rrs added a comment.
To answer jhb and kostikbel
1) Yes this does address the two issues that Hans re-write of the callout
system did without changing the KPI.
There may be other bugs as well, but with the test framework and the old
code I could reproduce both issues
(spin lock held to long and broken callout_active() semantic). The spin
lock held to long I was able
to get a vmcore and actually pinpoint *exactly* why it happened and
*exactly* how the patch fixes it (as
Gleb would say, no mystery ...)
2) Yes, its a good idea to split the test framework out as imp suggested, which
I will do into test framework and callout changes.
3) As to simpler testing, yes we originally were running the test out of the
kldload, but this causes
at least two problems. It appears that the GIANT lock is held during
kldload, any test that needs to do locking
can experience LOR's and other fun stuff with Witness on. Second it becomes
*much* harder to do multi-threaded testing.
This is/was essential to find the bugs and reproduce them in the callout
system. Basically you just put
num_threads = 2 and you have two kthreads running the tests which helps
located nasty races more easily ;-)
4) Imp, in an early version of the patch, complained I was not using the macros
and what he had
complained about was just code that I cut and paste. Thats what prompted me
to fix and use
the macro's properly throughout the code. Backing that out would be painful
at this point so
I see no need to split it out except to do a *lot* of extra work.
REVISION DETAIL
https://reviews.freebsd.org/D1711
To: rrs, gnn, rwatson, adrian, sbruno, lstewart, imp, hselasky
Cc: jhb, kostikbel, emaste, delphij, neel, erj, freebsd-net
_______________________________________________
freebsd-net@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-net
To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"
