Re: [PATCH] mm/vmalloc: Introduce DEBUG_VMALLOCINFO to reduce spinlock contention

[Andrew Morton]

On Thu, 10 Apr 2014 12:40:58 -0400 Richard Yao <ryao@xxxxxxxxxx> wrote:

> Performance analysis of software compilation by Gentoo portage on an
> Intel E5-2620 with 64GB of RAM revealed that a sizeable amount of time,
> anywhere from 5% to 15%, was spent in get_vmalloc_info(), with at least
> 40% of that time spent in the _raw_spin_lock() invoked by it.

This means that something in userspace is beating the crap out of
/proc/meminfo.  What is it and why is it doing this?

/proc/meminfo reads a large amount of stuff and gathering it will
always be expensive.  I don't think we really want to be doing
significant work and adding significant complexity to optimize meminfo.

If there really is a legitimate need to be reading meminfo with this
frequency then it would be pretty simple to optimise
get_vmalloc_info(): all it does is to return two ulongs and we could
maintain those at vmalloc/vfree time rather than doing the big list
walk.

If we can address these things then the vmap_area_lock problem should
just go away - the kernel shouldn't be calling vmalloc/vfree at high
frequency, especially during a compilation workload.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/