The purpose here is to allow to optimize exec_queue_set_job_timeout()
in follow-on patch. Currently it does q->ops->set_job_timeout(...).
But we'd like to apply exec_queue_user_extensions much earlier and
q->ops cannot be called before __xe_exec_queue_init().
It will be much more efficient to instead only have to set
q->sched_props.job_timeout_ms when applying user extensions. That value
will then be used during q->ops->init().
Signed-off-by: Brian Welty <brian.welty@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Split __xe_exec_queue_create() into two functions, alloc and init.
We have an issue in that exec_queue_user_extensions are applied too late.
In the case of USM properties, these need to be set prior to xe_lrc_init().
Refactor the logic here, so we can resolve this in follow-on. We only need
the xe_vm_lock held during the exec_queue_init function.
Signed-off-by: Brian Welty <brian.welty@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Building drivers/gpu/drm/xe/xe_gt_pagefault.c with GCC 11 results
in the following build errors:
./include/linux/fortify-string.h:57:33: error: writing 16 bytes into a region of size 0 [-Werror=stringop-overflow=]
57 | #define __underlying_memcpy __builtin_memcpy
| ^
./include/linux/fortify-string.h:644:9: note: in expansion of macro ‘__underlying_memcpy’
644 | __underlying_##op(p, q, __fortify_size); \
| ^~~~~~~~~~~~~
./include/linux/fortify-string.h:689:26: note: in expansion of macro ‘__fortify_memcpy_chk’
689 | #define memcpy(p, q, s) __fortify_memcpy_chk(p, q, s, \
| ^~~~~~~~~~~~~~~~~~~~
drivers/gpu/drm/xe/xe_gt_pagefault.c:340:17: note: in expansion of macro ‘memcpy’
340 | memcpy(pf_queue->data + pf_queue->tail, msg, len * sizeof(u32));
| ^~~~~~
In file included from drivers/gpu/drm/xe/xe_device_types.h:17,
from drivers/gpu/drm/xe/xe_vm_types.h:16,
from drivers/gpu/drm/xe/xe_bo.h:13,
from drivers/gpu/drm/xe/xe_gt_pagefault.c:16:
drivers/gpu/drm/xe/xe_gt_types.h:102:25: note: at offset [1144, 265324] into destination object ‘tile’ of size 8
102 | struct xe_tile *tile;
| ^~~~
Fix these by removing -Wstringop-overflow from drm/xe builds.
Closes: https://lore.kernel.org/all/45ad1d0f-a10f-483e-848a-76a30252edbe@paulmck-laptop/
Fixes: 7a8bc11782 ("drm/xe: Enable W=1 warnings by default")
Suggested-by: Stephen Rothwell <sfr@rothwell.id.au>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
[ This particular warning is broken on GCC11. In future changes it will
be moved to the normal C flags in the top level Makefile (out of
Makefile.extrawarn), but accounting for the compiler support. Just
remove it out of xe's forced extra warnings for now ]
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Add vram frequency sysfs attributes under the below hierarchy;
/device/tile#/memory/freq0
|-max_freq
|-min_freq
v2: Drop "vram" from attribute names (Rodrigo)
v3: Add documentation for new sysfs (Riana)
Drop prefix from XEHP_PCODE_FREQUENCY_CONFIG (Riana)
v4: Create sysfs under tile#/freq0 after removal of
physical_memsize attrbute
v5: Revert back to creating sysfs under tile#/memory/freq0
Remove definition of GT_FREQUENCY_MULTIPLIER (Rodrigo)
v6: Rename attributes to max/min_freq (Anshuman)
Fix review comments (Rodrigo)
v7: Make documentation more verbose
Move sysfs to separate file (Anshuman)
v8: Fix platform specific conditions and add kernel doc (Anshuman)
Fix typos and remove redundant headers (Riana)
v9: Fix typo (Riana)
Change function name to include "sysfs" (Lucas)
Signed-off-by: Sujaritha Sundaresan <sujaritha.sundaresan@intel.com>
Reviewed-by: Anshuman Gupta <anshuman.gupta@intel.com>
Link: https://lore.kernel.org/r/20240109110418.2065101-1-sujaritha.sundaresan@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
After exec_queue has been created, we cannot simply modify q->priority.
This needs to be done by the backend via q->ops. However in this case,
it would be more efficient to simply pass a flag when creating the
exec_queue and set the desired priority upfront during queue creation.
To that end: new flag EXEC_QUEUE_FLAG_HIGH_PRIORITY is introduced.
The priority field is moved to be with other scheduling properties and
is now exec_queue.sched_props.priority. This is no longer set to initial
value by the backend, but is now set within __xe_exec_queue_create().
Fixes: b4eecedc75 ("drm/xe: Fix potential deadlock handling page faults")
Signed-off-by: Brian Welty <brian.welty@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
We need to set q->priority prior to calling guc_exec_queue_add_msg() as
that will call init_policies() and sets the scheduling properties to those
stored in the exec_queue.
Fixes: dd08ebf6c3 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Signed-off-by: Brian Welty <brian.welty@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Protect entire IRQ handler by CT being enabled rather than just G2H
handler.
v2: Return on not enabled in CT irq handler (Michal)
Suggested-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
The intent is to return -EWOULDBLOCK to the user if a long running exec
queue is full during the exec IOCTL. -EWOULDBLOCK aliases to -EAGAIN
which results in the exec IOCTL doing a retry loop. Fix this by ensuring
the retry loop is broken when returning -EWOULDBLOCK.
Fixes: 8ae8a2e8dd ("drm/xe: Long running job update")
Reported-by: Sai Gowtham Ch <sai.gowtham.ch@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Brian Welty <brian.welty@intel.com>
We currently assume that we can upfront know exactly how many fence
slots we will need at the start of the exec, however the TTM bo_validate
can itself consume numerous fence slots, and due to how the
dma_resv_reserve_fences() works it only ensures that at least that many
fence slots are available. With this it is quite possible that TTM
steals some of the fence slots and then when it comes time to do the vma
binding and final exec stage we are lacking enough fence slots, leading
to some nasty BUG_ON(). A simple fix is to reserve our own fences later,
after the validate stage.
References: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/698
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Tested-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Release all mmap mappings for all vram objects which are associated
with userfault such that, while pcie function in D3hot, any access
to memory mappings will raise a userfault.
Upon userfault, in order to access memory mappings, if graphics
function is in D3 then runtime resume of dgpu will be triggered to
transition to D0.
v2:
- Avoid iomem check before bo migration check as bo can migrate
to system memory (Matthew Auld)
v3:
- Delete bo userfault link during bo destroy
- Upon bo move (vram-smem), do bo userfault link deletion in
xe_bo_move_notify instead of xe_bo_move (Thomas Hellström)
- Grab lock in rpm hook while deleting bo userfault link (Matthew Auld)
v4:
- Add kernel doc and wrap vram_userfault related
stuff in the structure (Matthew Auld)
- Get rpm wakeref before taking dma reserve lock (Matthew Auld)
- In suspend path apply lock for entire list op
including list iteration (Matthew Auld)
v5:
- Use mutex lock instead of spin lock
v6:
- Fix review comments (Matthew Auld)
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Anshuman Gupta <anshuman.gupta@intel.com>
Signed-off-by: Badal Nilawar <badal.nilawar@intel.com>
Acked-by: Thomas Hellström <thomas.hellstrom@linux.intel.com> #For the xe_bo_move_notify() changes
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://lore.kernel.org/r/20240104130702.950078-1-badal.nilawar@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Recommendation is to read FUSE4 register to check if WMTP has been
enabled/disabled by HW. If enabled we don't need to do anything special,
however if disabled recommendation is to also disable the WMTP mode in
the FF_SLICE_CS_CHICKEN2 register, falling back to thread-group and
mid-batch preemption only. However on Linux, the per-context CS_CHICKEN1
is how userspace controls pre-emption, so instead use the default lrc to
disable WMTP using CS_CHICKEN1, if disabled by HW. Userspace is still
free to set CS_CHICKEN1 to whatever they want later.
v2: remove redundant version check and also add descriptive name(Matt)
v3: remove usage of REG_FIELD_GET(Matt)
Cc: Matt Roper <matthew.d.roper@intel.com>
Co-developed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20240104182615.21327-1-nirmoy.das@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
It looks that declaration of kunit_get_current_test() was
available in xe_guc_relay.c only with CONFIG_KUNIT enabled.
If CONFIG_KUNIT is disabled we fail with:
In file included from ../include/linux/build_bug.h:5,
from ../include/linux/bitfield.h:10,
from ../drivers/gpu/drm/xe/xe_guc_relay.c:6:
../drivers/gpu/drm/xe/xe_guc_relay.c: In function ‘xe_guc_relay_process_guc2vf’:
../drivers/gpu/drm/xe/xe_guc_relay.c:863:52: error: implicit declaration of function ‘kunit_get_current_test’ [-Werror=implicit-function-declaration]
863 | if (unlikely(!IS_SRIOV_VF(relay_to_xe(relay)) && !kunit_get_current_test()))
| ^~~~~~~~~~~~~~~~~~~~~~
../include/linux/compiler.h:77:42: note: in definition of macro ‘unlikely’
77 | # define unlikely(x) __builtin_expect(!!(x), 0)
| ^
Reported-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Fixes: 811fe9f556 ("drm/xe/guc: Introduce Relay Communication for SR-IOV")
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Tested-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Link: https://lore.kernel.org/r/20240105171947.321-1-michal.wajdeczko@intel.com
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
There are scenarios where SR-IOV Virtual Function (VF) driver will
need to get additional data that is not available over VF MMIO BAR
nor could be queried from the GuC firmware and must be obtained
from the Physical Function (PF) driver.
To allow such communication between VF and PF drivers, GuC supports
set of H2G and G2H actions which allows relaying embedded messages,
that are otherwise opaque for the GuC.
To allow use of this communication mechanism, provide functions for
sending requests and handling replies and placeholder where we will
put handlers for incoming requests.
Reviewed-by: Piotr Piórkowski <piotr.piorkowski@intel.com>
Link: https://lore.kernel.org/r/20240104222031.277-8-michal.wajdeczko@intel.com
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Now intel_wakeref_t is a unsigned long and Xe KMD version of those
functions should use the same type, so changing from bool to
intel_wakeref_t.
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
i915 defines it as unsigned long so Xe should do the same to avoid
compilation warnings:
CC [M] drivers/gpu/drm/i915/i915_gem.o
CC [M] drivers/gpu/drm/xe/i915-display/intel_display_power_well.o
In file included from ./include/drm/drm_mm.h:51,
from drivers/gpu/drm/xe/xe_bo_types.h:11,
from drivers/gpu/drm/xe/xe_bo.h:11,
from ./drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object.h:11,
from ./drivers/gpu/drm/xe/compat-i915-headers/i915_drv.h:15,
from drivers/gpu/drm/i915/display/intel_display_power.c:8:
drivers/gpu/drm/i915/display/intel_display_power.c: In function ‘print_async_put_domains_state’:
drivers/gpu/drm/i915/display/intel_display_power.c:408:29: warning: format ‘%lu’ expects argument of type ‘long unsigned int’, but argument 5 has type ‘int’ [-Wformat=]
408 | drm_dbg(&i915->drm, "async_put_wakeref %lu\n",
| ^~~~~~~~~~~~~~~~~~~~~~~~~
409 | power_domains->async_put_wakeref);
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| |
| int
./include/drm/drm_print.h:410:39: note: in definition of macro ‘drm_dev_dbg’
410 | __drm_dev_dbg(NULL, dev, cat, fmt, ##__VA_ARGS__)
| ^~~
./include/drm/drm_print.h:510:33: note: in expansion of macro ‘drm_dbg_driver’
510 | #define drm_dbg(drm, fmt, ...) drm_dbg_driver(drm, fmt, ##__VA_ARGS__)
| ^~~~~~~~~~~~~~
drivers/gpu/drm/i915/display/intel_display_power.c:408:9: note: in expansion of macro ‘drm_dbg’
408 | drm_dbg(&i915->drm, "async_put_wakeref %lu\n",
| ^~~~~~~
drivers/gpu/drm/i915/display/intel_display_power.c:408:50: note: format string is defined here
408 | drm_dbg(&i915->drm, "async_put_wakeref %lu\n",
| ~~^
| |
| long unsigned int
| %u
CC [M] drivers/gpu/drm/i915/i915_gem_evict.o
CC [M] drivers/gpu/drm/i915/i915_gem_gtt.o
CC [M] drivers/gpu/drm/xe/i915-display/intel_display_trace.o
CC [M] drivers/gpu/drm/xe/i915-display/intel_display_wa.o
CC [M] drivers/gpu/drm/i915/i915_query.o
Fixes: 44e694958b ("drm/xe/display: Implement display support")
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Introduce a new DRM driver for Intel GPUs
Xe, is a new driver for Intel GPUs that supports both integrated and
discrete platforms. The experimental support starts with Tiger Lake.
i915 will continue be the main production driver for the platforms
up to Meteor Lake and Alchemist. Then the goal is to make this Intel
Xe driver the primary driver for Lunar Lake and newer platforms.
It uses most, if not all, of the key drm concepts, in special: TTM,
drm-scheduler, drm-exec, drm-gpuvm/gpuva and others.
Signed-off-by: Dave Airlie <airlied@redhat.com>
[airlied: add an extra X86 check, fix a typo, fix drm_exec_init interface
change].
From: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/ZYSwLgXZUZ57qGPQ@intel.com
We're currently sending non-blocking H2G messages using the EVENT type,
which suppresses all CTB protocol replies from the GuC, including the
failure cases. This might cause errors to slip through and manifest as
unexpected behavior (e.g. a context state might not be what the driver
thinks it is because the state change command was silently rejected by
the GuC). To avoid this kind of problems, we can use the FAST_REQUEST
type instead, which suppresses the reply only on success; this way we
still get the advantage of not having to wait for an ack from the GuC
(i.e. the H2G is still non-blocking) while still detecting errors.
Since we can't escalate to the caller when a non-blocking message
fails, we need to escalate to GT reset instead.
Note that FAST_REQUEST failures are NOT expected and are usually a sign
that the H2G was either malformed or requested an illegal operation.
v2: assign fence values to FAST_REQUEST messages, fix abi doc, use xe_gt
printers (Michal).
v3: fix doc alignment, fix and improve prints (Michal)
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com> #v2
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
