This patch fixes the SDMA queue initialization, when running in non-HWS mode.
The first fix is to move the initialization of SDMA VM parameters before the
initialization of the SDMA MQD.
The second fix is to load the MQD to an HQD after the initialization of the MQD.
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This patch adds a missing destruction of mqd, when destroying a kernel queue.
Without the destruction, there is a memory leakage when repeatedly creating and
destroying kernel queues.
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
get_pipes_num() calls BUG_ON so we can't set it as inline because it produces a
warning as BUG_ON() uses static variables when it is expanded.
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
This patch fixes a bug in the initialization of the pipelines. The
init_pipelines() function was called with a constant value of 0 in the
first_pipe argument. This is an error because amdkfd doesn't handle pipe 0.
The correct way is to pass the value that get_first_pipe() returns as the
argument for first_pipe.
This bug appeared in 3.19 (first version with amdkfd) and it causes around 15%
drop in CPU performance of Kaveri (A10-7850).
v2: Don't set get_first_pipe() as inline because it calls BUG_ON()
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Cc: stable@vger.kernel.org
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Pull drm updates from Dave Airlie:
"This is the main drm pull, it has a shared branch with some alsa
crossover but everything should be acked by relevant people.
New drivers:
- ATMEL HLCDC driver
- designware HDMI core support (used in multiple SoCs).
core:
- lots more atomic modesetting work, properties and atomic ioctl
(hidden under option)
- bridge rework allows support for Samsung exynos chromebooks to
work finally.
- some more panels supported
i915:
- atomic plane update support
- DSI uses shared DSI infrastructure
- Skylake basic support is all merged now
- component framework used for i915/snd-hda interactions
- write-combine cpu memory mappings
- engine init code refactored
- full ppgtt enabled where execlists are enabled.
- cherryview rps/gpu turbo and pipe CRC support.
radeon:
- indirect draw support for evergreen/cayman
- SMC and manual fan control for SI/CI
- Displayport audio support
amdkfd:
- SDMA usermode queue support
- replace suballocator usage with more suitable one
- rework for allowing interfacing to more than radeon
nouveau:
- major renaming in prep for later splitting work
- merge arm platform driver into nouveau
- GK20A reclocking support
msm:
- conversion to atomic modesetting
- YUV support for mdp4/5
- eDP support
- hw cursor for mdp5
tegra:
- conversion to atomic modesetting
- better suspend/resume support for child devices
rcar-du:
- interlaced support
imx:
- move to using dw_hdmi shared support
- mode_fixup support
sti:
- DVO support
- HDMI infoframe support
exynos:
- refactoring and cleanup, removed lots of internal unnecessary
abstraction
- exynos7 DECON display controller support
Along with the usual bunch of fixes, cleanups etc"
* 'drm-next' of git://people.freedesktop.org/~airlied/linux: (724 commits)
drm/radeon: fix voltage setup on hawaii
drm/radeon/dp: Set EDP_CONFIGURATION_SET for bridge chips if necessary
drm/radeon: only enable kv/kb dpm interrupts once v3
drm/radeon: workaround for CP HW bug on CIK
drm/radeon: Don't try to enable write-combining without PAT
drm/radeon: use 0-255 rather than 0-100 for pwm fan range
drm/i915: Clamp efficient frequency to valid range
drm/i915: Really ignore long HPD pulses on eDP
drm/exynos: Add DECON driver
drm/i915: Correct the base value while updating LP_OUTPUT_HOLD in MIPI_PORT_CTRL
drm/i915: Insert a command barrier on BLT/BSD cache flushes
drm/i915: Drop vblank wait from intel_dp_link_down
drm/exynos: fix NULL pointer reference
drm/exynos: remove exynos_plane_dpms
drm/exynos: remove mode property of exynos crtc
drm/exynos: Remove exynos_plane_dpms() call with no effect
drm/i915: Squelch overzealous uncore reset WARN_ON
drm/i915: Take runtime pm reference on hangcheck_info
drm/i915: Correct the IOSF Dev_FN field for IOSF transfers
drm/exynos: fix DMA_ATTR_NO_KERNEL_MAPPING usage
...
This patch changes a BUG_ON() statement to pr_debug, in case the user tries to
update a non-existing queue.
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Ben Goz <ben.goz@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This patch fixes a bug where the first_pipe index passed into init_pipelines()
was a #define instead of the value that is passed into amdkfd by radeon
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
This patch fixes a bug when calling to init_pipeline() interface.
The index that was passed to that function didn't take into account the
first_pipe value, which represents the first pipe index that is under amdkfd's
responsibility.
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
CP microcode uses undocumented bits in this register to record queue
state information. The KFD zeroes these bits in update_mqd, when invoked
through the UPDATE_QUEUE ioctl, causing incoherent state when the ioctl
is used to successively unmap and map a queue.
Since the queue type cannot be changed in this path, move the MQD write
to init_mqd.
Signed-off-by: Jay Cornwall <jay.cornwall@amd.com>
Reviewed-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
dqm->queue_count tracks queues in the active state only. In a few
places this count is modified unconditionally, leading to an incorrect
value when the UPDATE_QUEUE ioctl is used to make a queue inactive.
Signed-off-by: Jay Cornwall <jay.cornwall@amd.com>
Reviewed-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
This backmerges drm-fixes into drm-next mainly for the amdkfd
stuff, I'm not 100% confident, but it builds and the amdkfd
folks can fix anything up.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Conflicts:
drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
This patch handles a case where amdkfd tries to destroy a queue but the queue
type is invalid.
This case occurs in non-HWS path.
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Backmerge Linus tree after rc5 + drm-fixes went in.
There were a few amdkfd conflicts I wanted to avoid,
and Ben requested this for nouveau also.
Conflicts:
drivers/gpu/drm/amd/amdkfd/Makefile
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
drivers/gpu/drm/amd/amdkfd/kfd_priv.h
drivers/gpu/drm/amd/include/kgd_kfd_interface.h
drivers/gpu/drm/i915/intel_runtime_pm.c
drivers/gpu/drm/radeon/radeon_kfd.c
This patch replaces the two current amdkfd module parameters with a new one.
The current parameters that are being replaced are:
- Maximum number of HSA processes
- Maximum number of queues per process
The new parameter that replaces them is called "Maximum queues per device"
This replacement achieves two goals:
- Allows the user to have as many HSA processes as it wants (until
a maximum of 512 HSA processes in Kaveri).
- Removes the limitation the user had on maximum number of queues per HSA
process. E.g. the user can now have processes which only have one queue and
other processes which have hundreds of queues, while before the user
couldn't have more than 128 queues per process (as default).
The default value of the new parameter is 4096 (32 * 128, which were the
defaults of the old parameters). There is almost no additional GART memory
required for the default case. As a reminder, this amount of queues requires a
little bit below 4MB of GART memory.
v2:
In addition, This patch defines a new counter for queues accounting in the DQM
structure. This is done because the current counter only counts active queues
which allows the user to create more queues than the
max_num_of_queues_per_device module parameter allows.
However, we need the current counter for the runlist packet build process, so
the solution is to have a dedicated counter for this accounting.
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Ben Goz <ben.goz@amd.com>
If the first queue created was failed on DQM then PQM should
unregister the process from DQM.
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
This patch completely removes the sync_with_hw() because it was broken and
actually there is no point of using it.
This function was used to:
- Make sure that the submitted packet to the HIQ (which is a kernel queue) was
read by the CP. However, it was discovered that the method this function used
to do that (checking wptr == rptr) is not consistent with how the actual CP
firmware works in all cases.
- Make sure that the queue is empty before issuing the next packet. To achieve
that, the function blocked amdkfd from continuing until the recently
submitted packet was consumed. However, the acquire_packet_buffer() already
checks if there is enough room for a new packet so calling sync_with_hw() is
redundant.
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
In order not to occupy the current core and thus prevent the core from
servicing IOMMU PPR requests, this patch replaces the call in DQM to
cpu_relax() with a call to schedule().
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
This patch fixes a minor bug in allocate_hqd(), where the loop run from the
next-to-allocate pipe until the number of pipes.
This is wrong because we need to consider the possibility where
next-to-allocate pipe is not 0, and thus, the for-loop only checks part of the
pipes and doesn't wrap-around, as it supposed to do.
Therefore, we add another counting variable to make sure we go over all the
pipes, regardless of where we start to look at the first iteration of the loop.
This bug only affected non-HWS mode. In HWS mode, the CP fw is responsible for
allocating the HQD.
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
The work queue couldn't reliably prevent the SW ring buffer from
overflowing, so dmesg was spammed by
kfd kfd: Interrupt ring overflow, dropping interrupt.
messages when running e.g. the Atlantis Substance demo from
https://wiki.unrealengine.com/Linux_Demos on Kaveri.
Since the SW ring buffer doesn't actually do anything at this point, just
remove it for now. When actual interrupt processing code is added to
amdkfd, it should try to do things immediately and only defer to work
queues when necessary.
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
This patch changes kfd_ioctl() to be very similar to drm_ioctl().
The patch defines an array of amdkfd_ioctls, which maps IOCTL definition to the
ioctl function.
The kfd_ioctl() uses that mapping to call the appropriate ioctl function,
through a function pointer.
This patch also declares a new typedef for the ioctl function pointer.
v2: Renamed KFD_COMMAND_(START|END) to AMDKFD_...
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
This patch reformats the ioctl definitions in kfd_ioctl.h to be similar to the
drm ioctls definition style.
v2: Renamed KFD_COMMAND_(START|END) to AMDKFD_...
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
This patch moves the copy_to_user() and copy_from_user() calls from the
different ioctl functions in amdkfd to the general kfd_ioctl() function, as
this is a common code for all ioctls.
This was done according to example taken from drm_ioctl.c
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
This patch fixes a bug where deallocate_vmid() didn't actually unmap the
VMID<-->PASID mapping (in the registers).
That can cause undefined behavior.
This bug only occurs in non-HWS mode.
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
This patch fixes a bug in DQM, where the MQD of a newly created compute queue
is not loaded to an HQD slot. As a result, the CP never reads packets from this
queue.
This bug happens only in non-HWS (hardware scheduling) mode. In HWS mode, the
CP is responsible of loading MQDs to HQDs slots.
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Commit "amdkfd: use sizeof(long) granularity for the pasid bitmask" calculated
the number of longs it will need, but ended up allocating that number of
bytes rather than longs.
Fix that silly error and allocate the amount of data really required.
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Original code sent always 0 as the index number of the node. This patch fixes
this bug by sending a variable which is incremented per node.
Signed-off-by: Ben Goz <ben.goz@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
This patch fixes a device QCM bug, where the number of queues were not
counted correctly for the operation of update queue. The count was incorrect
as there was no regard to the previous state of the queue.
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
This patch starts to add support for the VI APU in the KQ (kernel queue)
module.
Because most (more than 90%) of the KQ code is shared among AMD's APUs, we
chose a design that performs most/all the code in the shared KQ file
(kfd_kernel_queue.c). If there is H/W specific code to be executed,
than it is written in an asic-specific extension function for that H/W.
That asic-specific extension function is called from the shared function at the
appropriate time. This requires that for every asic-specific extension function
that is implemented in a specific ASIC, there will be an equivalent
implementation in ALL ASICs, even if those implementations are just stubs.
That way we achieve:
- Maintainability: by having one copy of most of the code, we only need to
fix bugs at one locations
- Readability: very clear what is the shared code and what is done per ASIC
- Extensibility: very easy to add new H/W specific files/functions
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This patch does some re-org on the kernel_queue structure. It takes out
all the function pointers from the structure and puts them in a new structure,
called kernel_queue_ops. Then, it puts an instance of that structure
inside kernel_queue.
This re-org is done to prepare the KQ module to support more than one AMD APU
(Kaveri).
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This patch starts to add support for the VI APU in the DQM module.
Because most (more than 90%) of the DQM code is shared among AMD's APUs, we
chose a design that performs most/all the code in the shared DQM file
(kfd_device_queue_manager.c). If there is H/W specific code to be executed,
than it is written in an asic-specific extension function for that H/W.
That asic-specific extension function is called from the shared function at the
appropriate time. This requires that for every asic-specific extension function
that is implemented in a specific ASIC, there will be an equivalent
implementation in ALL ASICs, even if those implementations are just stubs.
That way we achieve:
- Maintainability: by having one copy of most of the code, we only need to
fix bugs at one locations
- Readability: very clear what is the shared code and what is done per ASIC
- Extensibility: very easy to add new H/W specific files/functions
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This patch does some re-org on the device_queue_manager structure. It takes out
all the function pointers from the structure and puts them in a new structure,
called device_queue_manager_ops. Then, it puts an instance of that structure
inside device_queue_manager.
This re-org is done to prepare the DQM module to support more than one AMD APU
(Kaveri).
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Instead of creating a BUG if trying to free a NULL GART sub-allocation object,
just return 0 (success).
This is done to mirror behavior of kfree.
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Since the user space may call open() more that once from the same process,
the aperture initialization should be moved from kfd_open()
Signed-off-by: Alexey Skidanov <Alexey.Skidanov@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
This patch displays the firmware version of the microcode that is currently
running in the MEC.
This is needed for the HSA RT, so it could differentiate its behavior based on
fw version. e.g. workarounds for bugs in fw
v2: Send the KGD_ENGINE_MEC1 as a parameter to the get_fw_version()
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This patch adds a new interface to the kfd-->kgd interface.
The new interface function retrieves the firmware version that is currently in
use by the MEC engine. The firmware was uploaded to the MEC engine by the kgd
(radeon).
v2: Added parameter of engine type to interface function
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This patch checks if the process that opens the /dev/kfd device is 32-bit
process. If so, it returns -EPERM and prints a warning message in dmesg.
This is done to prevent 32-bit user processes from using amdkfd, and hence, HSA
features.
AMD's HSA userspace stack will also support only 64-bit processes on Linux.
Reviewed-by: Alexey Skidanov <alexey.skidanov@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
In function acquire_packet_buffer() we may return -ENOMEM. In that case, we
should set the *buffer_ptr to NULL, so that calling functions which check the
*buffer_ptr value as a criteria for success, will know that
acquire_packet_buffer() failed.
Reviewed-by: Alexey Skidanov <alexey.skidanov@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
srcu callbacks are running in atomic context, we can't allocate using
__GFP_WAIT.
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
All the bit operations (such as find_first_zero_bit()) read sizeof(long) bytes
at a time. If we allocated less than sizeof(long) bytes for the bitmask we
would be accessing invalid memory when working with the bitmask.
Change the allocator to allocate sizeof(long) multiples for the bitmask.
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Ben Goz