If we issue a reset to a currently idle engine, leave it idle
afterwards. This is useful to excise a linkage between reset and the
shrinker. When waking the engine, we need to pin the default context
image which we use for overwriting a guilty context -- if the engine is
idle we do not need this pinned image! However, this pinning means that
waking the engine acquires the FS_RECLAIM, and so may trigger the
shrinker. The shrinker itself may need to wait upon the GPU to unbind
and object and so may require services of reset; ergo we should avoid
the engine wake up path.
The danger in skipping the recovery for idle engines is that we leave the
engine with no context defined, which may interfere with the operation of
the power context on some older platforms. In practice, we should only
be resetting an active GPU but it something to look out for on Ironlake
(if memory serves).
Fixes: 79ffac8599 ("drm/i915: Invert the GEM wakeref hierarchy")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190626154549.10066-2-chris@chris-wilson.co.uk
(cherry picked from commit 18398904ca)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Windows guest can't run after force-TDR with host log:
...
gvt: vgpu 1: workload shadow ppgtt isn't ready
gvt: vgpu 1: fail to dispatch workload, skip
...
The error is raised by set_context_ppgtt_from_shadow(), when it checks
and found the shadow_mm isn't marked as shadowed.
In work thread before each submission, a shadow_mm is set to shadowed in:
shadow_ppgtt_mm()
<-intel_vgpu_pin_mm()
<-prepare_workload()
<-dispatch_workload()
<-workload_thread()
However checking whether or not shadow_mm is shadowed is prior to it:
set_context_ppgtt_from_shadow()
<-dispatch_workload()
<-workload_thread()
In normal case, create workload will check the existence of shadow_mm,
if not it will create a new one and marked as shadowed. If already exist
it will reuse the old one. Since shadow_mm is reused, checking of shadowed
in set_context_ppgtt_from_shadow() actually always see the state set in
creation, but not the state set in intel_vgpu_pin_mm().
When force-TDR, all engines are reset, since it's not dmlr level, all
ppgtt_mm are invalidated but not destroyed. Invalidation will mark all
reused shadow_mm as not shadowed but still keeps in ppgtt_mm_list_head.
If workload submission phase those shadow_mm are reused with shadowed
not set, then set_context_ppgtt_from_shadow() will report error.
Pin for context after shadow_mm pinned and shadow pdps settled.
v2:
Move set_context_ppgtt_from_shadow() after prepare_workload(). (zhenyu)
v3:
Move set_context_ppgtt_from_shadow() after shadow pdps updated.(zhenyu)
Fixes: 4f15665ccb ("drm/i915: Add ppgtt to GVT GEM context")
Cc: stable@vger.kernel.org
Signed-off-by: Colin Xu <colin.xu@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
GPU hang observed during the guest OCL conformance test which is caused
by THP GTT feature used durning the test.
It was observed the same GFN with different size (4K and 2M) requested
from the guest in GVT. So during the guest page dma map stage, it is
required to unmap first with orginal size and then remap again with
requested size.
Fixes: b901b252b6 ("drm/i915/gvt: Add 2M huge gtt support")
Cc: stable@vger.kernel.org
Reviewed-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: Xiaolin Zhang <xiaolin.zhang@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Workload contains RB and WA_CTX which are in ggtt space,
if they aren't in valid ggtt space, the workload shouldn't be
shadowed and scanned. So checking them earlier to avoid shadow
them.
Reviewed-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: Xiong Zhang <xiong.y.zhang@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Instead of silently return virtual ggtt entries that guest is allowed
to access, this patch add extra range check. If guest read out of
range, it will print a warning and return 0. If guest write out
of range, the write will be dropped without any message.
Reviewed-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: Xiong Zhang <xiong.y.zhang@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
On VLV/CHV there is some kind of linkage between the cdclk frequency
and the DP link frequency. The spec says:
"For DP audio configuration, cdclk frequency shall be set to
meet the following requirements:
DP Link Frequency(MHz) | Cdclk frequency(MHz)
270 | 320 or higher
162 | 200 or higher"
I suspect that would more accurately be expressed as
"cdclk >= DP link clock", and in any case we can express it like
that in the code because of the limited set of cdclk (200, 266,
320, 400 MHz) and link frequencies (162 and 270 MHz) we support.
Without this we can end up in a situation where the cdclk
is too low and enabling DP audio will kill the pipe. Happens
eg. with 2560x1440 modes where the 266MHz cdclk is sufficient
to pump the pixels (241.5 MHz dotclock) but is too low for
the DP audio due to the link frequency being 270 MHz.
v2: Spell out the cdclk and link frequencies we actually support
Cc: stable@vger.kernel.org
Tested-by: Stefan Gottwald <gottwald@igel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111149
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190717114536.22937-1-ville.syrjala@linux.intel.com
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
(cherry picked from commit bffb31f73b)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
If we hit an error while allocating the page tables, we have to unwind
the incomplete updates, and wish to free the unused pd. However, we are
not allowed to be hoding the spinlock at that point, and so must use the
later free to defer it until after we drop the lock.
<3> [414.363795] BUG: sleeping function called from invalid context at drivers/gpu/drm/i915/i915_gem_gtt.c:472
<3> [414.364167] in_atomic(): 1, irqs_disabled(): 0, pid: 3905, name: i915_selftest
<4> [414.364406] 3 locks held by i915_selftest/3905:
<4> [414.364408] #0: 0000000034fe8aa8 (&dev->mutex){....}, at: device_driver_attach+0x18/0x50
<4> [414.364415] #1: 000000006bd8a560 (&dev->struct_mutex){+.+.}, at: igt_ctx_exec+0xb7/0x410 [i915]
<4> [414.364476] #2: 000000003dfdc766 (&(&pd->lock)->rlock){+.+.}, at: gen8_ppgtt_alloc_pdp+0x448/0x540 [i915]
<3> [414.364529] Preemption disabled at:
<4> [414.364530] [<0000000000000000>] 0x0
<4> [414.364696] CPU: 0 PID: 3905 Comm: i915_selftest Tainted: G U 5.2.0-rc7-CI-CI_DRM_6403+ #1
<4> [414.364698] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.10.1-0-g8891697-prebuilt.qemu-project.org 04/01/2014
<4> [414.364699] Call Trace:
<4> [414.364704] dump_stack+0x67/0x9b
<4> [414.364708] ___might_sleep+0x167/0x250
<4> [414.364777] vm_free_page+0x24/0xc0 [i915]
<4> [414.364852] free_pd+0xf/0x20 [i915]
<4> [414.364897] gen8_ppgtt_alloc_pdp+0x489/0x540 [i915]
<4> [414.364946] gen8_ppgtt_alloc_4lvl+0x8e/0x2e0 [i915]
<4> [414.364992] ppgtt_bind_vma+0x2e/0x60 [i915]
<4> [414.365039] i915_vma_bind+0xe8/0x2c0 [i915]
<4> [414.365088] __i915_vma_do_pin+0xa1/0xd20 [i915]
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111050
Fixes: 1d1b5490b9 ("drm/i915/gtt: Replace struct_mutex serialisation for allocation")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190703171913.16585-3-chris@chris-wilson.co.uk
(cherry picked from commit 068610895e)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
When a register is readonly there is not much we can tell about its
value (apart from its default value?). This can be covered by tests
exercising the value of the register from userspace.
For PS_INVOCATION_COUNT we've got the following piglit tests :
KHR-GL45.pipeline_statistics_query_tests_ARB.functional_fragment_shader_invocations
Vulkan CTS tests :
dEQP-VK.query_pool.statistics_query.fragment_shader_invocations.*
v2: Use a local to shrink under 80cols.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 86554f48e5 ("drm/i915/selftests: Verify whitelist of context registers")
Tested-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190629131350.31185-1-chris@chris-wilson.co.uk
(cherry picked from commit 361b690513)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Gen2 doesn't have a frame counter and apparently we no longer provide
a fake .get_vblank_counter() hook for it. That means all tracepoints
calling that hook will oops. Update the tracepoints to use
intel_crtc_get_vblank_counter() which will gracefully fall back to
using the software counter. This is actually a better approach since
we now get (hopefully accurate) frame numbers in the traces.
This also gets rid of the raw driver->get_vblank_counter() calls, which
we need to do in order to switch to the per-crtc vblank vfuncs.
v2: Deal with new tracepoints
v3: Use a distinct variable name for the internal crtc iterator (Chris)
Cc: Shawn Guo <shawn.guo@linaro.org>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Fixes: 967dd48417 ("drm: remove drm_vblank_no_hw_counter assignment from driver code")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190619170842.20579-2-ville.syrjala@linux.intel.com
(cherry picked from commit 4c888e7bd2)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
In preparation to enabling -Wimplicit-fallthrough, mark switch
cases where we are expecting to fall through.
This patch fixes the following warnings:
drivers/gpu/drm/i915/gem/i915_gem_mman.c: In function ‘i915_gem_fault’:
drivers/gpu/drm/i915/gem/i915_gem_mman.c:342:6: warning: this statement may fall through [-Wimplicit-fallthrough=]
if (!i915_terminally_wedged(i915))
^
drivers/gpu/drm/i915/gem/i915_gem_mman.c:345:2: note: here
case -EAGAIN:
^~~~
drivers/gpu/drm/i915/gem/i915_gem_pages.c: In function ‘i915_gem_object_map’:
./include/linux/compiler.h:78:22: warning: this statement may fall through [-Wimplicit-fallthrough=]
# define unlikely(x) __builtin_expect(!!(x), 0)
^~~~~~~~~~~~~~~~~~~~~~~~~~
./include/asm-generic/bug.h:136:2: note: in expansion of macro ‘unlikely’
unlikely(__ret_warn_on); \
^~~~~~~~
drivers/gpu/drm/i915/i915_utils.h:49:25: note: in expansion of macro ‘WARN’
#define MISSING_CASE(x) WARN(1, "Missing case (%s == %ld)\n", \
^~~~
drivers/gpu/drm/i915/gem/i915_gem_pages.c:270:3: note: in expansion of macro ‘MISSING_CASE’
MISSING_CASE(type);
^~~~~~~~~~~~
drivers/gpu/drm/i915/gem/i915_gem_pages.c:272:2: note: here
case I915_MAP_WB:
^~~~
drivers/gpu/drm/i915/i915_gpu_error.c: In function ‘error_record_engine_registers’:
./include/linux/compiler.h:78:22: warning: this statement may fall through [-Wimplicit-fallthrough=]
# define unlikely(x) __builtin_expect(!!(x), 0)
^~~~~~~~~~~~~~~~~~~~~~~~~~
./include/asm-generic/bug.h:136:2: note: in expansion of macro ‘unlikely’
unlikely(__ret_warn_on); \
^~~~~~~~
drivers/gpu/drm/i915/i915_utils.h:49:25: note: in expansion of macro ‘WARN’
#define MISSING_CASE(x) WARN(1, "Missing case (%s == %ld)\n", \
^~~~
drivers/gpu/drm/i915/i915_gpu_error.c:1196:5: note: in expansion of macro ‘MISSING_CASE’
MISSING_CASE(engine->id);
^~~~~~~~~~~~
drivers/gpu/drm/i915/i915_gpu_error.c:1197:4: note: here
case RCS0:
^~~~
drivers/gpu/drm/i915/display/intel_dp.c: In function ‘intel_dp_get_fia_supported_lane_count’:
./include/linux/compiler.h:78:22: warning: this statement may fall through [-Wimplicit-fallthrough=]
# define unlikely(x) __builtin_expect(!!(x), 0)
^~~~~~~~~~~~~~~~~~~~~~~~~~
./include/asm-generic/bug.h:136:2: note: in expansion of macro ‘unlikely’
unlikely(__ret_warn_on); \
^~~~~~~~
drivers/gpu/drm/i915/i915_utils.h:49:25: note: in expansion of macro ‘WARN’
#define MISSING_CASE(x) WARN(1, "Missing case (%s == %ld)\n", \
^~~~
drivers/gpu/drm/i915/display/intel_dp.c:233:3: note: in expansion of macro ‘MISSING_CASE’
MISSING_CASE(lane_info);
^~~~~~~~~~~~
drivers/gpu/drm/i915/display/intel_dp.c:234:2: note: here
case 1:
^~~~
drivers/gpu/drm/i915/display/intel_display.c: In function ‘check_digital_port_conflicts’:
CC [M] drivers/gpu/drm/nouveau/nvkm/engine/disp/cursgv100.o
drivers/gpu/drm/i915/display/intel_display.c:12043:7: warning: this statement may fall through [-Wimplicit-fallthrough=]
if (WARN_ON(!HAS_DDI(to_i915(dev))))
^
drivers/gpu/drm/i915/display/intel_display.c:12046:3: note: here
case INTEL_OUTPUT_DP:
^~~~
Also, notice that the Makefile is modified to stop ignoring
fall-through warnings. The -Wimplicit-fallthrough option
will be enabled globally in v5.3.
Warning level 3 was used: -Wimplicit-fallthrough=3
This patch is part of the ongoing efforts to enable
-Wimplicit-fallthrough.
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
In the sysctl code the proc_dointvec_minmax() function is often used to
validate the user supplied value between an allowed range. This
function uses the extra1 and extra2 members from struct ctl_table as
minimum and maximum allowed value.
On sysctl handler declaration, in every source file there are some
readonly variables containing just an integer which address is assigned
to the extra1 and extra2 members, so the sysctl range is enforced.
The special values 0, 1 and INT_MAX are very often used as range
boundary, leading duplication of variables like zero=0, one=1,
int_max=INT_MAX in different source files:
$ git grep -E '\.extra[12].*&(zero|one|int_max)' |wc -l
248
Add a const int array containing the most commonly used values, some
macros to refer more easily to the correct array member, and use them
instead of creating a local one for every object file.
This is the bloat-o-meter output comparing the old and new binary
compiled with the default Fedora config:
# scripts/bloat-o-meter -d vmlinux.o.old vmlinux.o
add/remove: 2/2 grow/shrink: 0/2 up/down: 24/-188 (-164)
Data old new delta
sysctl_vals - 12 +12
__kstrtab_sysctl_vals - 12 +12
max 14 10 -4
int_max 16 - -16
one 68 - -68
zero 128 28 -100
Total: Before=20583249, After=20583085, chg -0.00%
[mcroce@redhat.com: tipc: remove two unused variables]
Link: http://lkml.kernel.org/r/20190530091952.4108-1-mcroce@redhat.com
[akpm@linux-foundation.org: fix net/ipv6/sysctl_net_ipv6.c]
[arnd@arndb.de: proc/sysctl: make firmware loader table conditional]
Link: http://lkml.kernel.org/r/20190617130014.1713870-1-arnd@arndb.de
[akpm@linux-foundation.org: fix fs/eventpoll.c]
Link: http://lkml.kernel.org/r/20190430180111.10688-1-mcroce@redhat.com
Signed-off-by: Matteo Croce <mcroce@redhat.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Aaron Tomlin <atomlin@redhat.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull Kbuild updates from Masahiro Yamada:
- remove headers_{install,check}_all targets
- remove unreasonable 'depends on !UML' from CONFIG_SAMPLES
- re-implement 'make headers_install' more cleanly
- add new header-test-y syntax to compile-test headers
- compile-test exported headers to ensure they are compilable in
user-space
- compile-test headers under include/ to ensure they are self-contained
- remove -Waggregate-return, -Wno-uninitialized, -Wno-unused-value
flags
- add -Werror=unknown-warning-option for Clang
- add 128-bit built-in types support to genksyms
- fix missed rebuild of modules.builtin
- propagate 'No space left on device' error in fixdep to Make
- allow Clang to use its integrated assembler
- improve some coccinelle scripts
- add a new flag KBUILD_ABS_SRCTREE to request Kbuild to use absolute
path for $(srctree).
- do not ignore errors when compression utility is missing
- misc cleanups
* tag 'kbuild-v5.3' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild: (49 commits)
kbuild: use -- separater intead of $(filter-out ...) for cc-cross-prefix
kbuild: Inform user to pass ARCH= for make mrproper
kbuild: fix compression errors getting ignored
kbuild: add a flag to force absolute path for srctree
kbuild: replace KBUILD_SRCTREE with boolean building_out_of_srctree
kbuild: remove src and obj from the top Makefile
scripts/tags.sh: remove unused environment variables from comments
scripts/tags.sh: drop SUBARCH support for ARM
kbuild: compile-test kernel headers to ensure they are self-contained
kheaders: include only headers into kheaders_data.tar.xz
kheaders: remove meaningless -R option of 'ls'
kbuild: support header-test-pattern-y
kbuild: do not create wrappers for header-test-y
kbuild: compile-test exported headers to ensure they are self-contained
init/Kconfig: add CONFIG_CC_CAN_LINK
kallsyms: exclude kasan local symbols on s390
kbuild: add more hints about SUBDIRS replacement
coccinelle: api/stream_open: treat all wait_.*() calls as blocking
coccinelle: put_device: Add a cast to an expression for an assignment
coccinelle: put_device: Adjust a message construction
...
Despite what I think the prm recommends, commit f2253bd985
("drm/i915/ringbuffer: EMIT_INVALIDATE after switch context") turned out
to be a huge mistake when enabling Ironlake contexts as the GPU would
hang on either a MI_FLUSH or PIPE_CONTROL immediately following the
MI_SET_CONTEXT of an active mesa context (more vanilla contexts, e.g.
simple rendercopies with igt, do not suffer).
Ville found the following clue,
"[DevCTG+]: For the invalidate operation of the pipe control, the
following pointers are affected. The
invalidate operation affects the restore of these packets. If the pipe
control invalidate operation is completed
before the context save, the indirect pointers will not be restored from
memory.
1. Pipeline State Pointer
2. Media State Pointer
3. Constant Buffer Packet"
which suggests by us emitting the INVALIDATE prior to the MI_SET_CONTEXT,
we prevent the context-restore from chasing the dangling pointers within
the image, and explains why this likely prevents the GPU hang.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419111749.3910-1-chris@chris-wilson.co.uk
(cherry picked from commit 928f8f4231 in drm-intel-next)
Cc: stable@vger.kernel.org
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111014
Fixes: f2253bd985 ("drm/i915/ringbuffer: EMIT_INVALIDATE after switch context")
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Features:
- HDR support (Uma, Ville)
- Add I2C symlink under HDMI connector similar to DP (Oleg)
- Add ICL multi-segmented gamma support (Shashank, Uma)
- Update register whitelist support for new hardware (Robert, John)
- GuC firmware update with updated ABI interface (Michal, Oscar)
- Add support for new DMC header versions (Lucas)
- In-kernel blitter client for selftest use (Matthew)
- Add Mule Creec Canyon (MCC) PCH support to go with EHL (Matt)
- EHL platform feature updates (Matt)
- Use Command Transport Buffers with GuC on all gens (Daniele)
- New i915.force_probe module parameter to replace i915.alpha_support (Jani)
Refactoring:
- Better runtime PM code abstraction/encapsulation (Daniele)
- VBT parsing cleanup and improvements (Jani)
- Move display code to its own subdirectory (Jani)
- Header cleanup (Jani, Daniele)
- Prep work for subsclice mask expansion (Stuart)
- Use uncore mmio register accessors more, remove unused macro wrappers (Tvrtko)
- Remove unused atomic property get/set stubs (Maarten)
- GTT cleanups and improvements (Mika)
- Pass intel_ types instead of drm_ types in plenty of display code (Ville)
- Engine reset, hangcheck, fault code cleanups and improvements (Tvrtko)
- Consider AML variants simply as either KBL or CFL ULX (Ville)
- State checker cleanups and improvements (Ville)
- GEM code reorganization to more files under gem subdirectory (Chris)
- Reducing dependency on a coarse struct_mutex (Chris)
Fixes:
- Fix use of uninitialized/incorrect error pointers (Colin, Dan)
- Fix DSI fastboot on some VLV/CHV platforms (Hans)
- Fix DSI error path (Hans)
- Add ICL port A combo PHY HW state check (Imre)
- Fix ICL AUX-B HW not done issue (Imre)
- Fix perf whitelist on gen10+ (Lionel)
- Fix PSR exit by forcing manual exit on older gens (José)
- Match voltage ranges instead of exact values (Lucas)
- Fix SDVO HDMI audio, with cleanups (Ville)
- Fix plane state dumps (Ville)
- Fix driver cleanup code to support driver hot unbind (Janusz)
- Add checks for ICL memory bandwidth requirements (Ville)
- Fix toggling between no C8 planes vs. at least one C8 plane (Ville)
- Improved checks on PLL usage conditions, refactoring (Ville)
- Avoid clobbering M/N values in fastset fuzzy checks (Ville)
- Take a runtime pm wakeref for atomic commits (Chris)
- Do not allow runtime pm autosuspend to remove userspace GGTT mmaps too quickly (Chris)
- Avoid refcount_inc on known zero count to avoid debug flagging (Chris)
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/87v9x1lpdh.fsf@intel.com
The idea behind keeping the saturation mask local to a context backfired
spectacularly. The premise with the local mask was that we would be more
proactive in attempting to use semaphores after each time the context
idled, and that all new contexts would attempt to use semaphores
ignoring the current state of the system. This turns out to be horribly
optimistic. If the system state is still oversaturated and the existing
workloads have all stopped using semaphores, the new workloads would
attempt to use semaphores and be deprioritised behind real work. The
new contexts would not switch off using semaphores until their initial
batch of low priority work had completed. Given sufficient backload load
of equal user priority, this would completely starve the new work of any
GPU time.
To compensate, remove the local tracking in favour of keeping it as
global state on the engine -- once the system is saturated and
semaphores are disabled, everyone stops attempting to use semaphores
until the system is idle again. One of the reason for preferring local
context tracking was that it worked with virtual engines, so for
switching to global state we could either do a complete check of all the
virtual siblings or simply disable semaphores for those requests. This
takes the simpler approach of disabling semaphores on virtual engines.
The downside is that the decision that the engine is saturated is a
local measure -- we are only checking whether or not this context was
scheduled in a timely fashion, it may be legitimately delayed due to user
priorities. We still have the same dilemma though, that we do not want
to employ the semaphore poll unless it will be used.
v2: Explain why we need to assume the worst wrt virtual engines.
Fixes: ca6e56f654 ("drm/i915: Disable semaphore busywaits on saturated systems")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>
Cc: Dmitry Ermilov <dmitry.ermilov@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190618074153.16055-8-chris@chris-wilson.co.uk
Maarten needs -rc4 backmerged so he can pull in the fbcon notifier
removal topic branch into drm-misc-next.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
