Re: [PATCH] Mutilated form of Andi Kleen's AMD prefetch errata patch

[Jamie Lokier <jamie@shareable.org>]

Andrew Morton wrote:
> I'm a bit confused as to what significance the faulting address has btw:
> kernel code can raise prefetch faults against addresses which are less
> than, and presumably greater than TASK_SIZE.

The address itself is not important.

What's important is that if a prefetch erratum fault happens while
mmap_sem or any spinlock are held, then do_page_fault must not try to
acquire mmap_sem.  That can deadlock or schedule; both are bad.

Checking the address is an easy way to ensure that: addr >= TASK_SIZE
implies that no vma lookup is needed, hence mmap_sem is not needed.

There is already an addr >= TASK_SIZE test, so this can be hooked in
without any penalty, in fact it will streamline the existing code a bit.

-- Jamie