1config DRM_I915_USERFAULT_AUTOSUSPEND 2 int "Runtime autosuspend delay for userspace GGTT mmaps (ms)" 3 default 250 # milliseconds 4 help 5 On runtime suspend, as we suspend the device, we have to revoke 6 userspace GGTT mmaps and force userspace to take a pagefault on 7 their next access. The revocation and subsequent recreation of 8 the GGTT mmap can be very slow and so we impose a small hysteris 9 that complements the runtime-pm autosuspend and provides a lower 10 floor on the autosuspend delay. 11 12 May be 0 to disable the extra delay and solely use the device level 13 runtime pm autosuspend delay tunable. 14 15config DRM_I915_HEARTBEAT_INTERVAL 16 int "Interval between heartbeat pulses (ms)" 17 default 2500 # milliseconds 18 help 19 The driver sends a periodic heartbeat down all active engines to 20 check the health of the GPU and undertake regular house-keeping of 21 internal driver state. 22 23 This is adjustable via 24 /sys/class/drm/card?/engine/*/heartbeat_interval_ms 25 26 May be 0 to disable heartbeats and therefore disable automatic GPU 27 hang detection. 28 29config DRM_I915_PREEMPT_TIMEOUT 30 int "Preempt timeout (ms, jiffy granularity)" 31 default 640 # milliseconds 32 help 33 How long to wait (in milliseconds) for a preemption event to occur 34 when submitting a new context via execlists. If the current context 35 does not hit an arbitration point and yield to HW before the timer 36 expires, the HW will be reset to allow the more important context 37 to execute. 38 39 This is adjustable via 40 /sys/class/drm/card?/engine/*/preempt_timeout_ms 41 42 May be 0 to disable the timeout. 43 44 The compiled in default may get overridden at driver probe time on 45 certain platforms and certain engines which will be reflected in the 46 sysfs control. 47 48config DRM_I915_MAX_REQUEST_BUSYWAIT 49 int "Busywait for request completion limit (ns)" 50 default 8000 # nanoseconds 51 help 52 Before sleeping waiting for a request (GPU operation) to complete, 53 we may spend some time polling for its completion. As the IRQ may 54 take a non-negligible time to setup, we do a short spin first to 55 check if the request will complete in the time it would have taken 56 us to enable the interrupt. 57 58 This is adjustable via 59 /sys/class/drm/card?/engine/*/max_busywait_duration_ns 60 61 May be 0 to disable the initial spin. In practice, we estimate 62 the cost of enabling the interrupt (if currently disabled) to be 63 a few microseconds. 64 65config DRM_I915_STOP_TIMEOUT 66 int "How long to wait for an engine to quiesce gracefully before reset (ms)" 67 default 100 # milliseconds 68 help 69 By stopping submission and sleeping for a short time before resetting 70 the GPU, we allow the innocent contexts also on the system to quiesce. 71 It is then less likely for a hanging context to cause collateral 72 damage as the system is reset in order to recover. The corollary is 73 that the reset itself may take longer and so be more disruptive to 74 interactive or low latency workloads. 75 76 This is adjustable via 77 /sys/class/drm/card?/engine/*/stop_timeout_ms 78 79config DRM_I915_TIMESLICE_DURATION 80 int "Scheduling quantum for userspace batches (ms, jiffy granularity)" 81 default 1 # milliseconds 82 help 83 When two user batches of equal priority are executing, we will 84 alternate execution of each batch to ensure forward progress of 85 all users. This is necessary in some cases where there may be 86 an implicit dependency between those batches that requires 87 concurrent execution in order for them to proceed, e.g. they 88 interact with each other via userspace semaphores. Each context 89 is scheduled for execution for the timeslice duration, before 90 switching to the next context. 91 92 This is adjustable via 93 /sys/class/drm/card?/engine/*/timeslice_duration_ms 94 95 May be 0 to disable timeslicing. 96