linux/drivers/gpu/drm/i915/gt/selftest_rps.c
Chris Wilson a50717dbf4 drm/i915/selftests: Take the engine wakeref around __rps_up_interrupt
Since we are touching the device to read the registers, we are required
to ensure the device is awake at the time. Currently, we believe
ourselves to be inside the active request [thus an active engine
wakeref], but since that may be retired in the background, we can
spontaneously lose the wakeref and the ability to probe the HW.

<4> [379.686703] RPM wakelock ref not held during HW access
<4> [379.686805] WARNING: CPU: 7 PID: 4869 at ./drivers/gpu/drm/i915/intel_runtime_pm.h:115 gen12_fwtable_read32+0x233/0x300 [i915]
<4> [379.686808] Modules linked in: i915(+) vgem snd_hda_codec_hdmi mei_hdcp x86_pkg_temp_thermal coretemp crct10dif_pclmul crc32_pclmul ax88179_178a usbnet mii ghash_clmulni_intel snd_intel_dspcfg snd_hda_codec snd_hwdep snd_hda_core snd_pcm e1000e mei_me ptp mei pps_core intel_lpss_pci prime_numbers [last unloaded: i915]
<4> [379.686827] CPU: 7 PID: 4869 Comm: i915_selftest Tainted: G     U            5.7.0-rc1-CI-CI_DRM_8313+ #1
<4> [379.686830] Hardware name: Intel Corporation Tiger Lake Client Platform/TigerLake U DDR4 SODIMM RVP, BIOS TGLSFWI1.R00.2457.A13.1912190237 12/19/2019
<4> [379.686883] RIP: 0010:gen12_fwtable_read32+0x233/0x300 [i915]
<4> [379.686887] Code: d8 ea e0 0f 0b e9 19 fe ff ff 80 3d ad 12 2d 00 00 0f 85 17 fe ff ff 48 c7 c7 b0 32 3e a0 c6 05 99 12 2d 00 01 e8 2d d8 ea e0 <0f> 0b e9 fd fd ff ff 8b 05 c4 75 56 e2 85 c0 0f 85 84 00 00 00 48
<4> [379.686889] RSP: 0018:ffffc90000727970 EFLAGS: 00010286
<4> [379.686892] RAX: 0000000000000000 RBX: ffff88848cc20ee8 RCX: 0000000000000001
<4> [379.686894] RDX: 0000000080000001 RSI: ffff88843b1f0900 RDI: 00000000ffffffff
<4> [379.686896] RBP: 0000000000000000 R08: ffff88843b1f0900 R09: 0000000000000000
<4> [379.686898] R10: 0000000000000000 R11: 0000000000000000 R12: 000000000000a058
<4> [379.686900] R13: 0000000000000001 R14: ffff88848cc2bf30 R15: 00000000ffffffea
<4> [379.686902] FS:  00007f7d63f5e300(0000) GS:ffff8884a0180000(0000) knlGS:0000000000000000
<4> [379.686904] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
<4> [379.686907] CR2: 000055e5c30f4988 CR3: 000000042e190002 CR4: 0000000000760ee0
<4> [379.686910] PKRU: 55555554
<4> [379.686911] Call Trace:
<4> [379.686986]  live_rps_interrupt+0xb14/0xc10 [i915]
<4> [379.687051]  ? intel_rps_unpark+0xb0/0xb0 [i915]
<4> [379.687057]  ? __trace_bprintk+0x57/0x80
<4> [379.687143]  __i915_subtests+0xb8/0x210 [i915]
<4> [379.687222]  ? __i915_live_teardown+0x50/0x50 [i915]
<4> [379.687291]  ? __intel_gt_live_setup+0x30/0x30 [i915]
<4> [379.687361]  __run_selftests+0x112/0x170 [i915]
<4> [379.687431]  i915_live_selftests+0x2c/0x60 [i915]
<4> [379.687491]  i915_pci_probe+0x93/0x1b0 [i915]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Andi Shyti <andi.shyti@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200417093928.17822-2-chris@chris-wilson.co.uk
2020-04-17 14:47:30 +01:00

233 lines
5.7 KiB
C

// SPDX-License-Identifier: MIT
/*
* Copyright © 2020 Intel Corporation
*/
#include "intel_engine_pm.h"
#include "intel_gt_pm.h"
#include "intel_rc6.h"
#include "selftest_rps.h"
#include "selftests/igt_flush_test.h"
#include "selftests/igt_spinner.h"
static void dummy_rps_work(struct work_struct *wrk)
{
}
static void sleep_for_ei(struct intel_rps *rps, int timeout_us)
{
/* Flush any previous EI */
usleep_range(timeout_us, 2 * timeout_us);
/* Reset the interrupt status */
rps_disable_interrupts(rps);
GEM_BUG_ON(rps->pm_iir);
rps_enable_interrupts(rps);
/* And then wait for the timeout, for real this time */
usleep_range(2 * timeout_us, 3 * timeout_us);
}
static int __rps_up_interrupt(struct intel_rps *rps,
struct intel_engine_cs *engine,
struct igt_spinner *spin)
{
struct intel_uncore *uncore = engine->uncore;
struct i915_request *rq;
u32 timeout;
if (!intel_engine_can_store_dword(engine))
return 0;
mutex_lock(&rps->lock);
GEM_BUG_ON(!rps->active);
intel_rps_set(rps, rps->min_freq);
mutex_unlock(&rps->lock);
rq = igt_spinner_create_request(spin, engine->kernel_context, MI_NOOP);
if (IS_ERR(rq))
return PTR_ERR(rq);
i915_request_get(rq);
i915_request_add(rq);
if (!igt_wait_for_spinner(spin, rq)) {
pr_err("%s: RPS spinner did not start\n",
engine->name);
i915_request_put(rq);
intel_gt_set_wedged(engine->gt);
return -EIO;
}
if (!rps->active) {
pr_err("%s: RPS not enabled on starting spinner\n",
engine->name);
igt_spinner_end(spin);
i915_request_put(rq);
return -EINVAL;
}
if (!(rps->pm_events & GEN6_PM_RP_UP_THRESHOLD)) {
pr_err("%s: RPS did not register UP interrupt\n",
engine->name);
i915_request_put(rq);
return -EINVAL;
}
if (rps->last_freq != rps->min_freq) {
pr_err("%s: RPS did not program min frequency\n",
engine->name);
i915_request_put(rq);
return -EINVAL;
}
timeout = intel_uncore_read(uncore, GEN6_RP_UP_EI);
timeout = GT_PM_INTERVAL_TO_US(engine->i915, timeout);
sleep_for_ei(rps, timeout);
GEM_BUG_ON(i915_request_completed(rq));
igt_spinner_end(spin);
i915_request_put(rq);
if (rps->cur_freq != rps->min_freq) {
pr_err("%s: Frequency unexpectedly changed [up], now %d!\n",
engine->name, intel_rps_read_actual_frequency(rps));
return -EINVAL;
}
if (!(rps->pm_iir & GEN6_PM_RP_UP_THRESHOLD)) {
pr_err("%s: UP interrupt not recorded for spinner, pm_iir:%x, prev_up:%x, up_threshold:%x, up_ei:%x\n",
engine->name, rps->pm_iir,
intel_uncore_read(uncore, GEN6_RP_PREV_UP),
intel_uncore_read(uncore, GEN6_RP_UP_THRESHOLD),
intel_uncore_read(uncore, GEN6_RP_UP_EI));
return -EINVAL;
}
return 0;
}
static int __rps_down_interrupt(struct intel_rps *rps,
struct intel_engine_cs *engine)
{
struct intel_uncore *uncore = engine->uncore;
u32 timeout;
mutex_lock(&rps->lock);
GEM_BUG_ON(!rps->active);
intel_rps_set(rps, rps->max_freq);
mutex_unlock(&rps->lock);
if (!(rps->pm_events & GEN6_PM_RP_DOWN_THRESHOLD)) {
pr_err("%s: RPS did not register DOWN interrupt\n",
engine->name);
return -EINVAL;
}
if (rps->last_freq != rps->max_freq) {
pr_err("%s: RPS did not program max frequency\n",
engine->name);
return -EINVAL;
}
timeout = intel_uncore_read(uncore, GEN6_RP_DOWN_EI);
timeout = GT_PM_INTERVAL_TO_US(engine->i915, timeout);
sleep_for_ei(rps, timeout);
if (rps->cur_freq != rps->max_freq) {
pr_err("%s: Frequency unexpectedly changed [down], now %d!\n",
engine->name,
intel_rps_read_actual_frequency(rps));
return -EINVAL;
}
if (!(rps->pm_iir & (GEN6_PM_RP_DOWN_THRESHOLD | GEN6_PM_RP_DOWN_TIMEOUT))) {
pr_err("%s: DOWN interrupt not recorded for idle, pm_iir:%x, prev_down:%x, down_threshold:%x, down_ei:%x [prev_up:%x, up_threshold:%x, up_ei:%x]\n",
engine->name, rps->pm_iir,
intel_uncore_read(uncore, GEN6_RP_PREV_DOWN),
intel_uncore_read(uncore, GEN6_RP_DOWN_THRESHOLD),
intel_uncore_read(uncore, GEN6_RP_DOWN_EI),
intel_uncore_read(uncore, GEN6_RP_PREV_UP),
intel_uncore_read(uncore, GEN6_RP_UP_THRESHOLD),
intel_uncore_read(uncore, GEN6_RP_UP_EI));
return -EINVAL;
}
return 0;
}
int live_rps_interrupt(void *arg)
{
struct intel_gt *gt = arg;
struct intel_rps *rps = &gt->rps;
void (*saved_work)(struct work_struct *wrk);
struct intel_engine_cs *engine;
enum intel_engine_id id;
struct igt_spinner spin;
u32 pm_events;
int err = 0;
/*
* First, let's check whether or not we are receiving interrupts.
*/
if (!rps->enabled || rps->max_freq <= rps->min_freq)
return 0;
intel_gt_pm_get(gt);
pm_events = rps->pm_events;
intel_gt_pm_put(gt);
if (!pm_events) {
pr_err("No RPS PM events registered, but RPS is enabled?\n");
return -ENODEV;
}
if (igt_spinner_init(&spin, gt))
return -ENOMEM;
intel_gt_pm_wait_for_idle(gt);
saved_work = rps->work.func;
rps->work.func = dummy_rps_work;
for_each_engine(engine, gt, id) {
/* Keep the engine busy with a spinner; expect an UP! */
if (pm_events & GEN6_PM_RP_UP_THRESHOLD) {
intel_gt_pm_wait_for_idle(engine->gt);
GEM_BUG_ON(rps->active);
intel_engine_pm_get(engine);
err = __rps_up_interrupt(rps, engine, &spin);
intel_engine_pm_put(engine);
if (err)
goto out;
intel_gt_pm_wait_for_idle(engine->gt);
}
/* Keep the engine awake but idle and check for DOWN */
if (pm_events & GEN6_PM_RP_DOWN_THRESHOLD) {
intel_engine_pm_get(engine);
intel_rc6_disable(&gt->rc6);
err = __rps_down_interrupt(rps, engine);
intel_rc6_enable(&gt->rc6);
intel_engine_pm_put(engine);
if (err)
goto out;
}
}
out:
if (igt_flush_test(gt->i915))
err = -EIO;
igt_spinner_fini(&spin);
intel_gt_pm_wait_for_idle(gt);
rps->work.func = saved_work;
return err;
}