Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 13 Jun 2022 07:42:45 -0700
From:   Joe Damato <jdamato@fastly.com>
To:     David Laight <David.Laight@ACULAB.COM>
Cc:     Alexander Viro <viro@zeniv.linux.org.uk>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: [RFC,iov_iter v2 3/8] iov_iter: add copyin_iovec helper
Message-ID: <20220613144244.GA77534@fastly.com>
References: <1655024280-23827-1-git-send-email-jdamato@fastly.com>
 <1655024280-23827-4-git-send-email-jdamato@fastly.com>
 <8f6771a01fe74cde86641e780d31a2ce@AcuMS.aculab.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <8f6771a01fe74cde86641e780d31a2ce@AcuMS.aculab.com>
User-Agent: Mutt/1.5.24 (2015-08-30)
Precedence: bulk

On Mon, Jun 13, 2022 at 07:53:19AM +0000, David Laight wrote:
> From: Joe Damato
> > Sent: 12 June 2022 09:58
> > 
> > copyin_iovec is a helper which wraps copyin and selects the right copy
> > method based on the iter_copy_type.
> 
> A pretty bad description.

Thanks, David. I'll be sure to fix the commit description in the next
revision.

> > Signed-off-by: Joe Damato <jdamato@fastly.com>
> > ---
> >  lib/iov_iter.c | 19 ++++++++++++++-----
> >  1 file changed, 14 insertions(+), 5 deletions(-)
> > 
> > diff --git a/lib/iov_iter.c b/lib/iov_iter.c
> > index d32d7e5..6720cb2 100644
> > --- a/lib/iov_iter.c
> > +++ b/lib/iov_iter.c
> > @@ -168,6 +168,15 @@ static int copyin(void *to, const void __user *from, size_t n)
> >  	return n;
> >  }
> > 
> > +static int copyin_iovec(void *to, const void __user *from, size_t n,
> > +			struct iov_iter *i)
> > +{
> > +	if (unlikely(iov_iter_copy_is_nt(i)))
> > +		return __copy_from_user_nocache(to, from, n);
> > +	else
> > +		return copyin(to, from, n);
> > +}
> 
> Isn't this extra conditional going to have a measurable impact
> on all the normal copy paths?

The kernel already does a conditional for tx-nocache-copy on TCP sockets
when copying skbs to check for the NETIF_F_NOCACHE_COPY bit, but I hear
what you are saying.

I suppose I could push the NT copy check logic out of iov_iter, but to do
that I think I'd probably have to significantly refactor the iov code to
break apart copy_page_from_iter_iovec.

I'll spend a bit more time thinking through this, but I'm open to
suggestions if you have one; the benefit of supporting non-temporal copies
in this path is pretty significant, so I hope a path forward can be found.

> The additional costs of all the 'iovec' types is bad enough
> already.

Do you have data you can share on this?

Thanks for taking a look!