- 22 Nov, 2016 14 commits
-
-
Chris Wilson authored
Since the submit/execute split in commit d55ac5bf ("drm/i915: Defer transfer onto execution timeline to actual hw submission") the global seqno advance was deferred until the submit_request callback. After wedging the GPU, we were installing a nop_submit_request handler (to avoid waking up the dead hw) but I had missed converting this over to the new scheme. Under the new scheme, we have to explicitly call i915_gem_submit_request() from the submit_request handler to mark the request as on the hardware. If we don't the request is always pending, and any waiter will continue to wait indefinitely and hangcheck will not be able to resolve the lockup. References: https://bugs.freedesktop.org/show_bug.cgi?id=98748 Testcase: igt/gem_eio/in-flight Fixes: d55ac5bf ("drm/i915: Defer transfer onto execution timeline to actual hw submission") Signed-off-by:
Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by:
Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20161122144121.7379-3-chris@chris-wilson.co.uk
-
Chris Wilson authored
If the gpu reset fails and the machine is terminally wedged, further hangchecks achieve nothing but noise. Disable them, with a corollary that we re-enable hangchecking after a successful GPU reset in case the user is artificially bringing the machine back to life through the debug interface. Signed-off-by:
Mika Kuoppala <mika.kuoppala@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20161122144121.7379-2-chris@chris-wilson.co.uk
-
Chris Wilson authored
When a user context is closed, it's file_priv backpointer is replaced by ERR_PTR(-EBADF); be careful not to chase this invalid pointer after a hang and a GPU reset. Signed-off-by:
-
Robert Bragg authored
In particular this tries to capture for posterity some of the early challenges we had with using the core perf infrastructure in case we ever want to revisit adapting perf for device metrics. Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
Robert Bragg <robert@sixbynine.org> Reviewed-by:
Matthew Auld <matthew.auld@intel.com> Reviewed-by:
Sourab Gupta <sourab.gupta@intel.com> Signed-off-by:
Daniel Vetter <daniel.vetter@ffwll.ch> Link: http://patchwork.freedesktop.org/patch/msgid/20161107194957.3385-12-robert@sixbynine.org
-
Robert Bragg authored
This adds 'compute', 'compute extended', 'memory reads', 'memory writes' and 'sampler balance' metric sets for Haswell. The code is auto generated from an XML description of metric sets, currently maintained in gputop, ref: https://github.com/rib/gputop > gputop-data/oa-*.xml > scripts/i915-perf-kernelgen.py $ make -C gputop-data -f Makefile.xml Signed-off-by:
-
Robert Bragg authored
The maximum OA sampling frequency is now configurable via a dev.i915.oa_max_sample_rate sysctl parameter. Following the precedent set by perf's similar kernel.perf_event_max_sample_rate the default maximum rate is 100000Hz Signed-off-by:
-
Robert Bragg authored
Consistent with the kernel.perf_event_paranoid sysctl option that can allow non-root users to access system wide cpu metrics, this can optionally allow non-root users to access system wide OA counter metrics from Gen graphics hardware. Signed-off-by:
-
Robert Bragg authored
Each metric set is given a sysfs entry like: /sys/class/drm/card0/metrics/<guid>/id This allows userspace to enumerate the specific sets that are available for the current system. The 'id' file contains an unsigned integer that can be used to open the associated metric set via DRM_IOCTL_I915_PERF_OPEN. The <guid> is a globally unique ID for a specific OA unit register configuration that can be reliably used by userspace as a key to lookup corresponding counter meta data and normalization equations. The guid registry is currently maintained as part of gputop along with the XML metric set descriptions and code generation scripts, ref: https://github.com/rib/gputop > gputop-data/guids.xml > scripts/update-guids.py > gputop-data/oa-*.xml > scripts/i915-perf-kernelgen.py $ make -C gputop-data -f Makefile.xml SYSFS=1 WHITELIST=RenderBasic Signed-off-by:
-
Robert Bragg authored
Gen graphics hardware can be set up to periodically write snapshots of performance counters into a circular buffer via its Observation Architecture and this patch exposes that capability to userspace via the i915 perf interface. v2: Make sure to initialize ->specific_ctx_id when opening, without relying on _pin_notify hook, in case ctx already pinned. v3: Revert back to pinning ctx upfront when opening stream, removing need to hook in to pinning and to update OACONTROL on the fly. Signed-off-by:
Zhenyu Wang <zhenyuw@linux.intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by:
-
Robert Bragg authored
Adds a static OA unit, MUX + B Counter configuration for basic render metrics on Haswell. This is auto generated from an XML description of metric sets, currently maintained in gputop, ref: https://github.com/rib/gputop > gputop-data/oa-*.xml > scripts/i915-perf-kernelgen.py $ make -C gputop-data -f Makefile.xml SYSFS=0 WHITELIST=RenderBasic Signed-off-by:
-
Robert Bragg authored
Being able to program OACONTROL from a non-privileged batch buffer is not sufficient to be able to configure the OA unit. This was originally allowed to help enable Mesa to expose OA counters via the INTEL_performance_query extension, but the current implementation based on programming OACONTROL via a batch buffer isn't able to report useable data without a more complete OA unit configuration. Mesa handles the possibility that writes to OACONTROL may not be allowed and so only advertises the extension after explicitly testing that a write to OACONTROL succeeds. Based on this; removing OACONTROL from the whitelist should be ok for userspace. Removing this simplifies adding a new kernel api for configuring the OA unit without needing to consider the possibility that userspace might trample on OACONTROL state which we'd like to start managing within the kernel instead. In particular running any Mesa based GL application currently results in clearing OACONTROL when initializing which would disable the capturing of metrics. v2: This bumps the command parser version from 8 to 9, as the change is visible to userspace. Signed-off-by: -
Robert Bragg authored
check_cmd() is checking whether a command adheres to certain restrictions that ensure it's safe to execute within a privileged batch buffer. Returning false implies a privilege problem, not that the command is invalid. The distinction makes the difference between allowing the buffer to be executed as an unprivileged batch buffer or returning an EINVAL error to userspace without executing anything. In a case where userspace may want to test whether it can successfully write to a register that needs privileges the distinction may be important and an EINVAL error may be considered fatal. In particular this is currently true for Mesa, which includes a test for whether OACONTROL can be written too, but Mesa treats any error when flushing a batch buffer as fatal, calling exit(1). As it is currently Mesa can gracefully handle a failure to write to OACONTROL if the command parser is disabled, but if we were to remove OACONTROL from the parser's whitelist then the returned EINVAL would break Mesa applications as they attempt an OACONTROL write. This bumps the command parser version from 7 to 8, as the change is visible to userspace. Signed-off-by:
-
Robert Bragg authored
OACONTROL changes quite a bit for gen8, with some bits split out into a per-context OACTXCONTROL register. Rename now before adding more gen7 OA registers Signed-off-by:
-
Robert Bragg authored
Adds base i915 perf infrastructure for Gen performance metrics. This adds a DRM_IOCTL_I915_PERF_OPEN ioctl that takes an array of uint64 properties to configure a stream of metrics and returns a new fd usable with standard VFS system calls including read() to read typed and sized records; ioctl() to enable or disable capture and poll() to wait for data. A stream is opened something like: uint64_t properties[] = { /* Single context sampling */ DRM_I915_PERF_PROP_CTX_HANDLE, ctx_handle, /* Include OA reports in samples */ DRM_I915_PERF_PROP_SAMPLE_OA, true, /* OA unit configuration */ DRM_I915_PERF_PROP_OA_METRICS_SET, metrics_set_id, DRM_I915_PERF_PROP_OA_FORMAT, report_format, DRM_I915_PERF_PROP_OA_EXPONENT, period_exponent, }; struct drm_i915_perf_open_param parm = { .flags = I915_PERF_FLAG_FD_CLOEXEC | I915_PERF_FLAG_FD_NONBLOCK | I915_PERF_FLAG_DISABLED, .properties_ptr = (uint64_t)properties, .num_properties = sizeof(properties) / 16, }; int fd = drmIoctl(drm_fd, DRM_IOCTL_I915_PERF_OPEN, ¶m); Records read all start with a common { type, size } header with DRM_I915_PERF_RECORD_SAMPLE being of most interest. Sample records contain an extensible number of fields and it's the DRM_I915_PERF_PROP_SAMPLE_xyz properties given when opening that determine what's included in every sample. No specific streams are supported yet so any attempt to open a stream will return an error. v2: use i915_gem_context_get() - Chris Wilson v3: update read() interface to avoid passing state struct - Chris Wilson fix some rebase fallout, with i915-perf init/deinit v4: s/DRM_IORW/DRM_IOW/ - Emil Velikov Signed-off-by:
-
- 21 Nov, 2016 11 commits
-
-
Mika Kuoppala authored
Bannable property, banned status, guilty and active counts are properties of i915_gem_context. Make them so. v2: rebase Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by:
-
Mika Kuoppala authored
If we have a bad client submitting unfavourably across different contexts, creating new ones, the per context scoring of badness doesn't remove the root cause, the offending client. To counter, keep track of per client context bans. Deny access if client is responsible for more than 3 context bans in it's lifetime. v2: move ban check to context create ioctl (Chris) v3: add commentary about hangs needed to reach client ban (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by:
-
Mika Kuoppala authored
Now when driver has per context scoring of 'hanging badness' and also subsequent hangs during short windows are allowed, if there is progress made in between, it does not make sense to expose a ban timing window as a context parameter anymore. Let the scoring be the sole indicator for ban policy and substitute ban period context parameter as a boolean to get/set context bannable property. v2: allow non root to opt into being banned (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Suggested-by:
-
Mika Kuoppala authored
As hangcheck score was removed, the active decay of score was removed also. This removed feature for hangcheck to detect if the gpu client was accidentally or maliciously causing intermittent hangs. Reinstate the scoring as a per context property, so that if one context starts to act unfavourably, ban it. v2: ban_period_secs as a gate to score check (Chris) v3: decay in proper spot. scores as tunables (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Mika Kuoppala authored
Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: -
Mika Kuoppala authored
In order to simplify hangcheck state keeping, split hangcheck per engine loop in three phases: state load, action, state save. Add few more hangcheck actions to separate between seqno, head and subunit movements. This helps to gather all the hangcheck actions under a single switch umbrella. Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Mika Kuoppala authored
To find out what firmware we actually loaded (from dmesg) the explicit 'dmc' and 'firmware' are missing from the info printout. Add them. Cc: Imre Deak <imre.deak@intel.com> Reviewed-by:
Imre Deak <imre.deak@intel.com> Signed-off-by:
-
A.Sunil Kamath authored
Better to use num_scaler itself while printing scaler_info. This fixes a bug of printing information for the missing second scaler on pipe C for SKL platform. Signed-off-by:
A.Sunil Kamath <sunil.kamath@intel.com> Reviewed-by:
Ander Conselvan de Oliveira <conselvan2@gmail.com> Signed-off-by:
Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1479664226-22307-1-git-send-email-sunil.kamath@intel.com
-
Chris Wilson authored
When unloading the module, it is expected that we have finished executing all requests and so the signal threads should be idle. Add a warning in case there are any residual requests in the signaler rbtrees at that point. v2: We can also warn if there are any waiters Signed-off-by:
-
Chris Wilson authored
For user actions, such as the context ioctls, we prefer to use DRM_DEBUG rather than DRM_DEBUG_DRIVER as currently used. Signed-off-by:
Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
-
Daniel Vetter authored
Signed-off-by:
-
- 18 Nov, 2016 9 commits
-
-
Chris Wilson authored
If the LLC is coherent with the object, we do not need to worry about whether main memory and cache mismatch when we hand the object back to the system. Signed-off-by:
Ville Syrjälä <ville.syrjala@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20161118211747.25197-2-chris@chris-wilson.co.uk
-
Chris Wilson authored
Currently we only clflush the scanout if it is in the CPU domain. Also flush if we have a pending CPU clflush. We also want to treat the dirtyfb path similar, and flush any pending writes there as well. v2: Only send the fb flush message if flushing the dirt on flip v3: Make flush-for-flip and dirtyfb look more alike since they serve similar roles as end-of-frame marker. Signed-off-by:
-
Chris Wilson authored
On the DMA mapping error path, sg may be NULL (it has already been marked as the last scatterlist entry), and we should avoid dereferencing it again. Reported-by:
Dan Carpenter <dan.carpenter@oracle.com> Fixes: e2273302 ("drm/i915: avoid leaking DMA mappings") Signed-off-by:
-
Chris Wilson authored
Trying to chase an impossible bug (ivb): [ 207.765411] [drm:i915_reset_and_wakeup [i915]] resetting chip [ 207.765734] [drm:i915_gem_reset [i915]] resetting render ring to restart from tail of request 0x4ee834 [ 207.765791] [drm:intel_print_rc6_info [i915]] Enabling RC6 states: RC6 on RC6p on RC6pp off [ 207.767213] [drm:intel_guc_setup [i915]] GuC fw status: path (null), fetch NONE, load NONE [ 207.767515] kernel BUG at drivers/gpu/drm/i915/i915_gem_request.c:203! [ 207.767551] invalid opcode: 0000 [#1] PREEMPT SMP [ 207.767576] Modules linked in: snd_hda_intel i915 cdc_ncm usbnet mii x86_pkg_temp_thermal coretemp crct10dif_pclmul crc32_pclmul ghash_clmulni_intel lpc_ich snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic snd_hda_codec snd_hwdep snd_hda_core mei_me mei snd_pcm sdhci_pci sdhci mmc_core e1000e ptp pps_core [last unloaded: i915] [ 207.767808] CPU: 3 PID: 8855 Comm: gem_ringfill Tainted: G U 4.9.0-rc5-CI-Patchwork_3052+ #1 [ 207.767854] Hardware name: LENOVO 2356GCG/2356GCG, BIOS G7ET31WW (1.13 ) 07/02/2012 [ 207.767894] task: ffff88012c82a740 task.stack: ffffc9000383c000 [ 207.767927] RIP: 0010:[<ffffffffa00a0a3a>] [<ffffffffa00a0a3a>] i915_gem_request_retire+0x2a/0x4b0 [i915] [ 207.767999] RSP: 0018:ffffc9000383fb20 EFLAGS: 00010293 [ 207.768027] RAX: 00000000004ee83c RBX: ffff880135dcb480 RCX: 00000000004ee83a [ 207.768062] RDX: ffff88012fea42a8 RSI: 0000000000000001 RDI: ffff88012c82af68 [ 207.768095] RBP: ffffc9000383fb48 R08: 0000000000000000 R09: 0000000000000000 [ 207.768129] R10: 0000000000000000 R11: 0000000000000000 R12: ffff880135dcb480 [ 207.768163] R13: ffff88012fea42a8 R14: 0000000000000000 R15: 00000000000001d8 [ 207.768200] FS: 00007f955f658740(0000) GS:ffff88013e2c0000(0000) knlGS:0000000000000000 [ 207.768239] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 207.768258] CR2: 0000555899725930 CR3: 00000001316f6000 CR4: 00000000001406e0 [ 207.768286] Stack: [ 207.768299] ffff880135dcb480 ffff880135dcbe00 ffff88012fea42a8 0000000000000000 [ 207.768350] 00000000000001d8 ffffc9000383fb70 ffffffffa00a1339 0000000000000000 [ 207.768402] ffff88012f296c88 00000000000003f0 ffffc9000383fbb0 ffffffffa00b582d [ 207.768453] Call Trace: [ 207.768493] [<ffffffffa00a1339>] i915_gem_request_retire_upto+0x49/0x90 [i915] [ 207.768553] [<ffffffffa00b582d>] intel_ring_begin+0x15d/0x2d0 [i915] [ 207.768608] [<ffffffffa00b59cb>] intel_ring_alloc_request_extras+0x2b/0x40 [i915] [ 207.768667] [<ffffffffa00a2fd9>] i915_gem_request_alloc+0x359/0x440 [i915] [ 207.768723] [<ffffffffa008bd03>] i915_gem_do_execbuffer.isra.15+0x783/0x1a10 [i915] [ 207.768766] [<ffffffff811a6a2e>] ? __might_fault+0x3e/0x90 [ 207.768816] [<ffffffffa008d380>] i915_gem_execbuffer2+0xc0/0x250 [i915] [ 207.768854] [<ffffffff815532a6>] drm_ioctl+0x1f6/0x480 [ 207.768900] [<ffffffffa008d2c0>] ? i915_gem_execbuffer+0x330/0x330 [i915] [ 207.768939] [<ffffffff81202f6e>] do_vfs_ioctl+0x8e/0x690 [ 207.768972] [<ffffffff818193ac>] ? retint_kernel+0x2d/0x2d [ 207.769004] [<ffffffff810d6ef2>] ? trace_hardirqs_on_caller+0x122/0x1b0 [ 207.769039] [<ffffffff812035ac>] SyS_ioctl+0x3c/0x70 [ 207.769068] [<ffffffff818189ae>] entry_SYSCALL_64_fastpath+0x1c/0xb1 [ 207.769103] Code: 90 55 48 89 e5 41 57 41 56 41 55 41 54 49 89 fc 53 8b 35 fa 7b e1 e1 85 f6 0f 85 55 03 00 00 41 8b 84 24 80 02 00 00 85 c0 75 02 <0f> 0b 49 8b 94 24 a8 00 00 00 48 8b 8a e0 01 00 00 8b 89 c0 00 [ 207.769400] RIP [<ffffffffa00a0a3a>] i915_gem_request_retire+0x2a/0x4b0 [i915] [ 207.769463] RSP <ffffc9000383fb20> Let's add a couple more BUG_ONs before this to ascertain that the request did make it to hardware. The impossible part of this stacktrace is that request must have been considered completed by the i915_request_wait() before we tried to retire it. Signed-off-by:
-
Matthew Auld authored
When gathering the pages from our backing storage we expect get_pages() to either give us our sg_table or an err ptr. However when gathering our fake pages for stolen memory we may return NULL in the event of a failure. To prevent any funny business we should therefore return the proper err ptr value. Fixes: 03ac84f1 ("drm/i915: Pass around sg_table to get_pages/put_pages backend") Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Maarten Lankhorst authored
Signed-off-by:
Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1478609742-13603-12-git-send-email-maarten.lankhorst@linux.intel.com Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> #irc
-
Chris Wilson authored
Since we can retire requests from multiple paths, we cannot assume that i915_gem_retire_requests() is the sole path on which we can transition to gt.active_requests == 0. A consequence of this is that we would skip the function if we had already retired all the requests and not scheduled the idle worker. This is fallout from changing the routine from considering active_engines (for which it was the only consumer) to active_requests. v2: Move kicking the idle working to i915_gem_request_retire() otherwise we could postpone the idle callback everytime we called retire_requests even though we did no work. v3: We only need to move the idle work kicking! v4: Drop the BUG_ON(!awake) as we may be called from the shrinker in the middle of constructing a request before we have marked the device awake. v5: Add a BUG_ON() for active_requests underflow upon retirement (Joonas) Fixes: 28176ef4 ("drm/i915: Reserve space in the global seqno during request allocation") Signed-off-by:
-
Chris Wilson authored
I tried to avoid having to track the write for every VMA by only tracking writes to the ggtt. However, for the purposes of frontbuffer tracking this is insufficient as we need to invalidate around writes not just to the the ggtt but all aliased ppgtt views of the framebuffer. By moving the critical section to the object and only doing so for framebuffer writes we can reduce the tracking even further by only watching framebuffers and not vma. Signed-off-by:
Paulo Zanoni <paulo.r.zanoni@intel.com> Reviewed-by:
-
Tvrtko Ursulin authored
Otherwise it is just an useless empty line. Signed-off-by:
-
- 17 Nov, 2016 6 commits
-
-
Matthew Auld authored
We need to clean up the global_timeline in i915_gem_load_cleanup. v2: don't forget about the struct_mutex, and also WARN_ON if we have any remaining timelines before purging the global_timeline. v3: it might be a good idea to first remove the global_timeline...duh! Fixes: 73cb9701 ("drm/i915: Combine seqno + tracking into a global timeline struct") Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Matthew Auld authored
We already have an i915_address_space_init, so for symmetry we should also have a _fini, plus we already open code it twice. This then also fixes a bug where we leak the timeline for the ggtt vm. v2: don't forget about the struct_mutex for the ggtt path. Fixes: 80b204bc ("drm/i915: Enable multiple timelines") Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Chris Wilson authored
We should never be called via obj->ops->release() on anything other than a fully formed stolen object, so raise that to an assert. In the process tidy up a comment and variable no longer used outside of a conditional BUG. Reported-by:
kbuild test robot <fengguang.wu@intel.com> Signed-off-by:
-
Tvrtko Ursulin authored
We have to make sure there are no holes in the table in Gen9. Signed-off-by:
-
Tvrtko Ursulin authored
Similar to existing yesno and onoff and use it throughout the code. Signed-off-by:
-
Tvrtko Ursulin authored
Signed-off-by:
-
