On Fri, May 30, 2025 at 10:14:36AM +0900, YASUOKA Masahiko wrote:
> On Fri, 30 May 2025 10:08:38 +1000
> Jonathan Gray <j...@jsg.id.au> wrote:
> > On Fri, May 30, 2025 at 08:01:31AM +0900, YASUOKA Masahiko wrote:
> >>
> >> >Synopsis: inteldrm stop working after {hibernate,suspend}/resume
> >> >Category: kernel
> >> >Environment:
> >> System : OpenBSD 7.7
> >> Details : OpenBSD 7.7-current (GENERIC.MP) #117: Thu May 29
> >> 21:18:15 JST 2025
> >>
> >> yasuoka@xxx:/home/yasuoka/src/sys/arch/amd64/compile/GENERIC.MP
> >>
> >> Architecture: OpenBSD.amd64
> >> Machine : amd64
> >> >Description:
> >> After hibernate and resume, X11 stops working. Keyboard and
> >> mouse don't work, but Ctrl-Alt-F1 or Ctrl-Alt-Backspace works.
> >>
> >> errors in dmesg:
> >> ****
> >> drm:pid97650:__uc_init_hw *ERROR* [drm] *ERROR* GT0: GuC initialization
> >> failed 0xfffffffffffffffae
> >> drm:pid97650:intel_gt_init_hw *ERROR* [drm] *ERROR* GT0: Enabling uc
> >> failed (-5)
> >> drm:pid97650:intel_gt_resume *ERROR* [drm] *ERROR* GT0: Failed to
> >> initialize GPU, declaring it wedged!
> >> ****
> >>
> >> This happens because guc_wait_ucode() in i915/gt/uc/intel_guc_fw.c
> >> fails.
> >>
> >> The function is to wait for the GuC to start up by calling the inline
> >> function guc_load_done() and the function checks two regisiters.
> >>
> >> 97 static inline bool guc_load_done(struct intel_uncore *uncore,
> >> u32 *status, bool *success)
> >> 98 {
> >> 99 u32 val = intel_uncore_read(uncore, GUC_STATUS);
> >> 100 u32 uk_val = REG_FIELD_GET(GS_UKERNEL_MASK, val);
> >> 101 u32 br_val = REG_FIELD_GET(GS_BOOTROM_MASK, val);
> >> 102
> >> 103 *status = val;
> >> 104 switch (uk_val) {
> >> 105 case INTEL_GUC_LOAD_STATUS_READY:
> >> 106 *success = true;
> >> 107 return true;
> >> 108
> >> 109 case INTEL_GUC_LOAD_STATUS_ERROR_DEVID_BUILD_MISMATCH:
> >> 110 case INTEL_GUC_LOAD_STATUS_GUC_PREPROD_BUILD_MISMATCH:
> >> 111 case INTEL_GUC_LOAD_STATUS_ERROR_DEVID_INVALID_GUCTYPE:
> >>
> >> In my test, the functions fails with the resgisters:
> >>
> >> ukernel = INTEL_GUC_LOAD_STATUS_INIT_DATA_INVALID(0x71)
> >> bootrom = INTEL_BOOTROM_STATUS_JUMP_PASSED(0x76)
> >>
> >> When I was using 7.6, I didn't see this problem.
> >>
> >> >How-To-Repeat:
> >> 1. hibernate or suspend
> >> 2. resume
> >>
> >> the problem happens always (~10 times)
> >>
> >> After the workaround diff, not happen always (~3 times)
> >>
> >> >Fix:
> >> Also the diff attached at last, workaround the problem.
> >>
> >> The diff partially backouts the change on Feb 7 and add a printf().
> >>
> >> I don't understand it logically, but if the printf() is removed, the
> >> problem start happening.
> >
> > Thank you for the report.
> >
> > Does this smaller diff still workaround the problem?
>
> The smaller diff doesn't fix the problem. I tried 2 times.
Here is the other part of your initial diff.
The non-printf parts are a revert of
'drm/i915/guc: Change wa and EU_PERF_CNTL registers to MCR type'
linux 835e4d9bb3a13879031942ca6692d5a82ec00158
It would also be helpful if you could try raise the value
of GUC_LOAD_RETRY_LIMIT in intel_guc_fw.c without other patches,
to find a value that works.
Index: sys/dev/pci/drm/i915/gt/uc/intel_guc_ads.c
===================================================================
RCS file: /cvs/src/sys/dev/pci/drm/i915/gt/uc/intel_guc_ads.c,v
diff -u -p -r1.9 intel_guc_ads.c
--- sys/dev/pci/drm/i915/gt/uc/intel_guc_ads.c 7 Feb 2025 03:03:30 -0000
1.9
+++ sys/dev/pci/drm/i915/gt/uc/intel_guc_ads.c 30 May 2025 05:58:58 -0000
@@ -408,13 +408,8 @@ static int guc_mmio_regset_init(struct t
CCS_MASK(engine->gt))
ret |= GUC_MMIO_REG_ADD(gt, regset, GEN12_RCU_MODE, true);
- /*
- * some of the WA registers are MCR registers. As it is safe to
- * use MCR form for non-MCR registers, for code simplicity, all
- * WA registers are added with MCR form.
- */
for (i = 0, wa = wal->list; i < wal->count; i++, wa++)
- ret |= GUC_MCR_REG_ADD(gt, regset, wa->mcr_reg, wa->masked_reg);
+ ret |= GUC_MMIO_REG_ADD(gt, regset, wa->reg, wa->masked_reg);
/* Be extra paranoid and include all whitelist registers. */
for (i = 0; i < RING_MAX_NONPRIV_SLOTS; i++)
@@ -430,13 +425,13 @@ static int guc_mmio_regset_init(struct t
ret |= GUC_MMIO_REG_ADD(gt, regset, GEN9_LNCFCMOCS(i),
false);
if (GRAPHICS_VER(engine->i915) >= 12) {
- ret |= GUC_MCR_REG_ADD(gt, regset,
MCR_REG(i915_mmio_reg_offset(EU_PERF_CNTL0)), false);
- ret |= GUC_MCR_REG_ADD(gt, regset,
MCR_REG(i915_mmio_reg_offset(EU_PERF_CNTL1)), false);
- ret |= GUC_MCR_REG_ADD(gt, regset,
MCR_REG(i915_mmio_reg_offset(EU_PERF_CNTL2)), false);
- ret |= GUC_MCR_REG_ADD(gt, regset,
MCR_REG(i915_mmio_reg_offset(EU_PERF_CNTL3)), false);
- ret |= GUC_MCR_REG_ADD(gt, regset,
MCR_REG(i915_mmio_reg_offset(EU_PERF_CNTL4)), false);
- ret |= GUC_MCR_REG_ADD(gt, regset,
MCR_REG(i915_mmio_reg_offset(EU_PERF_CNTL5)), false);
- ret |= GUC_MCR_REG_ADD(gt, regset,
MCR_REG(i915_mmio_reg_offset(EU_PERF_CNTL6)), false);
+ ret |= GUC_MMIO_REG_ADD(gt, regset, EU_PERF_CNTL0, false);
+ ret |= GUC_MMIO_REG_ADD(gt, regset, EU_PERF_CNTL1, false);
+ ret |= GUC_MMIO_REG_ADD(gt, regset, EU_PERF_CNTL2, false);
+ ret |= GUC_MMIO_REG_ADD(gt, regset, EU_PERF_CNTL3, false);
+ ret |= GUC_MMIO_REG_ADD(gt, regset, EU_PERF_CNTL4, false);
+ ret |= GUC_MMIO_REG_ADD(gt, regset, EU_PERF_CNTL5, false);
+ ret |= GUC_MMIO_REG_ADD(gt, regset, EU_PERF_CNTL6, false);
}
return ret ? -1 : 0;
Index: sys/dev/pci/drm/i915/gt/uc/intel_guc_fw.c
===================================================================
RCS file: /cvs/src/sys/dev/pci/drm/i915/gt/uc/intel_guc_fw.c,v
diff -u -p -r1.8 intel_guc_fw.c
--- sys/dev/pci/drm/i915/gt/uc/intel_guc_fw.c 7 Feb 2025 03:03:30 -0000
1.8
+++ sys/dev/pci/drm/i915/gt/uc/intel_guc_fw.c 30 May 2025 05:59:52 -0000
@@ -197,6 +197,7 @@ static int guc_wait_ucode(struct intel_g
REG_FIELD_GET(GS_BOOTROM_MASK, status),
REG_FIELD_GET(GS_UKERNEL_MASK, status));
}
+ printf("%s: count = %d, ret = %d\n", __func__, count, ret);
after = ktime_get();
delta = ktime_sub(after, before);
delta_ms = ktime_to_ms(delta);
