The engines idle mask was increased to be an array of 4 u64 entries.
To make the print of this mask more readable, remove the "0x" prefix,
and zero-pad each u64 to 16 bytes if either it isn't zero or if any of
the higher-order u64's is not zero.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Failing to kill a user process during a hard reset can be due to a
reference to the user context which isn't released.
To make it easier to understand if this the reason for the failure and
not something else, add a print of the context refcount value.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Increase the size of the busy engines mask in 'struct hl_info_hw_idle',
for future ASICs with more than 128 engines.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Processes that use our device are likely to use at the same time other
devices such as remote storage.
In case our device is removed and a user process is still using the
device, we need to kill the user process. However, if that process
has a thread waiting for i/o to complete on remote storage, for example,
the process won't terminate.
Let's give it enough time to terminate before giving up.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
schedule_hard_reset can be true only if we didn't do hard-reset.
Therefore, no point of checking it in case hard_reset is true.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
If the device file is released while a context is still held, it won't
be possible to reopen it until the context is eventually released.
If that doesn't happen, only a device reset will revert it back to an
operational state, i.e. need to wait for a CS timeout or an error, or to
wait for an external intervention of injecting a reset via sysfs.
At this stage, after the device was released by user, context is held
either because of CS which were left running on the device and are not
relevant anymore, or due to missing cleanup steps from user side.
All of this is in any case handled in the device reset flow, so initiate
the reset at this point instead of waiting for it.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Each time razwi (read-only zero, write ignored) event happens, besides
capturing its data, also notify the user about it.
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Each time page fault happens, besides capturing its data, also notify
the user about it.
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Creating event queue workqueue using alloc_workqueue made it run in
multi threaded mode, which caused parallel dumping of events as well as
parallel events notifying to user, causing logs with multiple
events to be out of order.
Fixed by creating event queue workqueue as single threaded work queue.
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Each time razwi (read-only zero, write ignore) happens, besides
capturing its data, also notify the user about it.
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Add support for Gaudi2 Device with PCI revision 2.
Functionality is exactly the same as revision 1, the only difference
is device name exposed to user.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
hl_access_sram_dram_region() uses a region base which is set within the
hl_set_dram_bar() function. However, for SRAM access this function is
not called, and we end up with a wrong value of region base and with a
bad calculated address.
Fix it by initializing the region base value independently of whether
hl_set_dram_bar() is called or not.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Consecutive error protects a device reset loop from being triggered
due to h/w issues and enters the device into an unavailable state.
When user may cause the error, an unavailable state
will prevent the user from running its workloads.
The commit prevents entering consecutive state when a user context
is enabled.
Signed-off-by: Tal Cohen <talcohen@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Calling hl_device_reset() for a hard reset will lead to a quite
immediate device reset and to killing user process.
For resets that follow errors, it disables the option to debug the
errors on both the device side and the user application side.
This patch adds a 'graceful hard reset' option and a new
hl_device_cond_reset() function.
Under some conditions, mainly if there is no user process or if he is
not registered to driver notifications, this function will execute hard
reset as usual.
Otherwise, the reset will be postponed and a notification will be sent
to user, to let him perform post-error actions and then to release the
device, after which reset will take place.
If device is not released by user in some defined time, a watchdog work
will execute the reset in any case.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
As there are 2 types of user mappings, pmmu and hmmu, calculate
only the relevant mappings for the requested type.
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Up until now the use-case in the driver was that the HBM is accessed
using the HBM BAR, yet the BAR sometimes cannot cover the whole HBM and
so we needed to set the BAR to other HBM offset.
Now we are facing the need to access other PCI memory regions that can
be covered by the HBM BAR.
To answer that we are allowing the caller to determine if the HBM BAR
need to be set or not regardless of the PCI memory region.
Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
The code uses the pointer for trace purpose (without actually
dereference it) but still get static analysis warning.
This patch eliminate the warning.
Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
The event notifier mechanism should not raise an empty
event (event equals zero).
Signed-off-by: Tal Cohen <talcohen@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
'struct hl_device_reset_work' is used as a wrapper for the reset work
and its parameters, including the reset workqueue on which it runs.
In a future commit, another reset related work with similar parameters
is going to be added, but it won't use the reset workqueue.
As in any case there is a single reset workqueue, and to allow the resue
of this structure, move the reset workqueue to 'struct hl_device'.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
If reset upon device release is enabled, there is no need to check the
device idle status in hpriv_release(), because device is going to be
reset in any case.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
pf was an abbreviation for prefetch but because pf already stands
for 'physical function', we decided to change it to 'prefetch'.
Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Only the first page fault will be saved.
Besides the address which caused the page fault, the driver captures
all of the mmu user mappings.
User can retrieve this data via the new uapi (new opcode in INFO ioctl).
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
This event notification was compatible only with gaudi, where razwi
and page fault happens together.
To make it compatible with all ASICs, this refactor contains:
1. Razwi notification will only notify about razwi info.
New notification will be added in future patch, to retrieve data
about page fault error.
2. Changed razwi info structure to support all ASICs.
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Monitoring apps would like to query device state at any time so we
should allow it also during reset because it doesn't involve
accessing the h/w.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
When doing sizeof() and giving as argument a dereference of
a pointer-to-a-pointer object, clang will issue a warning.
Eliminate the warning by passing struct <name>*
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Current code does not takes into account the new DRAM region base
and so calculated address is wrong and can lead to crush.
Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
F/W security status might change after every reset.
Add the reading of the preboot status to the hard reset sequence, which
among others reads this security indication.
As this preboot status reading includes the waiting for the preboot to
be ready, it can be removed from the CPU init which is done in a later
stage.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
'Device activity open packet' should be sent outside of mutex as
there is no real necessity for a lock.
In addition 'device activity close packet' should be sent upon an
actual release of the device.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
As part of the RAS that is done by the f/w, we should send a message
to the f/w when a user either acquires or releases the device.
Signed-off-by: farah kassabri <fkassabri@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
This patch add tracepoints in the code for DMA allocation.
The main purpose is to be able to cross data with the map operations and
determine whether memory violation occurred, for example free DMA
allocation before unmapping it from device memory.
To achieve this the DMA alloc/free code flows were refactored so that a
single DMA tracepoint will catch many flows.
To get better understanding of what happened in the DMA allocations
the real allocating function is added to the trace as well.
Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
The calculation of the device DRAM base address before setting the
relevant PCIe BAR to point at it, has an assumption that this BAR is
used to access only the DRAM, and thus the covered DRAM size is a power
of 2.
In future ASICs it is not necessarily true, so need to update the
calculation to support also a non-power-of-2 size.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Instead of recalculating the cdev index, store it in a dedicated data
member. This data member is intended to be passed to other drivers using
the auxiliary bus infra and hence this new data member is necessary in
case that the calculation is changed in the future.
Signed-off-by: Omer Shpigelman <oshpigelman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
A common function that is called from multiple places can't be
located in degugfs.c because that file is only compiled if
debugfs is enabled in the kernel config file.
This can lead to undefined symbol compilation error.
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Add a missing lock in hl_device_resume() when it assigns a value to the
'in_reset' indication.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
In order to be more explicit we should use the term compute_reset
for describing the reset in which only the compute engines gets
reset.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Doing compute reset can be the traditional inference soft reset
that is supported only in Goya.
Or it can be the new reset upon device release, which is supported
in Gaudi2 and above.
Therefore, wherever suitable, use the terminology of compute reset
instead of soft reset.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
The user might want to know the device is in reset after device
release, which is not an erroneous event as a regular reset.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
reset_info.is_in_soft_reset should be updated both before in_reset
and inside the spin lock of the reset info structure.
The reasons are:
- When we are inside soft reset, it implies we are in reset. Therefore,
if someone checks if we are in soft reset, he can deduce we are
in reset, while the opposite is not correct and might be misleading.
- Both these flags are changed together so they must be changed
inside the reset info spinlock.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
For gaudi2 we need to send a value to F/W as part of the
PCI_ACCESS packet.
As a preparation, modify hl_fw_send_pci_access_msg() to have a 'value'
field.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Max power API is not supported in secured devices. Hence, we should
skip setting it during boot.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Enable the Gaudi2 ASIC code in the pci probe callback of the driver so
the driver will handle Gaudi2 ASICs.
Add the PCI ID to the PCI table and add the ASIC enum value to all
relevant places.
Fixup the device parameters initialization for Gaudi2.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
In Gaudi2 we moved to a different wait for command submission
completion model. Instead of receiving interrupt only on external
queues, we use the device's sync manager to notify us when the
entire command submission finishes.
This enables us to remove the categorization of queues to external
and internal, and treat each queue equally, without the need to parse
and patch any command buffer.
This change also requires refactoring to the IRQ handling of
CS completions.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
There are a couple of device variables that are used for testing
purposes and they are set to fixed values.
Remove the variables that are not relevant anymore and document the
remaining variables.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Add the ASIC-specific code for Gaudi2. Supply (almost) all of the
function callbacks that the driver's common code need to initialize,
finalize and submit workloads to the Gaudi2 ASIC.
It also contains the code to initialize the F/W of the Gaudi2 ASIC
and to receive events from the F/W.
It contains new debugfs entry to dump razwi events. razwi is a case
where the device's engines create a transaction that reaches an
invalid destination.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Region structure is derived from region type, hence no need to pass
it as an argument.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
This is a pre-requisite patch for adding tracepoints to the DMA memory
operations (allocation/free) in the driver.
The main purpose is to be able to cross data with the map operations and
determine whether memory violation occurred, for example free DMA
allocation before unmapping it from device memory.
To achieve this the DMA alloc/free code flows were refactored so that a
single DMA tracepoint will catch many flows.
Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
