Trace mailbox shows garbage

There're full of garbage in mailbox after system booted. this issue happens with kernel sof-v4.18 and sof-dev branch on APL and CNL, but cannot reproduce this issue on BYT.
Error log:
using 19.20MHz timestamp clock
0x0 [960767920505.705811] delta [********] value 0xffffffffffffffff
0x10 [960767920505.705811] delta [0.000000] value 0xffffffffffffffff
0x20 [960767920505.705811] delta [0.000000] value 0xffffffffffffffff
0x30 [960767920505.705811] delta [0.000000] value 0xffffffffffffffff
0x40 [960767920505.705811] delta [0.000000] value 0xffffffffffffffff
0x50 [960767920505.705811] delta [0.000000] value 0xffffffffffffffff
0x60 [960767920505.705811] delta [0.000000] value 0xffffffffffffffff
0x70 [960767920505.705811] delta [0.000000] value 0xffffffffffffffff
0x80 [960767920505.705811] delta [0.000000] value 0xffffffffffffffff
0x90 [960767920505.705811] delta [0.000000] value 0xffffffffffffffff
0xa0 [960767920505.705811] delta [0.000000] value 0xffffffffffffffff
rmbox.log

rmbox-t.log
dmesg.log

@lgirdwood

rmbox -t works fine (the rmbox-t.log file above shows the same)
rmbox without the -t doesn't work at all indeed.

If I am not mistaken, it means the DMA transfers work ok, but the memory window doesn't

@lgirdwood your input would be most useful here.

Yes, the rmbox -t works fine, but memory window is full of garbage. I have changed the title.

Looks PM related, ie. we are in S0ix when reading the SRAM window and should cache the trace as context.

I removed the RUNTIME_PM options and the error is still there.
I don't have time to look into this, and I don't have time to wait either, so I am tempted to push this v4.18 version as a force-in-function operation to fix this...

@keqiaozhang

@keqiaozhang @lgirdwood
it seems to be error caused by PM_RUNTIME
manually disable all RUNTIME code will fix this.
We have discuss this yesterday
@plbossart
I found that we may have some error in

int snd_sof_runtime_suspend(struct device *dev)
{
	return sof_suspend(dev);
}

int snd_sof_suspend(struct device *dev)
{
	/* suspend if dev is runtime_active */
	if (pm_runtime_active(dev))
		return sof_suspend(dev);

	return 0;
}

RUNTIME_PM option will not influence the runtime pm from the above code.

well, I've lost five hours of my time due to a misleading bug report . This rmbox issue already exists for topic/sof-dev, meaning there was no new bug introduced by the 4.18 transition. GAH!

@plbossart

@plbossart We apologize for the misleading message, I just confirmed that this issue also happens for sof-dev branch. We will review our work.

I will add a feature enhancement for caching context prior to D3 / S0ix.

@keqiaozhang

@keqiaozhang no hard feelings, it's a good finding, and we may be able to bisect to see where the problem happens.

Git bisect points to the following bad commit - definitively linked to runtime_pm.

64980d0c14a5257e85763b6b77f0d0bb29812a16 is the first bad commit
commit 64980d0c14a5257e85763b6b77f0d0bb29812a16
Author: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Date:   Fri Jul 20 00:12:15 2018 -0700

    ASoC: SOF: decrement device usage counter after booting firmware
    
    Decrement device usage counter for both the sof device
    and the pci/acpi/spi device.
    Without this change the runtime_usage count never reaches 0
    and the device will not suspend.
    
    Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>

Couple of findings on the resume flow:
PCI_TCSEL is not updated on resume - different from Skylake driver
CGCTL.MISCBDCGE is not updated on resume - different from Skylake driver.
PPCTL_GPROCEN PPCTL_PIE are not enable on resume - different from Skylake driver
GPROCEN is a smoking gun, it defined how the BAR access maps to memory mapped configuration registers.

On suspend:
PGCTL register not updated to disable LP retention mode (LSRMD)
MISCBDCGE changes before resetting the link

@plbossart

@plbossart
I also run a git bisect and the first bad commit points to:
commit 7cd8049
Author: Ranjani Sridharan ranjani.sridharan@linux.intel.com
Date: Tue Jul 24 19:54:00 2018 -0700

ASoC: SOF: allow runtime_pm for pci/acpi/spi device

Allow runtime_pm to be enabled after probe is finished for
pci/acpi/spi devices.

I also checked with 64980d0, but failed to reproduce this issue.

@mengdonglin

@mengdonglin @ZhendanYang @keqiaozhang The issue seems not to be reproducible now. Could you check it?

@bardliao

@bardliao Yes, I haven't seen this issue for a long time. But I'm not sure which patch fixed this. I'll close it. We can reopen it if it occurs in the future.

bpf: sockmap, restore sk_write_space when psock gets dropped

scsi: target: fix hang when multiple threads try to destroy the same …

btrfs: fix warning when creating a directory with smack enabled

Merge pull request thesofproject#81 from Rust-for-Linux/this-module-r…

wifi: mac80211: fix memory corruption in minstrel_ht_update_rates()

LoongArch: BPF: Fix sign-extension mov instructions

bpftool: Fix missing pids during link show

mm/pagewalk: fix usage of pmd_leaf()/pud_leaf() without present check

LoongArch: BPF: Optimize sign-extention mov instructions

ALSA: usb-audio: Add quirks for HP Elite x3 Lap Dock

Navigation Menu

Repository navigation

Trace mailbox shows garbage #81

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Trace mailbox shows garbage #81

Description

Activity

keqiaozhang commented on Aug 14, 2018

plbossart commented on Aug 14, 2018

keqiaozhang commented on Aug 14, 2018

lgirdwood commented on Aug 14, 2018

plbossart commented on Aug 15, 2018

xiulipan commented on Aug 15, 2018

plbossart commented on Aug 15, 2018

keqiaozhang commented on Aug 15, 2018

lgirdwood commented on Aug 15, 2018

plbossart commented on Aug 15, 2018

plbossart commented on Aug 15, 2018

plbossart commented on Aug 15, 2018

keqiaozhang commented on Aug 16, 2018

bardliao commented on Sep 14, 2018

ZhendanYang commented on Sep 14, 2018

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions