linux-kernel - Re: [PATCH rdma-next] RDMA/rdmavt: Decouple QP and SGE lists allocations

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: <YJvGWPimIFbptgdC@unreal>
Date:   Wed, 12 May 2021 15:13:12 +0300
From:   Leon Romanovsky <leon@...nel.org>
To:     Dennis Dalessandro <dennis.dalessandro@...nelisnetworks.com>
Cc:     "Marciniszyn, Mike" <mike.marciniszyn@...nelisnetworks.com>,
        Haakon Bugge <haakon.bugge@...cle.com>,
        Doug Ledford <dledford@...hat.com>,
        Jason Gunthorpe <jgg@...dia.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        OFED mailing list <linux-rdma@...r.kernel.org>
Subject: Re: [PATCH rdma-next] RDMA/rdmavt: Decouple QP and SGE lists
 allocations

On Wed, May 12, 2021 at 12:08:59AM -0400, Dennis Dalessandro wrote:
> 
> On 5/11/21 3:27 PM, Leon Romanovsky wrote:
> > On Tue, May 11, 2021 at 07:15:09PM +0000, Marciniszyn, Mike wrote:
> > > > > 
> > > > > Why not kzalloc_node() here?
> > > 
> > > I agree here.
> > > 
> > > Other allocations that have been promoted to the core have lost the node attribute in the allocation.
> > 
> > Did you notice any performance degradation?
> > 
> 
> So what's the motivation to change it from the way it was originally
> designed? It seems to me if the original implementation went to the trouble
> to allocate the memory on the local node, refactoring the code should
> respect that.

I have no problem to make rdma_zalloc_*() node aware, but would like to get
real performance justification. My assumption is that rdmavt use kzalloc_node
for the control plane based on some internal performance testing and we finally
can see the difference between kzalloc and kzalloc_node, am I right?

Is the claim of performance degradation backed by data?

The main reason (maybe I'm wrong here) is to avoid _node() allocators
because they increase chances of memory allocation failure due to not
doing fallback in case node memory is depleted.

Again, I'm suggesting to do plain kzalloc() for control part of QP.

Thanks

> 
> -Denny