From: "Nikolova, Tatyana E" <tatyana.e.nikolova@intel.com>
To: Jason Gunthorpe <jgg@nvidia.com>
Cc: "dledford@redhat.com" <dledford@redhat.com>,
"leon@kernel.org" <leon@kernel.org>,
"linux-rdma@vger.kernel.org" <linux-rdma@vger.kernel.org>
Subject: RE: [PATCH rdma-core] irdma: Restore full memory barrier for doorbell optimization
Date: Thu, 2 Sep 2021 16:27:37 +0000 [thread overview]
Message-ID: <DM6PR11MB4692BD4BB222DAA58C264000CBCE9@DM6PR11MB4692.namprd11.prod.outlook.com> (raw)
In-Reply-To: <20210819224408.GE1721383@nvidia.com>
> -----Original Message-----
> From: Jason Gunthorpe <jgg@nvidia.com>
> Sent: Thursday, August 19, 2021 5:44 PM
> To: Nikolova, Tatyana E <tatyana.e.nikolova@intel.com>
> Cc: dledford@redhat.com; leon@kernel.org; linux-rdma@vger.kernel.org
> Subject: Re: [PATCH rdma-core] irdma: Restore full memory barrier for
> doorbell optimization
>
> On Thu, Aug 19, 2021 at 10:01:50PM +0000, Nikolova, Tatyana E wrote:
> >
> >
> > > From: Jason Gunthorpe <jgg@nvidia.com>
> > > Sent: Wednesday, August 18, 2021 11:50 AM
> > > To: Nikolova, Tatyana E <tatyana.e.nikolova@intel.com>
> > > Cc: dledford@redhat.com; leon@kernel.org; linux-rdma@vger.kernel.org
> > > Subject: Re: [PATCH rdma-core] irdma: Restore full memory barrier
> > > for doorbell optimization
> > >
> > > On Fri, Aug 13, 2021 at 05:25:49PM -0500, Tatyana Nikolova wrote:
> > > > >> 1. Software writing the valid bit in the WQE.
> > > > >> 2. Software reading shadow memory (hw_tail) value.
> > > >
> > > > > You are missing an ordered atomic on this read it looks like
> > > >
> > > > Hi Jason,
> > > >
> > > > Why do you think we need atomic ops in this case? We aren't trying
> > > > to protect from multiple threads but CPU re-ordering of a write
> > > > and a read.
> > >
> > > Which is what the atomics will do.
> > >
> > > Barriers are only appropriate when you can't add atomic markers to
> > > the actual data that needs ordering.
> >
> > Hi Jason,
> >
> > We aren't sure what you mean by atomic markers. We ran a few
> > experiments with atomics, but none of the barriers we tried
> > smp_mb__{before,after}_atomic(), smp_load_acquire() and
> > smp_store_release() translates to a full memory barrier on X86.
>
> Huh? Those are kernel primitives, this is a userspace patch.
>
> Userspace follows the C11 atomics memory model.
>
> So I'd expect
>
> atomic_store_explicit(tail, memory_order_release)
> atomic_load_explicit(tail, memory_order_acquire)
>
> To be the atomics you need. This will ensure that the read/writes to valid
> before the atomics are sequenced correctly, eg no CPU thread can observe
> an updated tail without also observing the set valid.
>
Hi Jason,
We tried these atomic ops as shown bellow, but they don't fix the issue.
atomic_store_explicit(hdr, memory_order_release) atomic_load_explicit(tail, memory_order_acquire)
In assembly they look like this:
//set_64bit_val(wqe, 24, hdr);
atomic_store_explicit((_Atomic(uint64_t) *)(wqe + (24 >> 3)), hdr, memory_order_release);
2130: 49 89 5f 18 mov %rbx,0x18(%r15)
/root/CVL-3.0-V26.4C00390/rdma-core-27.0/build/../providers/irdma/uk.c:747
/root/CVL-3.0-V26.4C00390/rdma-core-27.0/build/../providers/irdma/uk.c:123
temp = atomic_load_explicit((_Atomic(__u64) *)qp->shadow_area, memory_order_acquire);
1c32: 15 00 00 28 84 adc $0x84280000,%eax
However, the following works:
atomic_store_explicit(hdr, memory_order_seq_cst)
//set_64bit_val(wqe, 24, hdr);
atomic_store_explicit((_Atomic(uint64_t) *)(wqe + (24 >> 3)), hdr, memory_order_seq_cst);
2130: 49 89 5f 18 mov %rbx,0x18(%r15)
2134: 0f ae f0 mfence
/root/CVL-3.0-V26.4C00390/rdma-core-27.0/build/../providers/irdma/uk.c:748
atomic_load_explicit(tail, memory_order_seq_cst) - same assembly as with memory_order_acquire
Thank you,
Tatyana
next prev parent reply other threads:[~2021-09-02 16:27 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-08-05 16:11 [PATCH rdma-core] irdma: Restore full memory barrier for doorbell optimization Tatyana Nikolova
2021-08-06 1:28 ` Jason Gunthorpe
2021-08-09 20:07 ` Nikolova, Tatyana E
2021-08-10 11:59 ` Jason Gunthorpe
2021-08-13 22:25 ` Tatyana Nikolova
2021-08-18 16:49 ` Jason Gunthorpe
2021-08-19 22:01 ` Nikolova, Tatyana E
2021-08-19 22:44 ` Jason Gunthorpe
2021-09-02 16:27 ` Nikolova, Tatyana E [this message]
2021-09-02 17:09 ` Jason Gunthorpe
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=DM6PR11MB4692BD4BB222DAA58C264000CBCE9@DM6PR11MB4692.namprd11.prod.outlook.com \
--to=tatyana.e.nikolova@intel.com \
--cc=dledford@redhat.com \
--cc=jgg@nvidia.com \
--cc=leon@kernel.org \
--cc=linux-rdma@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).