Re: [RFC 3/3] nvme: wire up support for async passthrough

[Kanchan Joshi <joshiiitr@gmail.com>]

On Fri, Mar 5, 2021 at 8:11 AM Keith Busch <kbusch@kernel.org> wrote:
>
> On Thu, Mar 04, 2021 at 04:31:11PM +0530, Kanchan Joshi wrote:
> > On Thu, Mar 4, 2021 at 3:14 AM Chaitanya Kulkarni
> > <Chaitanya.Kulkarni@wdc.com> wrote:
> > >
> > > On 3/2/21 23:22, Kanchan Joshi wrote:
> > > > +     if (!ioucmd)
> > > > +             cptr = &c;
> > > > +     else {
> > > > +             /*for async - allocate cmd dynamically */
> > > > +             cptr = kmalloc(sizeof(struct nvme_command), GFP_KERNEL);
> > > > +             if (!cptr)
> > > > +                     return -ENOMEM;
> > > > +     }
> > > > +
> > > > +     memset(cptr, 0, sizeof(c));
> > > Why not kzalloc and remove memset() ?
> >
> > Yes sure. Ideally I want to get rid of the allocation cost. Perhaps
> > employing kmem_cache/mempool can help. Do you think there is a better
> > way?
>
> I'll need to think on this to consider if the memory cost is worth it
> (8b to 64b), but you could replace nvme_request's 'struct nvme_command'
> pointer with the struct itself and not have to allocate anything per IO.
> An added bonus is that sync and async handling become more the same.

Indeed, thanks for this. I will fold that change as a prep in the next version.