Disable Intel GPU power management in low-latency/realtime modes

[rtzoeller]

The Intel i915 driver can contribute considerable latency when doing power management, which is visible with rtla timerlat top.

We've observed this issue on various Tiger Lake systems which are running headlessly, specifically with Xeon W-11865MLE and i5-11500HE CPUs.

Disabling power management, e.g. via echo on > /sys/bus/pci/devices/0000\:00\:02.0/power/control or using powertop, removes the latency spikes. While we can work around the issue, it'd be nice if tuned handled this automatically, especially given the magnitude of the spikes.

`rtla timerlat` output with stack trace

                                     Timer Latency
  0 00:00:06   |          IRQ Timer Latency (us)        |         Thread Timer Latency (us)
CPU COUNT      |      cur       min       avg       max |      cur       min       avg       max
  0 #5823      |       24         0         1        24 |     1035         3         7      1035
  1 #5824      |        1         0         1         4 |        6         2         6        11
  2 #5824      |        1         0         1         5 |        6         2         6        10
  3 #5824      |        1         0         1         5 |        6         2         6        13
  4 #5823      |        1         0         1         5 |        6         2         6        10
  5 #5823      |        1         0         1         5 |        7         2         6        13
  6 #5823      |        1         0         1         5 |        6         2         6        15
  7 #5823      |        1         0         1         5 |        7         2         6        13
  8 #5823      |       37         0         1        37 |       55         3         7        55
  9 #5823      |        1         0         1         5 |        6         2         6        13
 10 #5823      |        1         0         1         4 |        6         2         6        13
 11 #5823      |        1         0         1         5 |        7         2         6        13
 12 #5824      |        1         0         1         5 |        6         2         6        13
 13 #5824      |        1         0         1         5 |        6         2         6        14
 14 #5824      |        1         0         1         5 |        6         2         6        13
 15 #5824      |        1         0         1         5 |        6         2         6        15
---------------|----------------------------------------|---------------------------------------
ALL #93175  e0 |                  0         1        37 |                  2         6      1035
rtla timerlat hit stop tracing
## CPU 0 hit stop tracing, analyzing it ##
  IRQ handler delay:                                        23.99 us (2.32 %)
  IRQ latency:                                              24.71 us
  Timerlat IRQ duration:                                    10.37 us (1.00 %)
  Blocking thread:                                         997.61 us (96.30 %)
                     kworker/0:2:9651                      997.61 us
    Blocking thread stack trace
                -> timerlat_irq
                -> __hrtimer_run_queues
                -> hrtimer_interrupt
                -> __sysvec_apic_timer_interrupt
                -> sysvec_apic_timer_interrupt
                -> asm_sysvec_apic_timer_interrupt
                -> intel_dmc_load_program
                -> icl_display_core_init
                -> intel_display_power_resume
                -> intel_runtime_resume
                -> __rpm_callback
                -> rpm_callback
                -> rpm_resume
                -> __pm_runtime_resume
                -> intel_runtime_pm_get
                -> intel_display_power_get
                -> icl_tc_phy_hpd_live_status
                -> tc_phy_hpd_live_status
                -> tc_phy_get_target_mode
                -> __intel_tc_port_lock
                -> intel_digital_port_connected
                -> intel_dp_detect
                -> drm_helper_probe_detect_ctx
                -> output_poll_execute
                -> process_one_work
                -> worker_thread
                -> kthread
                -> ret_from_fork
  IRQ interference                                           2.44 us (0.24 %)
               local_timer:236                               2.44 us
------------------------------------------------------------------------
  Thread latency:                                         1035.95 us (100%)

Max timerlat IRQ latency from idle: 37.31 us in cpu 8