Re: Fixing excessive shootdowns during FFS read (kern/53124) on x86 - emap, direct map

[Eric Hawicz <erh%nimenees.com@localhost>]

On 6/10/2018 3:02 PM, Jaromír Doleček wrote:
> 2018-05-12 21:24 GMT+02:00 Chuck Silvers <chuq%chuq.com@localhost>:
> > the problem with keeping the pages locked (ie. PG_BUSY) while accessing
> > the user address space is that it can lead to deadlock if the page
> Meanwhile I've tested this scenario - wrote a test program to do
> mmap(2) for file, then calling write() using the memory from mmap as
> the buffer for write(2) for the same file offset.
>
> It actually caused the process to get SIGSEGV with new code, where the
> old code made write(2) just returned with EINVAL. I think this happens
> because fault resolution returns EINVAL when the page has PG_BUSY.
> Funny thing is that actually on Mac OS X that code actually does end
> up with deadlock, and unkillable process :D
>
> Maybe the SIGSEGV is actually acceptable behaviour? I don't think
> there would be valid reason to do mmap(2)/write(2) combination like
> that.

FWIW, a similar test on a Linux box (Debian 8, 3.16 kernel) works w/o 
errors but behaves rather oddly if the offsets don't exactly line up.  
(e.g. w/ offset 1, the first 8 chars written repeat the first read, then 
every 32 chars one char is repeated)
As long as we fail somehow, whether EINVAL or SIGSEGV, at least it'll be 
better than *that*. :)

Eric