Messages in this thread | | | Date | Fri, 28 Mar 2008 13:24:56 -0700 (PDT) | From | Linus Torvalds <> | Subject | Re: 2.6.25-rc7-git2: Reported regressions from 2.6.24 |
| |
On Fri, 28 Mar 2008, Pekka Enberg wrote: > > Because I didn't spot that we're calling kmalloc_large() in the fallback path > too. I was referring to the page allocator pass-through kmalloc_large() call > that happens before we hit __slab_alloc. My bad, sorry.
So this is kind of my point of the whole thing. It's really hard to even realize that there are multiple points - unless you do it at the top level.
And yeah, the double clearing is just a performance issue in a rare and odd case, and if it hadn't been for the very anal debug checks we would never have even noticed. But code that is hard to follow and has really subtle non-local implications is fundamentally more likely to have these kinds of problems. Which is why I'm arguing against code like that.
Linus
| |