Page MenuHomeFreeBSD

sysctl: make epp select proc serialized.
Needs ReviewPublic

Authored by weike.chen_dell.com on Dec 11 2023, 7:53 AM.
Tags
None
Referenced Files
Unknown Object (File)
Fri, Jan 10, 7:58 PM
Unknown Object (File)
Dec 6 2024, 12:23 PM
Unknown Object (File)
Nov 5 2024, 9:20 AM
Unknown Object (File)
Oct 4 2024, 10:28 PM
Unknown Object (File)
Oct 1 2024, 9:32 PM
Unknown Object (File)
Sep 26 2024, 5:57 AM
Unknown Object (File)
Sep 19 2024, 9:56 PM
Unknown Object (File)
Sep 18 2024, 1:53 AM
Subscribers

Details

Reviewers
cem
Summary

On Dell Lati5440 laptop (RaptorLake U), 'sysctl -a' hang due to epp select.

Test Plan

'sysctl -a'

Diff Detail

Repository
rG FreeBSD src repository
Lint
Lint Skipped
Unit
Tests Skipped

Event Timeline

@weike.chen_dell.com
I'm running sysctl dev.hwpstate_intel | grep epp or sysctl -a and all looks sane.

I'm not understanding how CTLFLAG_NEEDGIANT solves the problem. Maybe we firstly need to know where and why sysctl_epp_select() got blocked infinitely.

In D42993#982163, @zlei wrote:

@weike.chen_dell.com
I'm running sysctl dev.hwpstate_intel | grep epp or sysctl -a and all looks sane.

I'm not understanding how CTLFLAG_NEEDGIANT solves the problem. Maybe we firstly need to know where and why sysctl_epp_select() got blocked infinitely.

Were you trying it on RaptorLake U CPU? So far, for this is issue, we only reproduce it on RaptorLake U (Dell Latitude 5440). This could be some risk conditions since I check then original code is serialized the process.

In D42993#982163, @zlei wrote:

@weike.chen_dell.com
I'm running sysctl dev.hwpstate_intel | grep epp or sysctl -a and all looks sane.

I'm not understanding how CTLFLAG_NEEDGIANT solves the problem. Maybe we firstly need to know where and why sysctl_epp_select() got blocked infinitely.

Were you trying it on RaptorLake U CPU?

No, I tested on old I5 CPUs.

'sysctl -a' hang
So far, for this is issue, we only reproduce it on RaptorLake U (Dell Latitude 5440)

May you please file a problem report to https://bugs.freebsd.org/bugzilla/ ? Phabricator is really not good place to discussing bugs like this.

This could be some risk conditions since I check then original code is serialized the process.

Possibly true if it could not be repeated reliably.