linux-kernel - Re: [PATCH net-next 1/9] netmem: introduce struct netmem

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] [day] [month] [year] [list]

Message-ID: <20250620041208.GA11405@system.software.com>
Date: Fri, 20 Jun 2025 13:12:08 +0900
From: Byungchul Park <byungchul@...com>
To: Mina Almasry <almasrymina@...gle.com>
Cc: Jakub Kicinski <kuba@...nel.org>, willy@...radead.org,
	netdev@...r.kernel.org, linux-kernel@...r.kernel.org,
	linux-mm@...ck.org, kernel_team@...ynix.com,
	ilias.apalodimas@...aro.org, harry.yoo@...cle.com, hawk@...nel.org,
	akpm@...ux-foundation.org, davem@...emloft.net,
	john.fastabend@...il.com, andrew+netdev@...n.ch,
	asml.silence@...il.com, toke@...hat.com, tariqt@...dia.com,
	edumazet@...gle.com, pabeni@...hat.com, saeedm@...dia.com,
	leon@...nel.org, ast@...nel.org, daniel@...earbox.net,
	david@...hat.com, lorenzo.stoakes@...cle.com,
	Liam.Howlett@...cle.com, vbabka@...e.cz, rppt@...nel.org,
	surenb@...gle.com, mhocko@...e.com, horms@...nel.org,
	linux-rdma@...r.kernel.org, bpf@...r.kernel.org,
	vishal.moola@...il.com
Subject: Re: [PATCH net-next 1/9] netmem: introduce struct netmem_desc
 mirroring struct page

On Fri, Jun 13, 2025 at 07:19:07PM -0700, Mina Almasry wrote:
> On Thu, Jun 12, 2025 at 6:13 PM Byungchul Park <byungchul@...com> wrote:
> >
> > On Wed, Jun 11, 2025 at 06:55:42PM -0700, Jakub Kicinski wrote:
> > > On Tue, 10 Jun 2025 10:30:01 +0900 Byungchul Park wrote:
> > > > > What's the intended relation between the types?
> > > >
> > > > One thing I'm trying to achieve is to remove pp fields from struct page,
> > > > and make network code use struct netmem_desc { pp fields; } instead of
> > > > sturc page for that purpose.
> > > >
> > > > The reason why I union'ed it with the existing pp fields in struct
> > > > net_iov *temporarily* for now is, to fade out the existing pp fields
> > > > from struct net_iov so as to make the final form like:
> > >
> > > I see, I may have mixed up the complaints there. I thought the effort
> > > was also about removing the need for the ref count. And Rx is
> > > relatively light on use of ref counting.
> > >
> > > > > netmem_ref exists to clearly indicate that memory may not be readable.
> > > > > Majority of memory we expect to allocate from page pool must be
> > > > > kernel-readable. What's the plan for reading the "single pointer"
> > > > > memory within the kernel?
> > > > >
> > > > > I think you're approaching this problem from the easiest and least
> > > >
> > > > No, I've never looked for the easiest way.  My bad if there are a better
> > > > way to achieve it.  What would you recommend?
> > >
> > > Sorry, I don't mean that the approach you took is the easiest way out.
> > > I meant that between Rx and Tx handling Rx is the easier part because
> > > we already have the suitable abstraction. It's true that we use more
> > > fields in page struct on Rx, but I thought Tx is also more urgent
> > > as there are open reports for networking taking references on slab
> > > pages.
> > >
> > > In any case, please make sure you maintain clear separation between
> > > readable and unreadable memory in the code you produce.
> >
> > Do you mean the current patches do not?  If yes, please point out one
> > as example, which would be helpful to extract action items.
> >
> 
> I think one thing we could do to improve separation between readable
> (pages/netmem_desc) and unreadable (net_iov) is to remove the struct
> netmem_desc field inside the net_iov, and instead just duplicate the
> pp/pp_ref_count/etc fields. The current code gives off the impression
> that net_iov may be a container of netmem_desc which is not really
> accurate.
> 
> But I don't think that's a major blocker. I think maybe the real issue
> is that there are no reviews from any mm maintainers? So I'm not 100%
> sure this is in line with their memdesc plans. I think probably
> patches 2->8 are generic netmem-ifications that are good to merge
> anyway, but I would say patch 1 and 9 need a reviewed by from someone
> on the mm side. Just my 2 cents.
> 
> Btw, this series has been marked as changes requested on patchwork, so
> it is in need of a respin one way or another:

Some can be improved but the others not.  For example:

   +static noinline netmem_ref __page_pool_alloc_netmems_slow(struct page_pool *pool,
   +							  gfp_t gfp)

It complains about the long line length but no idea how to avoid it :(
I can do nothing but to ignore..

	Byungchul

> https://patchwork.kernel.org/project/netdevbpf/list/?series=&submitter=byungchul&state=*&q=&archive=&delegate=
> 
> https://docs.kernel.org/process/maintainer-netdev.html#patch-status
> 
> -- 
> Thanks,
> Mina