Re: [PATCH] fdtable: Eradicate fdarray overflow.

[Eric Dumazet]

Vadim Lobanov a écrit :
> On Wednesday 11 October 2006 22:19, Eric Dumazet wrote:
> > Hi Vadim
> >
> > I find your PAGE_SIZE/4 minimum allocation quite unjustified.
> >
> > For architectures with 64K PAGE_SIZE, we endup allocating 16K, for poor
> > tasks that happen to touch a not so high (>= 64) file descriptor...
> >
> > I would vote for a fixed size, like 1024
>
> In my opinion, always picking 1024 would be highly suboptimal for some 
> architectures (x86-64 in particular -- that's a whole page, just for the 
> fdarray!). If anything, I'd prefer something similar to this pseudo-code:

I was speaking of 1024 bytes.

I was the guy who made fdset going from PAGE_SIZE to 64 bytes (L1_CACHE_BYTES 
if you dare), I wont be the guy responsible for a reverse path on fdtable :)

That is replace your (PAGE_SIZE/4)  by 1024, wich was you probably meant
No archi has a smaller page, so no need to play with min_t() macro...

> #define FDTABLE_MIN min_t(uint, PAGE_SIZE / 4 / sizeof(struct file *), 1024)
> ...
> nr /= FDTABLE_MIN;
> nr = roundup_pow_of_two(nr + 1);
> nr *= FDTABLE_MIN;
>
> gcc should be smart enough to optimize that expression into a single constant. 
> At least it did (version 4.1.0) in my quick test here.
>
> > Eric
>
> Let me know what you think. Please don't just go radio-silent on me. ;)

radio-silent ? well, it seems I already sent you many mails about your patches :)

Eric

-
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/