GPU accumulates the context runtime in a 32 bit counter - CTX_TIMESTAMP
in the context image. This value is saved/restored on context switches.
KMD accumulates these values into a 64 bit counter taking care of any
overflows as needed. This count provides the basis for client specific
busyness in the fdinfo interface.
KMD accumulation happens just before the context is unpinned and when
context switches out. This works for execlist back-end since execlist
scheduling has visibility into context switches. With GuC mode, KMD does
not have visibility into context switches and this counter is
accumulated only when context is unpinned. Context is unpinned once the
context scheduling is successfully disabled. Disabling context
scheduling is an asynchronous operation. Also if a context is servicing
frequent requests, scheduling may never be disabled on it.
For GuC mode, since updates to the context runtime may be delayed, add
hooks to update the context runtime in a worker thread as well as when
a user queries for it.
Limitation:
- If a context is never switched out or runs for a long period of time,
the runtime value of CTX_TIMESTAMP may never be updated, so the
counter value may be unreliable. This patch does not support such
cases. Such support must be available from the GuC FW and it is WIP.
This patch is an extract from previous work authored by John/Umesh here -
https://patchwork.freedesktop.org/patch/496441/?series=105085&rev=4
v2: (Ashutosh)
- Drop COPS_RUNTIME_ACTIVE_TOTAL
- s/guc_context_update_clks/__guc_context_update_stats
- Pin context before accessing in guc_timestamp_ping
- In guc_context_unpin, use spinlock to serialize access to runtime stats
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Co-developed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230427224705.2785566-2-umesh.nerlige.ramappa@intel.com
SLPC enables use of efficient freq at init by default. It is
possible for GuC to request frequencies that are higher than
the 'software' max if user has set it lower than the efficient
level.
Scenarios/tests that require strict fixing of freq below the efficient
level will need to disable it through this interface.
v2: Keep just one interface to toggle sysfs. With this, user will
be completely responsible for toggling efficient frequency if need
be. There will be no implicit disabling when user sets min < RP1 (Ashutosh)
v3: Remove unused label, review comments (Ashutosh)
v4: Toggle efficient freq usage in SLPC selftest and checkpatch fixes
v5: Review comments (Andi) and add a separate patch for selftest updates
Fixes: 95ccf312a1 ("drm/i915/guc/slpc: Allow SLPC to use efficient frequency")
Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230426003942.1924347-1-vinay.belgaumkar@intel.com
MTL currently uses gen8_ppgtt_insert_huge when managing huge pages.
This is because MTL reports as not supporting 64K pages, or more
accurately, the system that reports whether a platform has 64K pages
reports false for MTL. This is only half correct, as the 64K page support
reporting system only cares about 64K page support for LMEM, which MTL
doesn't have.
MTL should be using xehpsdv_ppgtt_insert_huge. However, simply changing
over to using that manager doesn't resolve the issue because MTL is
expecting the virtual address space for the page table to be flushed after
initialization, so we must also add a flush statement there.
Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230425-hugepage-migrate-v8-2-7868d54eaa27@intel.com
On dGfx, the PL1 power limit being enabled and set to a low value results
in a low GPU operating freq. It also negates the freq raise operation which
is done before GuC firmware load. As a result GuC firmware load can time
out. Such timeouts were seen in the GL #8062 bug below (where the PL1 power
limit was enabled and set to a low value). Therefore disable the PL1 power
limit when allowed by HW when loading GuC firmware.
v2:
- Take mutex (to disallow writes to power1_max) across GuC reset/fw load
- Add hwm_power_max_restore to error return code path
v3 (Jani N):
- Add/remove explanatory comments
- Function renames
- Type corrections
- Locking annotation
v4:
- Don't hold the lock across GuC reset (Rodrigo)
- New locking scheme (suggested by Rodrigo)
- Eliminate rpm_get in power_max_disable/restore, not needed (Tvrtko)
v5:
- Fix uninitialized pl1en variable compile warning reported by kernel
build robot by creating new err_rps label
Link: https://gitlab.freedesktop.org/drm/intel/-/issues/8062
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230420164041.1428455-3-ashutosh.dixit@intel.com
This patch implements Wa_22016122933.
In MTL, memory writes initiated by the Media tile update the whole
cache line, even for partial writes. This creates a coherency
problem for cacheable memory if both CPU and GPU are writing data
to different locations within a single cache line.
This patch circumvents the issue by making CPU/GPU shared memory
uncacheable (WC on CPU side, and PAT index 2 for GPU). Additionally,
it ensures that CPU writes are visible to the GPU with an
intel_guc_write_barrier().
While fixing the CTB issue, we noticed some random GSC firmware
loading failure because the share buffers are cacheable (WB) on CPU
side but uncached on GPU side. To fix these issues we need to map
such shared buffers as WC on CPU side. Since such allocations are
not all done through GuC allocator, to avoid too many code changes,
the i915_coherent_map_type() is now hard coded to return WC for MTL.
v2: Simplify the commit message(Matt).
BSpec: 45101
Signed-off-by: Fei Yang <fei.yang@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Acked-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230424182902.3663500-3-fei.yang@intel.com
When HuC is loaded by GSC, there is no header definition for the kernel
to look at and firmware is just handed to GSC. However when reading the
version, it should still check the size of the blob to guarantee it's not
incurring into out-of-bounds array access.
If firmware is smaller than expected, the following message is now
printed:
# echo boom > /lib/firmware/i915/dg2_huc_gsc.bin
# dmesg | grep -i huc
[drm] GT0: HuC firmware i915/dg2_huc_gsc.bin: invalid size: 5 < 184
[drm] *ERROR* GT0: HuC firmware i915/dg2_huc_gsc.bin: fetch failed -ENODATA
...
Even without this change the size, header and signature are still
checked by GSC when loading, so this only avoids the out-of-bounds array
access.
Fixes: a7b516bd98 ("drm/i915/huc: Add fetch support for gsc-loaded HuC binary")
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230413200349.3492571-1-lucas.demarchi@intel.com
main pull request for v6.4
Core Display:
============
* Bugfixes for error handling during probe
* rework UBWC decoder programming
* prepare_commit cleanup
* bindings for SM8550 (MDSS, DPU), SM8450 (DP)
* timeout calculation fixup
* atomic: use drm_crtc_next_vblank_start() instead of our own
custom thing to calculate the start of next vblank
DP:
==
* interrupts cleanup
DPU:
===
* DSPP sub-block flush on sc7280
* support AR30 in addition to XR30 format
* Allow using REC_0 and REC_1 to handle wide (4k) RGB planes
* Split the HW catalog into individual per-SoC files
DSI:
===
* rework DSI instance ID detection on obscure platforms
GPU:
===
* uapi C++ compatibility fix
* a6xx: More robust gdsc reset
* a3xx and a4xx devfreq support
* update generated headers
* various cleanups and fixes
* GPU and GEM updates to avoid allocations which could trigger
reclaim (shrinker) in fence signaling path
* dma-fence deadline hint support and wait-boost
* a640 speedbin support
* a650 speedbin support
Conflicts in drivers/gpu/drm/msm/adreno/adreno_gpu.c:
Conflict between the 7fa5047a43 ("drm: Use of_property_present() for
testing DT property presence") and 9f251f9340 ("drm/msm/adreno: Use
OPP for every GPU generation"). The latter removed the of_ function
call outright, so I went with what's in the PR unchanged.
From: Rob Clark <robdclark@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/CAF6AEGvwuj5tabyW910+N-B=5kFNAC7QNYoQ=0xi3roBjQvFFQ@mail.gmail.com
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
This tag contains additional habanalabs driver changes for v6.4:
- uAPI changes:
- Add a definition of a new Gaudi2 server type. This is used by userspace
to know what is the connectivity between the accelerators inside the
server
- New features and improvements:
- speedup h/w queues test in Gaudi2 to reduce device initialization times.
- Firmware related fixes:
- Fixes to the handshake protocol during f/w initialization.
- Sync f/w events interrupt in hard reset to avoid warning message.
- Improvements to extraction of the firmware version.
- Misc bug fixes and code cleanups. Notable fixes are:
- Multiple fixes for interrupt handling in Gaudi2.
- Unmap mapped memory in case TLB invalidation fails.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
From: Oded Gabbay <ogabbay@kernel.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20230410124637.GA2441888@ogabbay-vm-u20.habana-labs.com
hl_sysfs_fini() is called only if hl_sysfs_init() completes
successfully. Therefore if hl_sysfs_init() fails, need to remove any
sysfs group that was added until that point.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
HW queues testing at driver load and after reset takes a substantial
amount of time.
This commit reduces the queues test time in Gaudi2 devices by running
all the tests in parallel instead of one after the other.
Time measurements on tests duration shows that the new method is almost
x100 faster than the serial approach.
Signed-off-by: Moti Haimovski <mhaimovski@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
There is only single eq entry for arc farm sei event which aggregates
events from the four arc farms.
Fix the code to handle this event according to this behavior.
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Multi MSI interrupts aren't working in Gaudi1 and because of that,
we are only using a single MSI interrupt. Therefore, let's remove this
dead code in order to avoid confusion.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Add definition of a new Gaudi2 server type. This represents
the connectivity between the cards in that server type.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Removing redundant asic prop variable as we don't need to expose this
to common code. In addition, fix some typos.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Sending COMMS_GOTO_WFE instructs the FW's CPU to halt (WFE state).
Once sent, FW's CPU isn't expected to continue communicating with LKD.
Therefore, the stage of waiting for COMMS_STS_OK should be skipped or
else waiting for COMMS_STS_OK will simply timeout, which will trigger
unexpected behavior.
Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Receiving events from FW, while the device is in hard reset, causes
a warning message in Driver log. The message may point to a
problem in the Driver or FW. But It also can appear as a result
of events that have been sent from FW just before the hard reset.
In order to avoid receiving events from FW while the device is in reset
and is already in 'disabled' mode, sync the f/w events interrupt right
before setting the device to 'disabled'.
Signed-off-by: Tal Cohen <talcohen@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
The decoder IRQ status register may have several set bits upon an
abnormal interrupt. Therefore, when setting the events mask, need to
check all bits and not using if-else.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Decoder abnormal interrupts are for errors and not for completion, so
rename the relevant work and work function to not include 'completion'.
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
There are rare cases of failures when cards are initialized due to
wrong values in efuse mappings that are parsed by firmware.
To help debug those cases, print (in debug level) the raw binning masks
as fetched from the firmware during device initialization.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Current mapping between HMMU event and HMMU block is wrong.
In addition the captured address in case of a page fault or
an access error is scrambled, Hence we must call the descramble
function.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
