On Sun, Nov 13, 2016 at 08:47:55AM -0700, Jonathan Corbet wrote: >On Wed, 26 Oct 2016 10:41:28 +0800 >Chao Fan <fanc.fnst@xxxxxxxxxxxxxx> wrote: > >> The iowait is not reliable by reading from /proc/stat, so this >> method to get iowait is not suggested. And we mark it in the >> document. > >Sorry for the delay on this. Life has been...challenging... > >> Signed-off-by: Cao Jin <caoj.fnst@xxxxxxxxxxxxxx> >> Signed-off-by: Chao Fan <fanc.fnst@xxxxxxxxxxxxxx> > >What does this signoff chain mean? Hi jon, I made this patch. He is my colleague who helped me a lot on investigating why this value decrease, so I added his s-o-b. I didn't know how to put the position, but both of us are responsible for this patch. Should I change and resend it? > >> --- >> Documentation/filesystems/proc.txt | 11 ++++++++++- >> 1 file changed, 10 insertions(+), 1 deletion(-) >> >> diff --git a/Documentation/filesystems/proc.txt b/Documentation/filesystems/proc.txt >> index 74329fd..71f5096 100644 >> --- a/Documentation/filesystems/proc.txt >> +++ b/Documentation/filesystems/proc.txt >> @@ -1305,7 +1305,16 @@ second). The meanings of the columns are as follows, from left to right: >> - nice: niced processes executing in user mode >> - system: processes executing in kernel mode >> - idle: twiddling thumbs >> -- iowait: waiting for I/O to complete >> +- iowait: In a word, iowait stands for waiting for I/O to complete. But there >> + are several problems: >> + 1. Cpu will not wait for I/O to complete, iowait is the time that a task is >> + waiting for I/O to complete. When cpu goes into idle state for >> + outstanding task io, another task will be scheduled on this CPU. >> + 2. In a multi-core CPU, the task waiting for I/O to complete is not running >> + on any CPU, so the iowait of each CPU is difficult to calculate. >> + 3. The value of iowait field in /proc/stat will decrease in certain >> + conditions. >> + So, the iowait is not reliable by reading from /proc/stat. >> - irq: servicing interrupts > >So I suppose I can apply this. But is there any chance of making it say >what iowait actually measures, rather than just saying that it's >unreliable? Current iowait definition and way of accounting is legacy from UP era, it works well in UP era, but when step into SMP era, both definition and accounting way are not suitable, one example is when task migrating between CPUs. In SMP era, this value should be a global one, not per-cpu. So, it is hard to say what iowait actually measures, the current problem is: a appropriate definition (iowait in SMP) is not clear. Quote someone's words, its accounting is "a steaming pile of crap". Whether the explanation is OK? Thanks, Chao Fan > >Thanks, > >jon > > -- To unsubscribe from this list: send the line "unsubscribe linux-doc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
