The current debugfs code is completely specific to FIFO's kickoff
profiler. But exposing these debugfs nodes is really a perfectly
generic operation to any given profiler.
Therefore add a generic debugfs interface for exposing profilers.
Any code that implements a profiler can now use a single function
call to export a profiler to the GPU debugfs area.
JIRA NVGPU-5606
Change-Id: I67a5bd9998fcfac94678e465442b9a38ab7e7612
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2358382
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add a generic profiler based on the channel kickoff profiler. This
aims to provide a mechanism to allow engineers to (more) easily profile
arbitrary software paths within nvgpu.
Usage of this profiler is still primarily through debugfs. Next up is
a generic debugfs interface for this profiler in the Linux code.
The end goal for this is to profile the recovery code and generate
interesting statistics.
JIRA NVGPU-5606
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Change-Id: I99783ec7e5143855845bde4e98760ff43350456d
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2355319
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
This adds a new device management unit in the common code responsible
for facilitating the parsing of the GPU top device list and providing
that info to other units in nvgpu.
The basic idea is to read this list once from HW and store it in a
set of lists corresponding to each device type (graphics, LCE, etc).
Many of the HALs in top can be deleted and instead implemented using
common code parsing the SW representation.
Every time the driver queries the device list it does so using a
device type and instance ID. This is common code. The HAL is responsible
for populating the device list in such a way that the driver can
query it in a chip agnostic manner.
Also delete some of the unit tests for functions that no longer
exist. This code will require new unit tests in time; those should be
quite simple to write once unit testing is needed.
JIRA NVGPU-5421
Change-Id: Ie41cd255404b90ae0376098a2d6e9f9abdd3f5ea
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2319649
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
In Bug 200588835, the spurious FBPA interrupts are seen on couple
of boards. These interrupts were found to be EDC (Error detection
and Correction) interrupts which are triggered due to ECC errors.
The EDC registers are not exposed to the driver, so the interrupt
status register cannot be cleared; resulting in interrupt storm.
Also, it was concluded that only bad HW can cause this failure
scenario. So, in the ISR for FBPA interrupts, get the GPU into
quiesce state as we don't expect the GPU to be in usable state post
such unrecoverable errors.
Adapt the quiesce code for Linux build too.
1. On Linux, we cannot exit the nvgpu process after quiesce like we
do on QNX. So, add nvgpu_disable_irqs() call to quiesce
implementation which is done as part of process exit handler on
QNX. Masking interrupts which is already done as part of quiesce
would be sufficient in most cases, but to be fail-safe
disable_irqs too.
3. Also, the IOCTL code looks at g->sw_ready, hence add
nvgpu_start_gpu_idle() to set g->sw_ready to false along with
setting NVGPU_DRIVER_IS_DYING = true.
We expect the nvgpu_sw_quiesce() call to finish before quiesce thread
wakes up from 50ms sleep. Hence, critical step like
nvgpu_start_gpu_idle() is added to nvgpu_sw_quiesce(), whereas the
somewhat redundant disable IRQs call is added to quiesce thread.
nvgpu_fifo_quiesce() was called twice by mistake; remove one of the
them.
Bug 2919899
Bug 200588835
Change-Id: I9beec688c2e1c0d8dfc1327ddf122684576f8684
Signed-off-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2354537
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: svc-mobile-misra <svc-mobile-misra@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The valid flag in struct nvgpu_fence_type is not very useful. It's set
when a fence is created on an allocated object and read in these three
scenarios:
- nvgpu_fence_install_fd() after a submit, if the submit was successful.
A successful submit implies that a post fence exists.
- nvgpu_fence_wait() for a copyengine job when synchronizing the ce
ringbuffer or when waiting for vidmem clears. In these cases the fence
is also clearly always valid.
- nvgpu_fence_is_expired() when testing whether a tracked job has
completed. Such jobs cannot exist without post fences that are
mandatory for tracking, so the fence must exist.
Remove the valid flag. Remove also the other init checks from the above
functions; they're equally unused and confusing implying that such calls
would be acceptable, causing sloppy code at best.
Jira NVGPU-5248
Jira NVGPU-5493
Change-Id: I52c5be1569b343024d2626bd9577f87b46064fba
Signed-off-by: Konsta Hölttä <kholtta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2357828
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-misra <svc-mobile-misra@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The differences between sync_fence ("android sync") and dma_fence are
abstracted away by nvhost in the nvhost_fence interface. There is no
need to have separate android and dma os fences for syncpoints; unify
the general implementation so that it's always used when requested for
the build.
Jira NVGPU-5386
Change-Id: Ia829e93e18d03064ff46ab1271547de2d1fb1cae
Signed-off-by: Konsta Hölttä <kholtta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2356158
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Implement empty stubs of the channel watchdog functions for when
watchdog is disabled from build. Add some forward declarations that were
missing. Now most call sites don't need #idefs for the build flag.
Add error checks for the wdt alloc failure.
Jira NVGPU-5494
Jira NVGPU-5493
Change-Id: I2d42e8ab4c5e045cd280b2e1f254396127bd154b
Signed-off-by: Konsta Hölttä <kholtta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2352050
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Currently, nvgpu_writel_loop() writes to a register and immediately
checks if register value is updated. It might take some time for
hardware registers to get updated with value written by software.
Modify nvgpu_writel_loop() to accept number of retries to check if
register value is updated and assert with nvgpu_assert().
Also, move nvgpu_writel_loop() to common code and use generic
nvgpu_readl() and nvgpu_writel() APIs.
JIRA NVGPU-5490
Change-Id: Iaaf24203a91eee3d05de7d0c7dea18113367de5f
Signed-off-by: Vedashree Vidwans <vvidwans@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2348628
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Vaibhav Kachore <vkachore@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Historically, nvgpu has supported a struct gk20a_dmabuf_priv and
associated it with a dmabuf instance. This was aided by Nvmap's
dma_buf_set_drv_data() and dma_buf_get_drvdata() APIs. gk20a_dmabuf_priv
is used to store Comptag IDs i.e. (1 per 64 kb) as well as can store the
dmabuf attachments to avoid multiple attach/detach calls. dma_buf_set_drv_data()
allows Nvgpu to associate an instance of struct gk20a_dmabuf_priv with the instance
of the dmabuf and also provide a release callback to delete the
instance when the last reference to the dmabuf is put. Nvmap accomplishes
this by modifying the struct dma_buf_ops definition to include the
set_drv_data and get_drv_data callbacks in the kernel code.
The above approach won't work for upstream Kstable and Nvmap
plans to remove these APIs for upcoming newer downstream kernels as
well.
In order to implement the same functionality without depending on Nvmap,
Nvgpu will implement a release chaining mechanism. Dmabuf's 'ops' pointer
points to a constant struct and hence a whole copy of the ops is made
followed by altering the new copy's release pointer.
struct gk20a_dmabuf_priv stores the new copy and the dmabuf's 'ops' is
changed to point to this. This allows Nvgpu to retrieve
the corresponding gk20a_dmabuf_priv instance using container_of.
Nvgpu's custom release callback will invoke the original release
callback of the dmabuf's producer as a last step, thus completing the
full circle. In case, the driver is removed, Nvgpu restores the
dmabuf's 'ops' back to the original state. In order to accomplish this,
every instance of a struct nvgpu_os_linux maintains a linkedlist of the
gk20a_dma_buf instances. During the driver removal, this linkedlist is
traversed and the corresponding dmabuf's 'ops' pointer is put back to
its original state followed by freeing of this instance.
Nvgpu is a producer of dmabuf's for vidmem and needs
a way to check whether the given dmabuf belongs to itself.
Its no longer reliable to depend on a comparision of
the 'ops' pointer. Instead dmabuf_export_info() allows a name to be set by the
exporter and this can be used to compare with a memory location
that belongs to Nvgpu. Similarly for sysmem dmabufs, Nvmap makes a
similar change in the way it identifies whether a dmabuf belongs to
itself.
Removed NVGPU_DMABUF_HAS_DRVDATA and moved to a unified mechanism for
both downstream as well as upstream kernel.
Some of the other changes in this file include the following.
1) Deletion of dmabuf.c and moving its contents over to dmabuf_priv.c
2) Replacing gk20a_mm_pin_has_drvdata with nvgpu_mm_pin_privdata and
vice-versa for unpin.
Bug 2878569
Change-Id: Icf8e79b05a25ad5a85f478c3ee0fc1eb7747e22d
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2341001
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: Puneet Saxena <puneets@nvidia.com>
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Currently, ltc fs_state is initialized during ltc init support. However,
ltc cbc_param and cbc_param2 registers do not seem to be providing
correct data if ltc.init_fs_state is called before fb.init_fs_state.
- Create fb.init_fb_support hal to initialize fb.
- Trigger init_fb_support before init_ltc_support.
Bug 2969956
Bug 2957808
JIRA NVGPU-4666
Change-Id: I54d697d27b9d9c6318c4ef459d215b6f82cd5571
Signed-off-by: Vedashree Vidwans <vvidwans@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2345673
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Make Recovery and quiesce co-exist to support quiesce state
on unrecoverrable errors. Currently, the quiesce code is wrapped
under ifndef CONFIG_NVGPU_RECOVERY. Isolate the quiesce code from
recovery config, thereby enabling it on all builds.
On Linux, the hung_task checker(check_hung_uninterruptible_tasks()
in kernel/hung_task.c) complains that quiesce thread is stuck for
more than 120 seconds.
INFO: task sw-quiesce:1068 blocked for more than 120 seconds.
The wait time of more than 120 seconds is expected as quiesce
thread will wait until quiesce call is triggered on fatal
unrecoverable errors. However, the INFO print upsets the
kernel_warning_test(KWT) on Linux builds. To fix the failing
KWT, change the quiesce task to interruptible instead of
uninterruptible as checker only looks at uninterruptible tasks.
Bug 2919899
JIRA NVGPU-5479
Change-Id: Ibd1023506859d8371998b785e881ace52cb5f030
Signed-off-by: tkudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2342774
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Disable the somewhat non-useful syncpoint debug spew in the nvgpu
debug spew. The GPU has it's own snapshot view of syncpoints so
visibility into other syncpoint data is often not very helpful.
However, there are plausibly times where this would be necessary.
For example debugging a sync issue between the GPU and some other
SoC engine. Therefore, the syncpoint debug spew can be enabled
again at runtime if necessary.
JIRA NVGPU-5541
Change-Id: I7028e2d6027e41835b2fed4f2bbb366c16b99967
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2349185
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
The gk20a_debug_dump() function implicitly adds a newline since it
uses nvgpu_err() under the hood (for uart destined prints). For the
seq_file destined writes it does not so there is an annoying inconsistency.
Remove the newline that many of the gk20a_debug_dump() calls add and add
the newline to the (now) seq_printf() call. This reduces the length of
debug dump logs and speeds them up - UART is _very_ slow after all.
Also cleanup some formatting issues in the various debug prints I
happened to notice.
JIRA NVGPU-5541
Change-Id: Iabf853d5c50214794fc4cbb602dfffabeb877132
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2347956
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
