[Ksummit-2013-discuss] [ATTEND] Printk softlockups, mmap_sem in ->fault & ->page_mkwrite

[Jan Kara]

  Hello,

  I'd like to attend Kernel Summit this year. I'm ext3, udf & quota
maintainer, working also on ext4, occasionally other filesystems,
writeback, VFS, IO scheduler and other related stuff.

Currently, I'm slowly working on modifications of get_user_pages() and
pagefault code so that filesystems can drop mmap_sem in their ->fault and
->page_mkwrite handlers. Depending on how far I get before KS, there might
be issues worth discussion (I don't want to burn people's time unless I
have some patches to show).

Other topic I'm working on and which I'd be happy to discuss if it's not
resolved by KS time are printk softlockup reports from systems with serial
console (e.g. it's really easy to kill machine with 9600 baud serial
console by triggering some heavier printk traffic, even with more realistic
115200 baud serial console I have machines that are not able to boot
because printk traffic during boot is too heavy - we loose too many
interrupts, etc.). My first attempt to solve the issue by deferring
printing to workqueue was shot down (for good reasons), suggested solution
to use irq_work didn't solve the issue for me in practice (irq_work is
executed on the same CPU with interrupts still disabled so the load
printing puts on the cpu doesn't really change). So I'm now experimenting
trying to find a solution that doesn't have the problems for which my
first attempt was refused.

>From the topics proposed so far I'm interested the most in discussing memcg
dirty throttling (if anyone has anything new to say to this long standing
issue) and huge page pagecache.

								Honza
-- 
Jan Kara <jack at suse.cz>
SUSE Labs, CR