Re: [PATCH v2 03/28] dcache: convert dentry_stat.nr_unused to per-cpu counters

[Dave Chinner <david-FqsqvQoI3Ljby3iVrkZq2A@public.gmane.org>]

On Mon, Apr 08, 2013 at 01:14:48PM +0400, Glauber Costa wrote:
> On 04/05/2013 05:15 AM, Dave Chinner wrote:
> > On Thu, Apr 04, 2013 at 06:09:31PM -0700, Greg Thelen wrote:
> >> On Fri, Mar 29 2013, Glauber Costa wrote:
> >>
> >>> From: Dave Chinner <dchinner-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
> >>>
> >>> Before we split up the dcache_lru_lock, the unused dentry counter
> >>> needs to be made independent of the global dcache_lru_lock. Convert
> >>> it to per-cpu counters to do this.
> >>>
> >>> Signed-off-by: Dave Chinner <dchinner-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
> >>> Reviewed-by: Christoph Hellwig <hch-jcswGhMUV9g@public.gmane.org>
> >>> ---
> >>>  fs/dcache.c | 17 ++++++++++++++---
> >>>  1 file changed, 14 insertions(+), 3 deletions(-)
> >>>
> >>> diff --git a/fs/dcache.c b/fs/dcache.c
> >>> index fbfae008..f1196f2 100644
> >>> --- a/fs/dcache.c
> >>> +++ b/fs/dcache.c
> >>> @@ -118,6 +118,7 @@ struct dentry_stat_t dentry_stat = {
> >>>  };
> >>>  
> >>>  static DEFINE_PER_CPU(unsigned int, nr_dentry);
> >>> +static DEFINE_PER_CPU(unsigned int, nr_dentry_unused);
> >>>  
> >>>  #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
> >>>  static int get_nr_dentry(void)
> >>> @@ -129,10 +130,20 @@ static int get_nr_dentry(void)
> >>>  	return sum < 0 ? 0 : sum;
> >>>  }
> >>>  
> >>> +static int get_nr_dentry_unused(void)
> >>> +{
> >>> +	int i;
> >>> +	int sum = 0;
> >>> +	for_each_possible_cpu(i)
> >>> +		sum += per_cpu(nr_dentry_unused, i);
> >>> +	return sum < 0 ? 0 : sum;
> >>> +}
> >>
> >> Just checking...  If cpu x is removed, then its per cpu nr_dentry_unused
> >> count survives so we don't leak nr_dentry_unused.  Right?  I see code in
> >> percpu_counter_sum_positive() to explicitly handle this case and I want
> >> to make sure we don't need it here.
> > 
> > DEFINE_PER_CPU() gives a variable per possible CPU, and we sum for
> > all possible CPUs. Therefore online/offline CPUs just don't matter.
> > 
> > The percpu_counter code uses for_each_online_cpu(), and so it has to
> > be aware of hotplug operations so taht it doesn't leak counts.
> 
> It is an unsigned quantity, however. Can't we go negative if it becomes
> unused in one cpu, but used in another?

Sure, but it's unsigned for the purposes of summing, not for the
purposes of having pos/neg values - they are just delta counters.

I'm just copying the code from fs/inode.c. I originally implemented
the fs/inode.c code using generic per-cpu counters, but there was
a hissy fit over "too much overhead" and so someone implemented
their own lightweight version. I've just copied the existing code to
code because I don't care to revisit this....

> Ex:
> 
> nr_unused/0: 0
> nr_unused/1: 0
> 
> dentry goes to the LRU at cpu 1:
> nr_unused/0: 0
> nr_unused/1: 1
> 
> CPU 1 goes down:
> nr_unused/0: 0

why?

> dentry goes out of the LRU at cpu 0:
> nr_unused/0: 1 << 32.

Sorry, where does that shift come from? Pulling from the LRU is just
a simple subtraction. (i.e. 0 - 1 = 0xffffffff), and so
when we sum them all up:

nr_unused/0: 1
nr_unused/0: -1 (0xffffffff)

sum = 1 + 0xffffffff = 0

> That would easily be fixed by using a normal signed long, and is in fact
> what the percpu code does in its internal operations.

Changing it to a long means it becomes at 64 bit value on 64 bit
machines (doubling memory usage), and now you're summing a 64 bit
values into a 32 bit integer. Something else to go wrong....

> Any reason not to do it? Something I am not seeing?

It's a direct copy of the counting code in fs/inode.c. That has not
demonstrated any problems in all my monitoring for the past coupl
eof years  (these are userspace visible stats) so AFAICT this code
is just fine...

Cheers,

Dave.
-- 
Dave Chinner
david-FqsqvQoI3Ljby3iVrkZq2A@public.gmane.org