling is done.
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
There may be a situation where drivers receives continuous fatal H/W
error events from FW immediately post reset cycle.
This may be due to some fault on the silicon itself.
In such case its better to bypass reset cycle so we won't be stuck in
endless loop of resets.
This commit bypasses reset request in case driver received two back to
back FW fatal error before first occurrence of heartbeat event.
Signed-off-by: Bharat Jauhari <bjauhari@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Taking an accurate timestamp in a close proximity of the interrupt is
required for user side statistics management.
Signed-off-by: Yuri Nudelman <ynudelman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
The driver allows only a single process to open a device's FD at any
single time. This is done by checking "hdev->compute_ctx" under mutex.
Therefore, to prevent a race between the moment a user closes it's FD
and when another user tries to open the device, we need to make sure
that clearing this variable is the very last thing that is done in the
code of the FD's release.
I'm moving the idle check before clearing this variable and the
"reset on device release". btw, if the reset happens it will prevent
any other user from opening the device until the reset is finished.
An important thing to note is that we need to remove the user process
that is closing the device from the process list BEFORE calling the
reset function. That is to prevent a case where the reset code will
try to kill that user process and it is unnecessary as the process
doesn't hold any device/driver resources anymore.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reset to the device is not necessarily due to an error, so print it
as info instead of error.
In addition, print the type of reset we are doing:
- reset of the entire device (aka hard reset)
- reset of the device after user have released it (less than hard reset)
- lighter reset of an inference device (aka soft reset)
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Soft-reset is the procedure where we reset only the compute/DMA engines
of the device, without requiring the current user-space process to
release the device.
This type of reset can happen if TDR event occurred (a workload got
stuck) or by a root request through sysfs.
This is only relevant for inference ASICs, as there is no real-world
use-case to do that in training, because training runs on multiple
devices.
In addition, we also do (in certain ASICs) a reset upon device release.
That reset uses the same code as the soft-reset.
Therefore, to better differentiate between the two resets, it is better
to rename the soft-reset support as "inference soft-reset", to make
the code more self-explanatory.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
The translation in debugfs of device memory MMU virtual addresses was
wrong as it did not take into consideration the fact that the page
sizes there can be not power of 2.
Signed-off-by: Yuri Nudelman <ynudelman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
In order to avoid user target value wraparound, we modify the
current interface so user will be able to wait for an 8-byte
target value rather than a 4-byte value.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
During TDR handling, we check multiple times if CS is valid.
No need to perform this check as CS must be valid at all time
during the TDR handling.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Add support to retrieve following power info via HWMON:
- instantaneous power value
- highest value since last reset
- reset the highest place holder
Signed-off-by: Rajaravi Krishna Katta <rkatta@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Instead of having dedicated function per message that we want to send
to the firmware in COMMS protocol, have a generic function that we can
call to from other parts of the driver
Signed-off-by: Alon Mizrahi <amizrahi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Instead of using the Linux kernel HWMON enums definition when
communicating with the firmware, use proprietary HWMON based enums
i.e. map hwmon.h header enum to cpucp_if.h based enum while.
This is needed because the HWMON enums are not forcing backward
compatibility and therefore changes can break compatibility between
newer driver and older firmware.
The driver will check for CPU_BOOT_DEV_STS0_MAP_HWMON_EN bit to
validate if f/w supports cpucp->hwmon enum mapping to support older
firmware where this mapping won't be available.
Signed-off-by: Rajaravi Krishna Katta <rkatta@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Command submission timeout is currently determined during driver
loading time. As some environments requires this timeout to be
modified in runtime, we introduce a new debugfs node that controls
the timeout value without the need to reload the driver.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
In wait for CS IOCTL code, the driver resets the incoming args structure
before returning to the user, regardless of the return value of the
IOCTL.
In case the IOCTL returns EINTR, resetting the args will result in error
in case the userspace will repeat the ioctl call immediately (which is
the behavior in the hl-thunk userspace library).
The solution is to reset the args only if the driver returns success (0)
as a return value for the IOCTL.
Signed-off-by: Rajaravi Krishna Katta <rkatta@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Staged submission consists of multiple command submissions.
In order to be explicit, driver should return a single cs sequence
for every cs in the submission, or else user may try to wait on
an internal CS rather than waiting for the whole submission.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Add handling for case where the user doesn't set wait offset,
and keeps it as 0. In such a case the driver will decrement one
from this zero value which will cause the code to wait for
wrong number of signals.
The solution is to treat this case as in legacy wait cs,
and wait for the next signal.
Signed-off-by: farah kassabri <fkassabri@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
As user can send wrong arguments to multi CS API, we rate limit
the amount of errors dumped to dmesg, in addition we change the
severity to warning.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Couple of fixes to the LBW RR configuration:
1. Add missing configuration of the SM RR registers in the DMA_IF.
2. Remove HBW range that doesn't belong.
3. Add entire gap + DBG area, from end of TPC7 to end of entire
DBG space.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
As collective wait operation is required only when NIC ports are
available, we disable the option to submit a CS in case all the ports
are disabled, which is the current situation in the upstream driver.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Due to FLR scenario when running inside a VM, we must not use indirect
MSI because it might cause some issues on VM destroy.
In a VM we use single MSI mode in contrary to multi MSI mode which is
used in bare-metal.
Hence direct MSI should be used in single MSI mode only.
Signed-off-by: Omer Shpigelman <oshpigelman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
In case of single staged cs with both first/last indications
set, we reach a scenario where in cs_release function flow
we don't cancel the TDR work before freeing the cs memory,
this lead to kernel OOPs since when the timer expires
the work pointer will be freed already.
In addition treat wait encaps cs "not found" handle
as "OK" for the user in order to keep the user interface
for both legacy and encpas signal/wait features the same.
Signed-off-by: farah kassabri <fkassabri@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
We have a potential race where a user interrupt can be received
in between user thread value comparison and before request was
added to wait list. This means that if no consecutive interrupt
will be received, user thread will timeout and fail.
The solution is to add the request to wait list before we
perform the comparison.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
When the f/w runs in secured mode, it can reset the ASIC when certain
events occur. In unsecured mode, the driver asks the f/w to reset the
ASIC for those events.
We need to perform the entire reset procedure but without accessing the
ASIC. i.e. without halting the engines and without sending messages
to the f/w.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
copy_from_user might sleep so we can never call it when we have
a spinlock.
Moreover, it is not necessary in waiting for user interrupt, because
if multiple threads will call this function on the same interrupt,
each one will have it's own fence object inside the driver. The
user address might be the same, but it doesn't really matter to us,
as we only read from it.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Checking if the device is operational when entering the function to
wait for user interrupt is not something that is useful or necessary.
It is not done in any other wait_for_cs ioctl path.
If the device becomes non-operational during the wait, the reset
function will make sure the process wait is interrupted.
Instead, move the check to the beginning of hl_wait_ioctl(). It will
block any attempt to wait on CS or user interrupt once the device
is already marked as non-operational.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Because this spinlock is taken in an interrupt handler, we must use
the spin_lock_irqsave/irqrestore version to disable the interrupts
on the local CPU. Otherwise, we can have a potential deadlock (if
the interrupt handler is scheduled to run on the same cpu that the
code who took the lock was running on).
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
On init, the disabled state is cleared right before hw_init and that
causes the device to report on "Operational" state before the device
initialization is finished. Although the char device is not yet exposed
to the user at this stage, the sysfs entries are exposed.
This can cause errors in monitoring applications that use the sysfs
entries.
In order to avoid this, a new state "in device creation" is introduced
to ne reported when the device is not disabled but is still in init
flow.
Signed-off-by: Omer Shpigelman <oshpigelman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
This must be done to clear the internal mem cache so we won't get
ecc errors on the first invalidation.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
In secured mode, the CGM is disabled. Therefore, the DC power is
higher. Without taking it into consideration, the utilization is
12-15% at idle.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
The out of bounds SLM access TPC interrupt indicates a severe compiler
bug and needs to be informed to user.
This interrupt is currently masked so unmask it.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
It is useful to have the ability to see which user address was pinned
to which physical address during the initial mapping. We already have
all that info stored, but no means to search this data (which may be
quite large).
Signed-off-by: Yuri Nudelman <ynudelman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
In case F/W security is enabled driver cannot access ECC registers,
hence driver must fetch the ECC info from F/W.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
During the integration, the multi-CS requirements were refined:
- The multi CS call shall wait on "per-ASIC" predefined stream masters
instead of set of streams.
- Stream masters are set of QIDs used by the upper SW layers (synapse)
for completion (must be an external/HW queue).
Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Current "state dump" is lacking of monitored SOB IDs. Add for
convenience.
Signed-off-by: Alon Mizrahi <amizrahi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
