Can't resume from suspend if a specific disk is added to DISK_DEVICES with AHCI_RUNTIME_PM_ON_BAT enabled #606

luckv · 2021-12-09T16:49:04Z

[x] I've read and accepted the Bug Reporting Howto
[x] I've provided all required tlp-stat outputs via Gist

Describe the bug
If the parameter DISK_DEVICES is specifically set with one of my disks (sdb) and AHCI_RUNTIME_PM_ON_BAT is set to auto, after suspend (systemctl suspend), the pc turn on but not completely. The screen turn on but remains blank (lit but black), and the wifi doesn't connect (I can't ping the pc after suspension).

I don't know much about the disk. Was shipped with the laptop and report this name: Micron 1100 MTFD

I excluded all other possible causes:

The issue doesn't happen without tlp, or with tlp disabled
The issue doesn't happen with tlp default configuration
The issue doesn't happen if that disk is excluded from DISK_DEVICES or AHCI_RUNTIME_PM_ON_BAT is set to on
The issue doesn't happen with all other settings enabled that you see in my tlp-stat. I enabled them one by one and rebooted to check where the problem was.
Gone through the troubleshooting guide, disabled both RUNTIME_PM_ON_BAT and USB_AUTOSUSPEND, but doesn't work
Searched in the web for similar issues
Searched in this repository for similar issues

Expected behavior
The pc to return from suspension normally

To Reproduce
Steps to reproduce the unexpected behavior:

Set sdb in the list of devices

The problem occurs only in battery mode
Full output of tlp-stat: link. All customizations are in luckv.conf

Additional context
OS: Fedora 35 (Workstation Edition)
Kernel version (uname -r): 5.15.6-200.fc35.x86_64

I have also Pop!_OS version 21.04 (same kernel version, 5.15) installed on the same pc (also the same boot disk) but with TLP 1.3.1. The issue doesn't present with it.

The text was updated successfully, but these errors were encountered:

linrunner · 2021-12-09T17:58:56Z

This is probably related to #587 (comment)

The AHCI autosuspend sysfiles associated to sdb are:

/sys/block/sdb/power/control = on, autosuspend_delay_ms = -1
/sys/bus/pci/devices/0000:00:17.0/ata2/power/control     = auto  -- sdb

The disk seems to refuse to accept auto to /sys/block/sdb/power/control.

Does the problem disappear after writing on to /sys/bus/pci/devices/0000:00:17.0/ata2/power/control?

luckv · 2021-12-09T18:18:49Z

@linrunner are you suggesting that should I try to write directly in /sys/bus/pci/devices/0000:00:17.0/ata2/power/control?
How can I do that? echo 'abc' >> /sys/bus/pci/devices/0000:00:17.0/ata2/power/control ?

linrunner · 2021-12-09T19:11:10Z

Almost. You need a root shell:

echo on > /sys/bus/pci/devices/0000:00:17.0/ata2/power/control

luckv · 2021-12-09T21:01:27Z

The situation has changed. Before I tried what you suggest I commented AHCI_RUNTIME_PM_ON_BAT in my configuration so its value was auto.
Without doing a reboot I done sudo tlp start, then I checked tlp-stat and what I saw was:

/sys/block/sdb/power/control = auto, autosuspend_delay_ms = 15000
/sys/bus/pci/devices/0000:00:17.0/ata2/power/control     = auto  -- sdb

but the problem persists, also after reboot. After reboot I took tlp-stat. See here

I wait for you response because this is so unexpected, I don't think doing what you suggest now it's the correct thing to do

linrunner · 2021-12-10T06:27:14Z

You tested the opposite of what I need to isolate the cause. So again properly please:

echo on > /sys/bus/pci/devices/0000:00:17.0/ata2/power/control
echo on > /sys/block/sdb/power/control

Then suspend and check if the problem occurs again.

luckv · 2021-12-12T10:50:40Z

Sorry @linrunner for answering only now.
I do what you suggested and it worked! My laptop resume from suspension correctly.
In tlp-stat the output remained the same as my last message (see here).
But if I try to read those values with cat I obtain those results:

[root@fedora luckv]# cat /sys/block/sdb/power/control
on
[root@fedora luckv]# cat /sys/bus/pci/devices/0000:00:17.0/ata2/power/control
auto

Hope this can help you to resolve the problem

linrunner · 2021-12-12T18:37:22Z

In tlp-stat the output remained the same as my last message (see here).
But if I try to read those values with cat I obtain those results:

tlp-stat always displays the same values as cat, it even uses cat internally.

The deviation in your test case is due to the fact that tlp is called again after the resume and restores the state before your manual write operations. I guess the next suspend will produce the problem again.

What I don't understand is why /sys/block/sdb/power/control stays on after boot and resume but is auto after tlp start.

Let's test this a little more thoroughly:

Enable trace mode by adding the following line to your config file

TLP_DEBUG="disk pm ps run"

Reboot and show the output of

sudo tlp-stat -d

Make your manual changes (in a root shell)

echo on > /sys/bus/pci/devices/0000:00:17.0/ata2/power/control
echo on > /sys/block/sdb/power/control

and show the output of

sudo tlp-stat -d

again.

Suspend and resume your laptop. Show the output of

sudo tlp-stat -d

once again. Finallly the trace output for the whole business

sudo tlp-stat -T

luckv · 2021-12-12T19:35:10Z

The deviation in your test case is due to the fact that tlp is called again after the resume and restores the state before your manual write operations. I guess the next suspend will produce the problem again.

You are right, only the first time the pc suspends and resumes correctly

luckv · 2021-12-12T20:00:26Z

@linrunner I do what as you written.
I created a Gist with all the output you requested here.
The number at the start of the file name should be the number in the numbered list of the passages. I also included a tlp-stat -c taken at the end to recheck my configuration.

linrunner · 2021-12-13T19:09:57Z

Your outputs show strange deviations.

The result of writing manually

echo on > /sys/bus/pci/devices/0000:00:17.0/ata2/power/control
echo on > /sys/block/sdb/power/control

is

/sys/block/sdb/power/control = auto, autosuspend_delay_ms = 15000
/sys/bus/pci/devices/0000:00:17.0/ata2/power/control = on -- sdb

In your start post the initial state is

/sys/block/sdb/power/control = on, autosuspend_delay_ms = -1
/sys/bus/pci/devices/0000:00:17.0/ata2/power/control = auto -- sdb

On the other hand, the result of the recent boot process is

/sys/block/sdb/power/control = auto, autosuspend_delay_ms = 15000
/sys/bus/pci/devices/0000:00:17.0/ata2/power/control = auto -- sdb

Is this still the same kernel version?

As a workaround I suggest you turn off AHCI runtime pm completely by configuring

AHCI_RUNTIME_PM_ON_AC= 
AHCI_RUNTIME_PM_ON_BAT=

Show your effective configuration (difference to TLP's defaults only)

tlp-stat -s --cdiff
Reboot and show

sudo tlp-stat -s -d
Suspend/resume and show

sudo tlp-stat -s -d
Another suspend/resume and show

sudo tlp-stat -s -d

luckv · 2021-12-13T19:41:07Z

Is this still the same kernel version?

Absolutely yes. I show you the output of my uname --all:

Linux xxx-fedora 5.15.6-200.fc35.x86_64 #1 SMP Wed Dec 1 13:41:10 UTC 2021 x86_64 x86_64 x86_64 GNU/Linux

In your start post the initial state is

/sys/block/sdb/power/control = on, autosuspend_delay_ms = -1
/sys/bus/pci/devices/0000:00:17.0/ata2/power/control = auto -- sdb

I wasn't able to obtain that output ever again. I don't know why onestly.

Your outputs show strange deviations.

I know. In fact one time I don't try what you asked because I was confused of the results of my actions.
One very very strange thing it's that, even if tlp uses cat internally as you said, tlp-stat -d and cat /sys/block/sdb/power/control show 2 different results, one said that /sys/block/sdb/power/control is on, the second say that is auto... even if executed one after the other.
This behaviour it's not deterministic... I don't know really what to think.

luckv · 2021-12-13T19:59:44Z

I follow the passages. Here the output.

The output after reboot and after suspends, remained the same, with this output for both disks:

/sys/block/sdb/power/control = on, autosuspend_delay_ms = -1
/sys/bus/pci/devices/0000:00:17.0/ata2/power/control     = on  -- sdb

@linrunner I started thinking this is a problem of the disk.. may be?

luckv · 2021-12-13T20:15:17Z

While in battery mode I restored AHCI_RUNTIME_PM_ON_BAT=on, shutdown the pc, then booted on.
I opened a terminal and launched only this commands. The outputs for /sys/block/... differs from tlp-stat

Gist here

/sys/block/<DISK>/device/power/control References: * #606 * https://lkml.org/lkml/2016/2/18/147

linrunner · 2021-12-14T07:03:10Z

Oops. In fact there is an error in the output. TLP internally reads and writes the correct path /sys/block/DISK/device/power/control, but tlp-stat -d displays it incorrectly as /sys/block/DISK/power/control.

I can't provide a Fedora package right away, so please edit the file /usr/share/tlp/func.d/tlp-func-stat Line 446 by hand according to the commit diff please.

References: * linrunner/TLP#587 (comment) * linrunner/TLP#593 * linrunner/TLP#606

linrunner · 2021-12-14T17:59:31Z

Workaround added to the FAQ.

luckv · 2021-12-14T20:26:22Z

I edited my files. Do you want that I test something else for you @linrunner ?

luckv · 2021-12-14T20:30:18Z

May be really a bug in the linux kernel??

linrunner · 2021-12-15T06:28:18Z

Just for the record: with AHCI_RUNTIME_PM_ON_BAT=on suspend/resume works without freeze now?

May be really a bug in the linux kernel??

Of course. But we will not solve it here, you will have to create a kernel bug report.

I edited my files. Do you want that I test something else for you @linrunner ?

I will let you know when the 1.5 beta is ready.

luckv · 2021-12-17T20:14:07Z

Just for the record: with AHCI_RUNTIME_PM_ON_BAT=on suspend/resume works without freeze now?

Yes

Of course. But we will not solve it here, you will have to create a kernel bug report.

Ok

I will let you know when the 1.5 beta is ready.

Perfect

linrunner · 2021-12-22T17:34:02Z

@luckv : Fedora F35 packages are available now. Enjoy.

References: * #587 * #593 * #606

linrunner added a commit that referenced this issue Dec 14, 2021

tlp-stat -d: display correct sysfs path for disk runtime pm

8663478

/sys/block/<DISK>/device/power/control References: * #606 * https://lkml.org/lkml/2016/2/18/147

linrunner added a commit to linrunner/tlp-doc that referenced this issue Dec 14, 2021

faq/op: system freezes on wakeup from suspend on battery

0786dbb

References: * linrunner/TLP#587 (comment) * linrunner/TLP#593 * linrunner/TLP#606

linrunner added documentation change kernel issue labels Dec 14, 2021

linrunner added the fixed/completed label Dec 15, 2021

linrunner added this to the 1.5 Release milestone Dec 15, 2021

linrunner added ahci pm regression labels Dec 15, 2021

luckv mentioned this issue Dec 22, 2021

[TLP 1.5 Beta] Call for testers #609

Closed

linrunner removed this from the 1.5 Release milestone Jun 10, 2022

linrunner closed this as completed Sep 25, 2022

linrunner added a commit that referenced this issue Mar 31, 2023

Deactivate AHCI_RUNTIME_PM before suspend to avoid resume freezes

56bb6ce

References: * #587 * #593 * #606

linrunner mentioned this issue Jul 21, 2023

[TLP 1.6 Beta] Call for testers #700

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Can't resume from suspend if a specific disk is added to DISK_DEVICES with AHCI_RUNTIME_PM_ON_BAT enabled #606

Can't resume from suspend if a specific disk is added to DISK_DEVICES with AHCI_RUNTIME_PM_ON_BAT enabled #606

luckv commented Dec 9, 2021 •

edited

Loading

linrunner commented Dec 9, 2021 •

edited

Loading

luckv commented Dec 9, 2021 •

edited

Loading

linrunner commented Dec 9, 2021 •

edited

Loading

luckv commented Dec 9, 2021 •

edited

Loading

linrunner commented Dec 10, 2021

luckv commented Dec 12, 2021

linrunner commented Dec 12, 2021

luckv commented Dec 12, 2021 •

edited

Loading

luckv commented Dec 12, 2021 •

edited

Loading

linrunner commented Dec 13, 2021

luckv commented Dec 13, 2021 •

edited

Loading

luckv commented Dec 13, 2021

luckv commented Dec 13, 2021

linrunner commented Dec 14, 2021

linrunner commented Dec 14, 2021

luckv commented Dec 14, 2021

luckv commented Dec 14, 2021

linrunner commented Dec 15, 2021

luckv commented Dec 17, 2021 •

edited

Loading

linrunner commented Dec 22, 2021

Can't resume from suspend if a specific disk is added to DISK_DEVICES with AHCI_RUNTIME_PM_ON_BAT enabled #606

Can't resume from suspend if a specific disk is added to DISK_DEVICES with AHCI_RUNTIME_PM_ON_BAT enabled #606

Comments

luckv commented Dec 9, 2021 • edited Loading

linrunner commented Dec 9, 2021 • edited Loading

luckv commented Dec 9, 2021 • edited Loading

linrunner commented Dec 9, 2021 • edited Loading

luckv commented Dec 9, 2021 • edited Loading

linrunner commented Dec 10, 2021

luckv commented Dec 12, 2021

linrunner commented Dec 12, 2021

luckv commented Dec 12, 2021 • edited Loading

luckv commented Dec 12, 2021 • edited Loading

linrunner commented Dec 13, 2021

luckv commented Dec 13, 2021 • edited Loading

luckv commented Dec 13, 2021

luckv commented Dec 13, 2021

linrunner commented Dec 14, 2021

linrunner commented Dec 14, 2021

luckv commented Dec 14, 2021

luckv commented Dec 14, 2021

linrunner commented Dec 15, 2021

luckv commented Dec 17, 2021 • edited Loading

linrunner commented Dec 22, 2021

luckv commented Dec 9, 2021 •

edited

Loading

linrunner commented Dec 9, 2021 •

edited

Loading

luckv commented Dec 9, 2021 •

edited

Loading

linrunner commented Dec 9, 2021 •

edited

Loading

luckv commented Dec 9, 2021 •

edited

Loading

luckv commented Dec 12, 2021 •

edited

Loading

luckv commented Dec 12, 2021 •

edited

Loading

luckv commented Dec 13, 2021 •

edited

Loading

luckv commented Dec 17, 2021 •

edited

Loading