netdev - Re: [PATCH RFC bpf-next 05/20] trait: Replace memcpy calls with inline copies

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <D8CPGEGQ4630.2MKAQH44PFCCO@arthurfabre.com>
Date: Mon, 10 Mar 2025 16:52:30 +0100
From: "Arthur Fabre" <arthur@...hurfabre.com>
To: "Lorenzo Bianconi" <lorenzo.bianconi@...hat.com>
Cc: <netdev@...r.kernel.org>, <bpf@...r.kernel.org>, <jakub@...udflare.com>,
 <hawk@...nel.org>, <yan@...udflare.com>, <jbrandeburg@...udflare.com>,
 <thoiland@...hat.com>, <lbiancon@...hat.com>, "Arthur Fabre"
 <afabre@...udflare.com>
Subject: Re: [PATCH RFC bpf-next 05/20] trait: Replace memcpy calls with
 inline copies

On Mon Mar 10, 2025 at 11:50 AM CET, Lorenzo Bianconi wrote:
> > From: Arthur Fabre <afabre@...udflare.com>
> > 
> > When copying trait values to or from the caller, the size isn't a
> > constant so memcpy() ends up being a function call.
> > 
> > Replace it with an inline implementation that only handles the sizes we
> > support.
> > 
> > We store values "packed", so they won't necessarily be 4 or 8 byte
> > aligned.
> > 
> > Setting and getting traits is roughly ~40% faster.
>
> Nice! I guess in a formal series this patch can be squashed with patch 1/20
> (adding some comments).

Happy to squash and add comments instead if that's better :)

>
> Regards,
> Lorenzo
>
> > 
> > Signed-off-by: Arthur Fabre <afabre@...udflare.com>
> > ---
> >  include/net/trait.h | 25 +++++++++++++++++++------
> >  1 file changed, 19 insertions(+), 6 deletions(-)
> > 
> > diff --git a/include/net/trait.h b/include/net/trait.h
> > index 536b8a17dbbc091b4d1a4d7b4b21c1e36adea86a..d4581a877bd57a32e2ad032147c906764d6d37f8 100644
> > --- a/include/net/trait.h
> > +++ b/include/net/trait.h
> > @@ -7,6 +7,7 @@
> >  #include <linux/errno.h>
> >  #include <linux/string.h>
> >  #include <linux/bitops.h>
> > +#include <linux/unaligned.h>
> >  
> >  /* Traits are a very limited KV store, with:
> >   * - 64 keys (0-63).
> > @@ -145,23 +146,23 @@ int trait_set(void *traits, void *hard_end, u64 key, const void *val, u64 len, u
> >  			memmove(traits + off + len, traits + off, traits_size(traits) - off);
> >  	}
> >  
> > -	/* Set our value. */
> > -	memcpy(traits + off, val, len);
> > -
> > -	/* Store our length in header. */
> >  	u64 encode_len = 0;
> > -
> >  	switch (len) {
> >  	case 2:
> > +		/* Values are least two bytes, so they'll be two byte aligned */
> > +		*(u16 *)(traits + off) = *(u16 *)val;
> >  		encode_len = 1;
> >  		break;
> >  	case 4:
> > +		put_unaligned(*(u32 *)val, (u32 *)(traits + off));
> >  		encode_len = 2;
> >  		break;
> >  	case 8:
> > +		put_unaligned(*(u64 *)val, (u64 *)(traits + off));
> >  		encode_len = 3;
> >  		break;
> >  	}
> > +
> >  	h->high |= (encode_len >> 1) << key;
> >  	h->low |= (encode_len & 1) << key;
> >  	return 0;
> > @@ -201,7 +202,19 @@ int trait_get(void *traits, u64 key, void *val, u64 val_len)
> >  	if (real_len > val_len)
> >  		return -ENOSPC;
> >  
> > -	memcpy(val, traits + off, real_len);
> > +	switch (real_len) {
> > +	case 2:
> > +		/* Values are least two bytes, so they'll be two byte aligned */
> > +		*(u16 *)val = *(u16 *)(traits + off);
> > +		break;
> > +	case 4:
> > +		*(u32 *)val = get_unaligned((u32 *)(traits + off));
> > +		break;
> > +	case 8:
> > +		*(u64 *)val = get_unaligned((u64 *)(traits + off));
> > +		break;
> > +	}
> > +
> >  	return real_len;
> >  }
> >  
> > 
> > -- 
> > 2.43.0
> > 
> >