> All the patches for fixing buffer cache problems were against 2.0.xx, while
> not all of them have been made against 2.1.xx and actually have made it into
> 2.1.xx. So, at the moment I'm experiencing buffer cache (supposedly)
> problems with 2.1.62 on sparc32 with 48M of RAM, once my filesystem gets
> hosed up, fsck never finishes his job under 2.1.62, while under 2.0.30 it
> does. It locks up (irq go, but current task doesn't change any more) when
> memory used in the system (most of it by fsck) gets close to freepages
> limits (in my case 88 132 176).
>
> Can anyone who was writing those patches look at current 2.1.62 and check if
> all their changes have made it in or have been solved in some other way?
Hi Jakub,
>From your post I wasn't sure why you suspect problems in buffer cache.
You might want to put some printks to see whether you're actually using
up pages with atomic allocations at the time the problem occurs.
I do have some further patches to the buffer code that I'll post soon.
Regards,
Bill