#14355 closed bug (fixed)

Unloading/reloading an atheroswifi device ends in vm_page_fault KDL

Reported by: v.vill Owned by: waddlesplash
Priority: normal Milestone: Unscheduled
Component: Drivers/Network/atheroswifi Version: R1/Development
Keywords: Cc:
Blocked By: Blocking: #14411
Has a Patch: no Platform: All

Description

Due to some problems with the updated wifi stack (see #14282 and #14349), sometimes wpa_supplicant or net_server crashes and then the network list becomes empty. Reopening the Network preferences window, selecting the wifi adapter and clicking on "Disable" then "Enable", sometimes makes the card available again. But nearly one time out of two or three attemps, it triggers a vm_page_fault KDL instead (see attached).

I'm running hrev 52205 x86_64 with an AR9565 wifi chip.

Attachments (2)

IMG_20180813_221010.jpg (1.7 MB ) - added by v.vill 15 months ago.
KDL photo
IMG_20180828_010441.jpg (731.9 KB ) - added by v.vill 15 months ago.
KDL threads

Change History (12)

by v.vill, 15 months ago

Attachment: IMG_20180813_221010.jpg added

KDL photo

comment:1 by miqlas, 15 months ago

Empty network list is not new, happend frequently with the old wlan drivers too. I think it is the wlan-supplicant. But it never KDL'd for me.

comment:2 by waddlesplash, 15 months ago

Component: - GeneralDrivers/Network/atheroswifi
Owner: changed from nobody to waddlesplash

comment:3 by waddlesplash, 15 months ago

Please try again after hrev52288 (and your other WiFi-related tickets also.)

If it still occurs, please run threads 1 at the KDL prompt and include the output of that here.

by v.vill, 15 months ago

Attachment: IMG_20180828_010441.jpg added

KDL threads

comment:4 by v.vill, 15 months ago

No problem; see attached KDL screenshot. Reproduced with hrev52291.

comment:5 by v.vill, 15 months ago

Darn. It appears this problem is now even more easily triggered than I thought. I opened a new ticket just in case (sorry for the noise if it wasn’t relevant); see #14411.

comment:6 by waddlesplash, 15 months ago

Blocking: 14411 added

comment:7 by waddlesplash, 15 months ago

OK, so, ath_reset_grablock just calls "pause", which is defined as a macro here. _pause then calls tsleep, also a macro, which calls msleep but with a NULL mutex. msleep apparently tries to unconditionally lock or unlock the mutex, which seems pretty strange. I need to look at what FreeBSD does here.

comment:8 by waddlesplash, 15 months ago

This (and #14411) should be fixed with hrev52297. Please test.

comment:9 by v.vill, 15 months ago

OK, we seem to be getting somewhere. I couldn't trigger the KDL anymore, no matter how hard I tried! This appears to be fixed (unfortunately it still leaves me with #14349, see my update there). Let me now if there's any additional steps I can take to verify further.

Kudos!

comment:10 by waddlesplash, 15 months ago

Resolution: fixed
Status: newclosed

Nice!

Note: See TracTickets for help on using tickets.