Re: aio: compat_ioctl issue?

[Jeff Moyer <jmoyer@redhat.com>]

Michael Tokarev <mjt@tls.msk.ru> writes:

> Jeff Moyer wrote:
> []
>> Sorry for taking so long on this.  I only tested the case where niovs >
>> fast_segs, and I missed an obvious thing: I didn't assign the return
>> pointer to the proper iovec.
>
> There's no need to be sorry really.  Because, well, the whole thing isn't
> quite useful anyway: running proper 64bit code is preferable ;)
>
> I actually tried the thing, running a guest right now, which in turn is
> running a quick benchmark and appears to perform quite good at it too.

OK, great.  I'm in the process of unifying the duplicated code, now, so
I might ask for one more sanity check if you have the time and patience
for it.

>> So, this patch should get you going.
>
> Well, I already switched to 64bit kvm binary for my case, and actually
> that one makes alot more sense anyway: there's no conversion like this
> needed, and no 32<=>64bit mode switching either.  (Actually 32bit code
> in this my case is slower elsewhere too).

OK, makes sense, but we should get this right.

> By the way, how about the case when we've several {write,read}v in the
> iocb array?  Will each use the same fast_segs array from the beginning,
> overwriting data of previous iocb element? :)  Just... curious :)

No, each iocb has a built-in iovec which gets specified for the
fast_iov.

> Thank you for your support!
>
> You can add my
> Tested-By: Michael Tokarev <mjt@tls.msk.ru>
> if you want.  Thanks!

Thanks!

-Jeff