lkml.org 
[lkml]   [2014]   [Mar]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2] Pre-emption control for userspace
On 03/25/2014 11:44 AM, Andrew Morton wrote:
> So the procfs file is written in binary format and is read back in
> ascii format. Seems odd.
>
> Perhaps this should all be done as a new syscall rather than some
> procfs thing.
>

I didn't want to add yet another syscall which will then need to be
added to glibc, but I am open to doing it through a syscall if that is
the consensus.

>> + struct preempt_delay {
>> + u32 __user *delay_req; /* delay request flag pointer */
>> + unsigned char delay_granted:1; /* currently in delay */
>> + unsigned char yield_penalty:1; /* failure to yield penalty */
>> + } sched_preempt_delay;
>
> The problem with bitfields is that a write to one bitfield can corrupt
> a concurrent write to the other one. So it's your responsibility to
> provide locking and/or to describe how this race is avoided. A comment
> here in the definition would be a suitable way of addressing this.
>

I do not have a strong reason to use a bitfield, just trying to not use
any more bytes than I need to. If using a char is safer, I would rather
use safer code.

>> + if (delay_req) {
>> + int ret;
>> +
>> + pagefault_disable();
>> + ret = __copy_from_user_inatomic(&delay_req_flag, delay_req,
>> + sizeof(u32));
>> + pagefault_enable();
>
> This all looks rather hacky and unneccesary. Can't we somehow use
> plain old get_user() and avoid such fuss?

get_user() takes longer and can sleep if page fault occurs. I need this
code to be very fast for it to be beneficial and am willing to ignore
page faults since page fault would imply the task has not touched
pre-emption delay request field and hence we can resched safely.

>> +#else
>> +#define delay_resched_task(curr) resched_task(curr)
>
> This could have been implemented in C...
>

Sure, I can do that.

Thanks, Andrew!

--
Khalid



\
 
 \ /
  Last update: 2014-03-25 21:01    [W:0.196 / U:0.204 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site