Track the priority of each request and use it to determine the order in
which we submit requests to the hardware via execlists.
The priority of the request is determined by the user (eventually via
the context) but may be overridden at any time by the driver. When we set
the priority of the request, we bump the priority of all of its
dependencies to match - so that a high priority drawing operation is not
stuck behind a background task.
When the request is ready to execute (i.e. we have signaled the submit
fence following completion of all its dependencies, including third
party fences), we put the request into a priority sorted rbtree to be
submitted to the hardware. If the request is higher priority than all
pending requests, it will be submitted on the next context-switch
interrupt as soon as the hardware has completed the current request. We
do not currently preempt any current execution to immediately run a very
high priority request, at least not yet.
One more limitation, is that this is first implementation is for
execlists only so currently limited to gen8/gen9.
v2: Replace recursive priority inheritance bumping with an iterative
depth-first search list.
v3: list_next_entry() for walking lists
v4: Explain how the dfs solves the recursion problem with PI.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161114204105.29171-8-chris@chris-wilson.co.uk
Defer the transfer from the client's timeline onto the execution
timeline from the point of readiness to the point of actual submission.
For example, in execlists, a request is finally submitted to hardware
when the hardware is ready, and only put onto the hardware queue when
the request is ready. By deferring the transfer, we ensure that the
timeline is maintained in retirement order if we decide to queue the
requests onto the hardware in a different order than fifo.
v2: Rebased onto distinct global/user timeline lock classes.
v3: Play with the position of the spin_lock().
v4: Nesting finally resolved with distinct sw_fence lock classes.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161114204105.29171-4-chris@chris-wilson.co.uk
Though we will have multiple timelines, we still have a single timeline
of execution. This we can use to provide an execution and retirement order
of requests. This keeps tracking execution of requests simple, and vital
for preserving a single waiter (i.e. so that we can order the waiters so
that only the earliest to wakeup need be woken). To accomplish this we
distinguish the seqno used to order requests per-context (external) and
that used internally for execution.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-26-chris@chris-wilson.co.uk
Our timelines are more than just a seqno. They also provide an ordered
list of requests to be executed. Due to the restriction of handling
individual address spaces, we are limited to a timeline per address
space but we use a fence context per engine within.
Our first step to introducing independent timelines per context (i.e. to
allow each context to have a queue of requests to execute that have a
defined set of dependencies on other requests) is to provide a timeline
abstraction for the global execution queue.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-23-chris@chris-wilson.co.uk
Driver accesses the ringbuffer pages, via GMADR BAR, if the pages are
pinned in mappable aperture portion of GGTT and for ringbuffer pages
allocated from Stolen memory, access can only be done through GMADR BAR.
In case of GuC based submission, updates done in ringbuffer via GMADR
may not get committed to memory by the time the Command streamer starts
reading them, resulting in fetching of stale data.
For Host based submission, such problem is not there as the write to Ring
Tail or ELSP register happens from the Host side prior to submission.
Access to any GFX register from CPU side goes to GTTMMADR BAR and Hw already
enforces the ordering between outstanding GMADR writes & new GTTMADR access.
MMIO writes from GuC side do not go to GTTMMADR BAR as GuC communication to
registers within GT is contained within GT, so ordering is not enforced
resulting in a race, which can manifest in form of a hang.
To ensure the flush of in-flight GMADR writes, a POSTING READ is done to
GuC register prior to doorbell ring.
There is already a similar WA in i915_gem_object_flush_gtt_write_domain(),
which takes care of GMADR writes from User space to GEM buffers, but not the
ringbuffer writes from KMD.
This WA is needed on all recent HW.
v2:
- Use POSTING_READ_FW instead of POSTING_READ as GuC register do not lie
in any forcewake domain range and so the overhead of spinlock & search
in the forcewake table is avoidable. (Chris)
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Sagar Arun Kamble <sagar.a.kamble@intel.com>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1477413323-1880-1-git-send-email-akash.goel@intel.com
The GuC log buffer flush work item has to do a register access to send the
ack to GuC and this work item, if not synced before suspend, can potentially
get executed after the GFX device is suspended. This work item function uses
rpm get/put calls around the Hw access, which covers the rpm suspend case
but for system suspend a sync would be required as kernel can potentially
schedule the work items even after some devices, including GFX, have been
put to suspend. But sync has to be done only for the system suspend case,
as sync along with rpm get/put can cause a deadlock for rpm suspend path.
To have the sync, but like a NOOP, for rpm suspend path also this work
item could have been queued from the irq handler only when the device is
runtime active & kept active while that work item is pending or getting
executed but an interrupt can come even after the device is out of use and
so can potentially lead to missing of this work item.
By marking the workqueue, dedicated for handling GuC log buffer flush
interrupts, as freezable we don't have to bother about flushing of this
work item from the suspend hooks, the pending work item if any will be
either executed before the suspend or scheduled later on resume. This way
the handling of log buffer flush work item can be kept same between system
suspend & rpm suspend.
Suggested-by: Imre Deak <imre.deak@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
As per the current i915 Driver load sequence, debugfs registration is done
at the end and so the relay channel debugfs file is also created after that
but the GuC firmware is loaded much earlier in the sequence.
As a result Driver could miss capturing the boot-time logs of GuC firmware
if there are flush interrupts from the GuC side.
Relay has a provision to support early logging where initially only relay
channel can be created, to have buffers for storing logs, and later on
channel can be associated with a debugfs file at appropriate time.
Have availed that, which allows Driver to capture boot time logs also,
which can be collected once Userspace comes up.
v2:
- Remove the couple of FIXMEs, as now the relay channel will be created
early before enabling the flush interrupts, so no possibility of relay
channel pointer being modified & read at the same time from 2 different
execution contexts.
- Rebase.
v3:
- Add a comment to justiy setting 'is_global' before the NULL check on the
parent directory dentry pointer.
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
To ensure that we always get the up-to-date data from log buffer, its
better to access the buffer through an uncached CPU mapping. Also the way
buffer is accessed from GuC & Host side, manually doing cache flush may
not be effective always if cached CPU mapping is used. In order to avoid
any performance drop & have fast reads from the GuC log buffer, used SSE4.1
movntdqa based memcpy function i915_memcpy_from_wc, as copying using
movntqda from WC type memory is almost as fast as reading from WB memory.
This way log buffer sampling time will not get increased and so would be
able to deal with the flush interrupt storm when GuC is generating logs at
a very high rate.
Ideally SSE 4.1 should be present on all chipsets supporting GuC based
submisssions, but if not then logging will not be enabled.
v2: Rebase.
v3: Squash the WC type vmalloc mapping patch with this patch. (Chris)
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
This patch provides debugfs interface i915_guc_output_control for
on the fly enabling/disabling of logging in GuC firmware and controlling
the verbosity level of logs.
The value written to the file, should have bit 0 set to enable logging and
bits 4-7 should contain the verbosity info.
v2: Add a forceful flush, to collect left over logs, on disabling logging.
Useful for Validation.
v3: Besides minor cleanup, implement read method for the debugfs file and
set the guc_log_level to -1 when logging is disabled. (Tvrtko)
v4: Minor cleanup & rebase. (Tvrtko)
v5:
- Lock struct_mutex after the NULL check for guc log buffer vma. (Chris)
- Rebase.
Signed-off-by: Sagar Arun Kamble <sagar.a.kamble@intel.com>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
GuC firmware sends a flush interrupt to Host when the log buffer is half
full and at that time only it updates the log buffer state.
But in certain cases, as described below, it could be useful to have all
that even when log buffer is only partially full. For that there is a force
log buffer flush Host2GuC action supported by GuC firmware.
For Validation requirements, a forceful flush is needed to collect the
left over logs on disabling logging. The same can be done before proceeding
with GPU/GuC reset as there could be some data in log buffer which is yet
to be captured and those logs would be particularly useful to understand
that why the reset was initiated.
Signed-off-by: Sagar Arun Kamble <sagar.a.kamble@intel.com>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
GuC firmware sends an interrupt to flush the log buffer when it becomes
half full, so Driver doesn't really need to sample the complete buffer
and can just copy only the newly written data by GuC into the local
buffer, i.e. as per the read & write pointer values.
Moreover the flush interrupt would generally come for one type of log
buffer, when it becomes half full, so at that time the other 2 types of
log buffer would comparatively have much lesser unread data in them.
In case of overflow reported by GuC, Driver do need to copy the entire
buffer as the whole buffer would contain the unread data.
v2: Rebase.
v3: Fix the blooper of doing the copy twice. (Tvrtko)
v4: Add curlies for 'else' case also, matching the 'if'. (Tvrtko)
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
GuC firmware sends an interrupt to flush the log buffer when it
becomes half full. GuC firmware also tracks how many times the
buffer overflowed.
It would be useful to maintain a statistics of how many flush
interrupts were received and for which type of log buffer,
along with the overflow count of each buffer type.
Augmented i915_log_info debugfs to report back these statistics.
v2:
- Update the logic to detect multiple overflows between the 2
flush interrupts and also log a message for overflow (Tvrtko)
- Track the number of times there was no free sub buffer to capture
the GuC log buffer. (Tvrtko)
v3:
- Fix the printf field width for overflow counter, set it to 10 as per the
max value of u32, which takes 10 digits in decimal form. (Tvrtko)
v4:
- Move the log buffer overflow handling to a new function for better
readability. (Tvrtko)
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
With the addition of new Host2GuC actions related to GuC logging, there
is a need of a lock to serialize them, as they can execute concurrently
with each other and also with other existing actions.
v2: Use mutex in place of spinlock to serialize, as sleep can happen
while waiting for the action's response from GuC. (Tvrtko)
v3: To conform to the general rules, acquire mutex before taking the
forcewake. (Tvrtko)
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Added a new debugfs interface '/sys/kernel/debug/dri/guc_log' for the
User to capture GuC firmware logs. Availed relay framework to implement
the interface, where Driver will have to just use a relay API to store
snapshots of the GuC log buffer in the buffer managed by relay.
The snapshot will be taken when GuC firmware sends a log buffer flush
interrupt and up to four snapshots could be stored in the relay buffer.
The relay buffer will be operated in a mode where it will overwrite the
data not yet collected by User.
Besides mmap method, through which User can directly access the relay
buffer contents, relay also supports the 'poll' method. Through the 'poll'
call on log file, User can come to know whenever a new snapshot of the
log buffer is taken by Driver, so can run in tandem with the Driver and
capture the logs in a sustained/streaming manner, without any loss of data.
v2: Defer the creation of relay channel & associated debugfs file, as
debugfs setup is now done at the end of i915 Driver load. (Chris)
v3:
- Switch to no-overwrite mode for relay.
- Fix the relay sub buffer switching sequence.
v4:
- Update i915 Kconfig to select RELAY config. (TvrtKo)
- Log a message when there is no sub buffer available to capture
the GuC log buffer. (Tvrtko)
- Increase the number of relay sub buffers to 8 from 4, to have
sufficient buffering for boot time logs
v5:
- Fix the alignment, indentation issues and some minor cleanup. (Tvrtko)
- Update the comment to elaborate on why a relay channel has to be
associated with the debugfs file. (Tvrtko)
v6:
- Move the write to 'is_global' after the NULL check on parent directory
dentry pointer. (Tvrtko)
v7: Add a BUG_ON to validate relay buffer allocation size. (Chris)
Testcase: igt/tools/intel_guc_logger
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Sourab Gupta <sourab.gupta@intel.com>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
GuC ukernel sends an interrupt to Host to flush the log buffer
and expects Host to correspondingly update the read pointer
information in the state structure, once it has consumed the
log buffer contents by copying them to a file or buffer.
Even if Host couldn't copy the contents, it can still update the
read pointer so that logging state is not disturbed on GuC side.
v2:
- Use a dedicated workqueue for handling flush interrupt. (Tvrtko)
- Reduce the overall log buffer copying time by skipping the copy of
crash buffer area for regular cases and copying only the state
structure data in first page.
v3:
- Create a vmalloc mapping of log buffer. (Chris)
- Cover the flush acknowledgment under rpm get & put.(Chris)
- Revert the change of skipping the copy of crash dump area, as
not really needed, will be covered by subsequent patch.
v4:
- Destroy the wq under the same condition in which it was created,
pass dev_piv pointer instead of dev to newly added GuC function,
add more comments & rename variable for clarity. (Tvrtko)
v5:
- Allocate & destroy the dedicated wq, for handling flush interrupt,
from the setup/teardown routines of GuC logging. (Chris)
- Validate the log buffer size value retrieved from state structure
and do some minor cleanup. (Tvrtko)
- Fix error/warnings reported by checkpatch. (Tvrtko)
- Rebase.
v6:
- Remove the interrupts_enabled check from guc_capture_logs_work, need
to process that last work item also, queued just before disabling the
interrupt as log buffer flush interrupt handling is a bit different
case where GuC is actually expecting an ACK from host, which should be
provided to keep the logging going.
Sync against the work will be done by caller disabling the interrupt.
- Don't sample the log buffer size value from state structure, directly
use the expected value to move the pointer & do the copy and that cannot
go wrong (out of bounds) as Driver only allocated the log buffer and the
relay buffers. Driver should refrain from interpreting the log packet,
as much possible and let Userspace parser detect the anomaly. (Chris)
v7:
- Use switch statement instead of 'if else' for retrieving the GuC log
buffer size. (Tvrtko)
- Refactored the log buffer copying function and shortended the name of
couple of variables for better readability. (Tvrtko)
v8:
- Make the dedicated wq as a high priority one to further reduce the
turnaround time of handing log buffer flush event from GuC.
Signed-off-by: Sagar Arun Kamble <sagar.a.kamble@intel.com>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
There are certain types of interrupts which Host can receive from GuC.
GuC ukernel sends an interrupt to Host for certain events, like for
example retrieve/consume the logs generated by ukernel.
This patch adds support to receive interrupts from GuC but currently
enables & partially handles only the interrupt sent by GuC ukernel.
Future patches will add support for handling other interrupt types.
v2:
- Use common low level routines for PM IER/IIR programming (Chris)
- Rename interrupt functions to gen9_xxx from gen8_xxx (Chris)
- Replace disabling of wake ref asserts with rpm get/put (Chris)
v3:
- Update comments for more clarity. (Tvrtko)
- Remove the masking of GuC interrupt, which was kept masked till the
start of bottom half, its not really needed as there is only a
single instance of work item & wq is ordered. (Tvrtko)
v4:
- Rebase.
- Rename guc_events to pm_guc_events so as to be indicative of the
register/control block it is associated with. (Chris)
- Add handling for back to back log buffer flush interrupts.
v5:
- Move the read & clearing of register, containing Guc2Host message
bits, outside the irq spinlock. (Tvrtko)
v6:
- Move the log buffer flush interrupt related stuff to the following
patch so as to do only generic bits in this patch. (Tvrtko)
- Rebase.
v7:
- Remove the interrupts_enabled check from gen9_guc_irq_handler, want to
process that last interrupt also before disabling the interrupt, sync
against the work queued by irq handler will be done by caller disabling
the interrupt.
Signed-off-by: Sagar Arun Kamble <sagar.a.kamble@intel.com>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
So far there were 2 fields related to GuC logs in 'intel_guc' structure.
For the support of capturing GuC logs & storing them in a local buffer,
multiple new fields would have to be added. This warrants a separate
structure to contain the fields related to GuC logging state.
Added a new structure 'intel_guc_log' and instance of it inside
'intel_guc' structure.
v2: Rebase.
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
GuC Log buffer allocation was tied up with verbosity level module param
i915.guc_log_level. User would be given a provision to enable firmware
logging at runtime, through a host2guc action, and not necessarily during
Driver load time. But the address of log buffer can be passed only in
init params, at firmware load time, so GuC has to be reset and firmware
needs to be reloaded to pass the log buffer address at runtime.
To avoid reset of GuC & reload of firmware, allocation of log buffer will
be done always but logging would be enabled initially on GuC side based on
the value of module parameter guc_log_level.
v2: Update commit message to describe the constraint with allocation of
log buffer at runtime. (Tvrtko)
v3: Rebase.
Signed-off-by: Sagar Arun Kamble <sagar.a.kamble@intel.com>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
With the possibility of addition of many more number of rings in future,
the drm_i915_private structure could bloat as an array, of type
intel_engine_cs, is embedded inside it.
struct intel_engine_cs engine[I915_NUM_ENGINES];
Though this is still fine as generally there is only a single instance of
drm_i915_private structure used, but not all of the possible rings would be
enabled or active on most of the platforms. Some memory can be saved by
allocating intel_engine_cs structure only for the enabled/active engines.
Currently the engine/ring ID is kept static and dev_priv->engine[] is simply
indexed using the enums defined in intel_engine_id.
To save memory and continue using the static engine/ring IDs, 'engine' is
defined as an array of pointers.
struct intel_engine_cs *engine[I915_NUM_ENGINES];
dev_priv->engine[engine_ID] will be NULL for disabled engine instances.
There is a text size reduction of 928 bytes, from 1028200 to 1027272, for
i915.o file (but for i915.ko file text size remain same as 1193131 bytes).
v2:
- Remove the engine iterator field added in drm_i915_private structure,
instead pass a local iterator variable to the for_each_engine**
macros. (Chris)
- Do away with intel_engine_initialized() and instead directly use the
NULL pointer check on engine pointer. (Chris)
v3:
- Remove for_each_engine_id() macro, as the updated macro for_each_engine()
can be used in place of it. (Chris)
- Protect the access to Render engine Fault register with a NULL check, as
engine specific init is done later in Driver load sequence.
v4:
- Use !!dev_priv->engine[VCS] style for the engine check in getparam. (Chris)
- Kill the superfluous init_engine_lists().
v5:
- Cleanup the intel_engines_init() & intel_engines_setup(), with respect to
allocation of intel_engine_cs structure. (Chris)
v6:
- Rebase.
v7:
- Optimize the for_each_engine_masked() macro. (Chris)
- Change the type of 'iter' local variable to enum intel_engine_id. (Chris)
- Rebase.
v8: Rebase.
v9: Rebase.
v10:
- For index calculation use engine ID instead of pointer based arithmetic in
intel_engine_sync_index() as engine pointers are not contiguous now (Chris)
- For appropriateness, rename local enum variable 'iter' to 'id'. (Joonas)
- Use for_each_engine macro for cleanup in intel_engines_init() and remove
check for NULL engine pointer in cleanup() routines. (Joonas)
v11: Rebase.
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1476378888-7372-1-git-send-email-akash.goel@intel.com
Currently the presumption is that the request construction and its
submission to the GuC are all under the same holding of struct_mutex. We
wish to relax this to separate the request construction and the later
submission to the GuC. This requires us to reserve some space in the
GuC command queue for the future submission. For flexibility to handle
out-of-order request submission we do not preallocate the next slot in
the GuC command queue during request construction, just ensuring that
there is enough space later.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20160909131201.16673-17-chris@chris-wilson.co.uk
Drive final request submission from a callback from the fence. This way
the request is queued until all dependencies are resolved, at which
point it is handed to the backend for queueing to hardware. At this
point, no dependencies are set on the request, so the callback is
immediate.
A side-effect of imposing a heavier-irqsafe spinlock for execlist
submission is that we lose the softirq enabling after scheduling the
execlists tasklet. To compensate, we manually kickstart the softirq by
disabling and enabling the bh around the fence signaling.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: John Harrison <john.c.harrison@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160909131201.16673-14-chris@chris-wilson.co.uk
Update reset path in preparation for engine reset which requires
identification of incomplete requests and associated context and fixing
their state so that engine can resume correctly after reset.
The request that caused the hang will be skipped and head is reset to the
start of breadcrumb. This allows us to resume from where we left-off.
Since this request didn't complete normally we also need to cleanup elsp
queue manually. This is vital if we employ nonblocking request
submission where we may have a web of dependencies upon the hung request
and so advancing the seqno manually is no longer trivial.
ABI: gem_reset_stats / DRM_IOCTL_I915_GET_RESET_STATS
We change the way we count pending batches. Only the active context
involved in the reset is marked as either innocent or guilty, and not
mark the entire world as pending. By inspection this only affects
igt/gem_reset_stats (which assumes implementation details) and not
piglit.
ARB_robustness gives this guide on how we expect the user of this
interface to behave:
* Provide a mechanism for an OpenGL application to learn about
graphics resets that affect the context. When a graphics reset
occurs, the OpenGL context becomes unusable and the application
must create a new context to continue operation. Detecting a
graphics reset happens through an inexpensive query.
And with regards to the actual meaning of the reset values:
Certain events can result in a reset of the GL context. Such a reset
causes all context state to be lost. Recovery from such events
requires recreation of all objects in the affected context. The
current status of the graphics reset state is returned by
enum GetGraphicsResetStatusARB();
The symbolic constant returned indicates if the GL context has been
in a reset state at any point since the last call to
GetGraphicsResetStatusARB. NO_ERROR indicates that the GL context
has not been in a reset state since the last call.
GUILTY_CONTEXT_RESET_ARB indicates that a reset has been detected
that is attributable to the current GL context.
INNOCENT_CONTEXT_RESET_ARB indicates a reset has been detected that
is not attributable to the current GL context.
UNKNOWN_CONTEXT_RESET_ARB indicates a detected graphics reset whose
cause is unknown.
The language here is explicit in that we must mark up the guilty batch,
but is loose enough for us to relax the innocent (i.e. pending)
accounting as only the active batches are involved with the reset.
In the future, we are looking towards single engine resetting (with
minimal locking), where it seems inappropriate to mark the entire world
as innocent since the reset occurred on a different engine. Reducing the
information available means we only have to encounter the pain once, and
also reduces the information leaking from one context to another.
v2: Legacy ringbuffer submission required a reset following hibernation,
or else we restore stale values to the RING_HEAD and walked over
stolen garbage.
v3: GuC requires replaying the requests after a reset.
v4: Restore engine IRQ after reset (so waiters will be woken!)
Rearm hangcheck if resetting with a waiter.
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Arun Siluvery <arun.siluvery@linux.intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160909131201.16673-13-chris@chris-wilson.co.uk
Where we're going to continue regardless of the problem, rather than
fail, then the message should be a WARNing rather than an ERROR.
Signed-off-by: Dave Gordon <david.s.gordon@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Rather than walk the full array of engines checking whether each is in
the mask in turn, we can use the mask to jump to the right engines. This
should quicker for a sparse array of engines or mask, whilst generating
smaller code:
text data bss dec hex filename
1251010 4579 800 1256389 132bc5 drivers/gpu/drm/i915/i915.ko
1250530 4579 800 1255909 1329e5 drivers/gpu/drm/i915/i915.ko
The downside is that we have to pass in a temporary, alas no C99
iterators yet.
[P.S. Joonas doesn't like having to pass extra temporaries into the
macro, and even less that I called them tmp. As yet, we haven't found a
macro that avoids passing in a temporary that is smaller. We probably
will get C99 iterators first!]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160827075401.16470-2-chris@chris-wilson.co.uk
This little helper only exists to safely discard the upper unused 32bits
of the general 64-bit VMA address - as we know that all Global GTT
currently are less than 4GiB in size and so that the upper bits must be
zero. In many places, we use a u32 for the global GTT offset and we want
to document where we are discarding the full VMA offset.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-28-git-send-email-chris@chris-wilson.co.uk
Use the GGTT VMA as the primary cookie for handing ring objects as
the most common action upon the ring is mapping and unmapping which act
upon the VMA itself. By restructuring the code to work with the ring
VMA, we can shrink the code and remove a few cycles from context pinning.
v2: Move the flush of the object back to before the first pin. We use
the am-I-bound? query to only have to check the flush on the first
bind and so avoid stalling on active rings.
Lots of little renames and small hoops.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-18-git-send-email-chris@chris-wilson.co.uk
When working with contexts, we most frequently want the GGTT VMA for the
context state, first and foremost. Since the object is available via the
VMA, we need only then store the VMA.
v2: Formatting tweaks to debugfs output, restored some comments removed
in the next patch
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-15-git-send-email-chris@chris-wilson.co.uk
Since the guc allocates and pins and object into the GGTT for its usage,
it is more natural to use that pinned VMA as our resource cookie.
v2: Embrace naming tautology
v3: Rewrite comments for guc_allocate_vma()
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-12-git-send-email-chris@chris-wilson.co.uk
Now that host structures are indexed by host engine-id rather than
guc_id, we can usefully convert some for_each_engine() loops to use
for_each_engine_id() and avoid multiple dereferences of engine->id.
Also a few related tweaks to cache structure members locally wherever
they're used more than once or twice, hopefully eliminating memory
references.
Signed-off-by: Dave Gordon <david.s.gordon@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
The Context Descriptor passed by the kernel to the GuC contains a field
specifying which engine(s) the context will use. Historically, this was
always set to "all of them", but if we had a separate client for each
engine, we could be more precise, and set only the bit for the engine
that the client was associated with. So this patch enables this usage,
in preparation for having multiple clients, though at this point there
is still only a single client used for all supported engines.
Signed-off-by: Dave Gordon <david.s.gordon@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
We have essentially the same code in each of two different
loops, so we can refactor it into a little helper function.
This also reduces the amount of work done during startup,
as we now only reprogram h/w found to be in a state other
than that expected, and so avoid the overhead of setting
doorbell registers to the state they're already in.
Signed-off-by: Dave Gordon <david.s.gordon@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
guc_init_doorbell_hw() borrows the (currently single) GuC client to use
in reinitialising ALL the doorbell registers (as the hardware doesn't
reset them when the GuC is reset). As a prerequisite for accommodating
multiple clients, it should only reset doorbells that are supposed to be
disabled, avoiding those that are marked as in use by any client.
Signed-off-by: Dave Gordon <david.s.gordon@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Since i915_gem_obj_ggtt_pin() is an idiom breaking curry function for
i915_gem_object_ggtt_pin(), spare us the confusion and remove it.
Removing it now simplifies later patches to change the i915_vma_pin()
(and friends) interface.
v2: Add a redundant GEM_BUG_ON(!view) to
i915_gem_obj_lookup_or_create_ggtt_vma()
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1470324762-2545-18-git-send-email-chris@chris-wilson.co.uk
dma-buf provides a generic fence class for interoperation between
drivers. Internally we use the request structure as a fence, and so with
only a little bit of interfacing we can rebase those requests on top of
dma-buf fences. This will allow us, in the future, to pass those fences
back to userspace or between drivers.
v2: The fence_context needs to be globally unique, not just unique to
this device.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1469002875-2335-4-git-send-email-chris@chris-wilson.co.uk
Rather than using wait_for_atomic() when chacking for a response from
the GuC, we can get the effect of a hybrid spin/sleep wait by breaking
it into two stages. First, spin-wait for up to 10us to minimise latency
for "quick" commands; then, if that times out, sleep-wait for up 10ms
(the maximum allowed for a "slow" command).
Being able to do this depends on the recent patch
18f4b84 drm/i915: Use atomic waits for short non-atomic ones
and is similar to the hybrid approach in
1758b90 drm/i915: Use a hybrid scheme for fast register waits
(although we can't use that as-is, because that interface doesn't quite
match what we need here).
Signed-off-by: Dave Gordon <david.s.gordon@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1467815411-21756-1-git-send-email-david.s.gordon@intel.com
Since drm_i915_private is now a subclass of drm_device we do not need to
chase the drm_i915_private->dev backpointer and can instead simply
access drm_i915_private->drm directly.
text data bss dec hex filename
1068757 4565 416 1073738 10624a drivers/gpu/drm/i915/i915.ko
1066949 4565 416 1071930 105b3a drivers/gpu/drm/i915/i915.ko
Created by the coccinelle script:
@@
struct drm_i915_private *d;
identifier i;
@@
(
- d->dev->i
+ d->drm.i
|
- d->dev
+ &d->drm
)
and for good measure the dev_priv->dev backpointer was removed entirely.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1467711623-2905-4-git-send-email-chris@chris-wilson.co.uk
Since we now subclass struct drm_device, we can save pointer dances by
noting the equivalence of struct drm_device and struct drm_i915_private,
i.e. by using to_i915().
text data bss dec hex filename
1073824 4562 416 1078802 107612 drivers/gpu/drm/i915/i915.ko
1068976 4562 416 1073954 106322 drivers/gpu/drm/i915/i915.ko
Created by the coccinelle script:
@@
expression E;
identifier p;
@@
- struct drm_i915_private *p = E->dev_private;
+ struct drm_i915_private *p = to_i915(E);
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Dave Gordon <david.s.gordon@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1467628477-25379-1-git-send-email-chris@chris-wilson.co.uk
No need for local struct drm_device * since dev_priv is the
correct thing to pass in to NEEDS_WaRsDisableCoarsePowerGating
anyway. Changed the macro definition for the latter to reflect
that as well.
v2: Alignment bikeshed.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Dave Gordon <david.s.gordon@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1466518034-24838-1-git-send-email-tvrtko.ursulin@linux.intel.com
The ONLY places that guc_id (aka hw_id) should be used are those where
the value or address is determined by and shared with the GuC firmware;
specifically, when filling in the GuC-context-descriptor or the GuC
addon data, or putting an entry in the GuC's work queue.
It need not (and therefore should not) be used to index GuC statistics
or similar host-managed tracking data. In particular, i915_guc_submit()
produces (and debugfs decodes) GuC submission statistics which should be
indexed by driver-engine-id rather then guc-engine-id.
Signed-off-by: Dave Gordon <david.s.gordon@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1466432287-5799-1-git-send-email-david.s.gordon@intel.com
During a hibernate/resume cycle, the whole system is reset, including
the GuC and the doorbell hardware. Then the system is booted up, drivers
are loaded, etc -- the GuC firmware may be loaded and set running at
this point. But then, the booted kernel is replaced by the hibernated
image, and this resumed kernel will also try to reload the GuC firmware
(which will fail). To recover, we reset the GuC and try again (which
should work). But this GuC reset doesn't also reset the doorbell
hardware, so it can be left in a state inconsistent with that assumed
by the driver and/or the newly-loaded GuC firmware.
It would be better if the GuC reset also cleared all doorbell state,
but that's not how the hardware currently works; also, the driver cannot
directly reprogram the doorbell hardware (only the GuC can do that).
So this patch cycles through all doorbells, assigning and releasing each
in turn, so that all the doorbell hardware is left in a consistent
state, no matter how it was programmed by the previously-running kernel
and/or GuC firmware.
v2: don't use kmap_atomic() now that client page 0 is kept mapped.
Signed-off-by: Dave Gordon <david.s.gordon@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1465837054-16245-2-git-send-email-david.s.gordon@intel.com
This version doesn't update the doorbell bitmap, as that will
be done when the selected doorbell is associated with a client.
The call is now slightly earlier, just on the general principle
that potentially-failing operations should be done as early as
possible, to eliminate late failures and simplify recovery.
Suggested-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Dave Gordon <david.s.gordon@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Chris Wilson