Nathan Scott wrote:
>
> hi,
>
> On Dec 6, 8:40am, Ken McDonell wrote:
> > Subject: Re: panic occurs on IA64
> > ...
> >
> > Will mkfs by default choose a blocksize that satisifies the
> > "restriction is that blocksize is tied to PAGE_SIZE &
> > PAGE_CACHE_SIZE"?
> >
>
> no - mkfs default blocksize is always 4K.
>
> it would be fairly trivial to add a mkfs getpagesize(2) check
> and print a warning if these don't match (we could remove it
> later when the work has been done to fix this)... shall I?
>
Indeed. You had the right question, Ken. Yes, we should
make the blocksize default to the pagesize ... regardless
of when we get non-page-sized blocks working, I can imagine
that page-sized blocks will be best performing ignoring
fragmentation issues.
BTW, by "tied to" I meant that the I/O routines only deal
with page-sized blocks ... it's probably not a huge change
to get less-than-page-sized blocks to work (we had 512-byte
blocks working about 6 months back). Large-than-page-sized
blocks is mostly an unknown quantity, implementation-wise.
Finally, a question for Hiroshi: can you please let us know
what block-size you used in mkfs? And, what was the
PAGE_SIZE in your kernel? If you let mkfs take the
defaults, can you please try again with whatever PAGE_SIZE
is for your machine?
thanks,
ananth.
|