From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id D5663C433F5 for ; Wed, 3 Nov 2021 11:36:01 +0000 (UTC) Received: from lists.gnu.org (lists.gnu.org [209.51.188.17]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.kernel.org (Postfix) with ESMTPS id 24777610E8 for ; Wed, 3 Nov 2021 11:36:01 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.4.1 mail.kernel.org 24777610E8 Authentication-Results: mail.kernel.org; dmarc=fail (p=none dis=none) header.from=redhat.com Authentication-Results: mail.kernel.org; spf=pass smtp.mailfrom=nongnu.org Received: from localhost ([::1]:51986 helo=lists1p.gnu.org) by lists.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1miEYh-0002La-SA for qemu-devel@archiver.kernel.org; Wed, 03 Nov 2021 07:35:59 -0400 Received: from eggs.gnu.org ([2001:470:142:3::10]:58560) by lists.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1miEXE-00019F-Mx for qemu-devel@nongnu.org; Wed, 03 Nov 2021 07:34:28 -0400 Received: from us-smtp-delivery-124.mimecast.com ([170.10.133.124]:20193) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1miEX9-0006WK-LK for qemu-devel@nongnu.org; Wed, 03 Nov 2021 07:34:27 -0400 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=redhat.com; s=mimecast20190719; t=1635939261; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: in-reply-to:in-reply-to:references:references; bh=84y9xvgroLEy2mY4wlZBqwN4qObJCFXhOJQzTROD0zE=; b=Ul/bEMnxhpbTIBV/u4JGV2k1FRRJ/o4PzY7omUSw2IXYOoYRiusrH91v+h17+yK9bPpvnI 3mgHVBhkVpD5KSJ8lE2OlJlhyuxsDHD5FgWv6C6z7rJKWKl2kVWTOsanoBvsDYP2DZR6fK IXQglFMny5QAIrYNKqeuEYknj9d1M7Q= Received: from mimecast-mx01.redhat.com (mimecast-mx01.redhat.com [209.132.183.4]) (Using TLS) by relay.mimecast.com with ESMTP id us-mta-116-4qMVOsRwNYyTAPWfCw2umw-1; Wed, 03 Nov 2021 07:34:18 -0400 X-MC-Unique: 4qMVOsRwNYyTAPWfCw2umw-1 Received: from smtp.corp.redhat.com (int-mx05.intmail.prod.int.phx2.redhat.com [10.5.11.15]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mimecast-mx01.redhat.com (Postfix) with ESMTPS id 731C2100C661; Wed, 3 Nov 2021 11:34:16 +0000 (UTC) Received: from localhost (unknown [10.39.194.230]) by smtp.corp.redhat.com (Postfix) with ESMTP id BFEF86A905; Wed, 3 Nov 2021 11:33:34 +0000 (UTC) Date: Wed, 3 Nov 2021 11:33:33 +0000 From: Stefan Hajnoczi To: Christian Schoenebeck Subject: Re: [PATCH v2 0/3] virtio: increase VIRTQUEUE_MAX_SIZE to 32k Message-ID: References: <10221570.6MffRmy8Bz@silver> <136317874.cAzoFBZ3UR@silver> MIME-Version: 1.0 In-Reply-To: <136317874.cAzoFBZ3UR@silver> X-Scanned-By: MIMEDefang 2.79 on 10.5.11.15 Authentication-Results: relay.mimecast.com; auth=pass smtp.auth=CUSA124A263 smtp.mailfrom=stefanha@redhat.com X-Mimecast-Spam-Score: 0 X-Mimecast-Originator: redhat.com Content-Type: multipart/signed; micalg=pgp-sha256; protocol="application/pgp-signature"; boundary="M4Sxhk+xVfddXjVp" Content-Disposition: inline Received-SPF: pass client-ip=170.10.133.124; envelope-from=stefanha@redhat.com; helo=us-smtp-delivery-124.mimecast.com X-Spam_score_int: -34 X-Spam_score: -3.5 X-Spam_bar: --- X-Spam_report: (-3.5 / 5.0 requ) BAYES_00=-1.9, DKIMWL_WL_HIGH=-0.717, DKIM_SIGNED=0.1, DKIM_VALID=-0.1, DKIM_VALID_AU=-0.1, DKIM_VALID_EF=-0.1, RCVD_IN_DNSWL_LOW=-0.7, RCVD_IN_MSPIKE_H2=-0.001, SPF_HELO_NONE=0.001, SPF_PASS=-0.001 autolearn=ham autolearn_force=no X-Spam_action: no action X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: Kevin Wolf , Laurent Vivier , qemu-block@nongnu.org, "Michael S. Tsirkin" , Jason Wang , Amit Shah , David Hildenbrand , qemu-devel@nongnu.org, Greg Kurz , virtio-fs@redhat.com, Eric Auger , Hanna Reitz , "Gonglei \(Arei\)" , Gerd Hoffmann , =?iso-8859-1?Q?Marc-Andr=E9?= Lureau , Paolo Bonzini , Fam Zheng , Raphael Norwitz , "Dr. David Alan Gilbert" Errors-To: qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org Sender: "Qemu-devel" --M4Sxhk+xVfddXjVp Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Mon, Nov 01, 2021 at 09:29:26PM +0100, Christian Schoenebeck wrote: > On Donnerstag, 28. Oktober 2021 11:00:48 CET Stefan Hajnoczi wrote: > > On Mon, Oct 25, 2021 at 05:03:25PM +0200, Christian Schoenebeck wrote: > > > On Montag, 25. Oktober 2021 12:30:41 CEST Stefan Hajnoczi wrote: > > > > On Thu, Oct 21, 2021 at 05:39:28PM +0200, Christian Schoenebeck wro= te: > > > > > On Freitag, 8. Oktober 2021 18:08:48 CEST Christian Schoenebeck w= rote: > > > > > > On Freitag, 8. Oktober 2021 16:24:42 CEST Christian Schoenebeck= wrote: > > > > > > > On Freitag, 8. Oktober 2021 09:25:33 CEST Greg Kurz wrote: > > > > > > > > On Thu, 7 Oct 2021 16:42:49 +0100 > > > > > > > >=20 > > > > > > > > Stefan Hajnoczi wrote: > > > > > > > > > On Thu, Oct 07, 2021 at 02:51:55PM +0200, Christian Schoe= nebeck wrote: > > > > > > > > > > On Donnerstag, 7. Oktober 2021 07:23:59 CEST Stefan Haj= noczi wrote: > > > > > > > > > > > On Mon, Oct 04, 2021 at 09:38:00PM +0200, Christian > > > > > > > > > > > Schoenebeck > > > > > >=20 > > > > > > wrote: > > > > > > > > > > > > At the moment the maximum transfer size with virtio= is > > > > > > > > > > > > limited > > > > > > > > > > > > to > > > > > > > > > > > > 4M > > > > > > > > > > > > (1024 * PAGE_SIZE). This series raises this limit t= o its > > > > > > > > > > > > maximum > > > > > > > > > > > > theoretical possible transfer size of 128M (32k pag= es) > > > > > > > > > > > > according > > > > > > > > > > > > to > > > > > > > > > > > > the > > > > > > > > > > > > virtio specs: > > > > > > > > > > > >=20 > > > > > > > > > > > > https://docs.oasis-open.org/virtio/virtio/v1.1/cs01= /virt > > > > > > > > > > > > io-v > > > > > > > > > > > > 1.1- > > > > > > > > > > > > cs > > > > > > > > > > > > 01 > > > > > > > > > > > > .html# > > > > > > > > > > > > x1-240006 > > > > > > > > > > >=20 > > > > > > > > > > > Hi Christian, > > > > > > > >=20 > > > > > > > > > > > I took a quick look at the code: > > > > > > > > Hi, > > > > > > > >=20 > > > > > > > > Thanks Stefan for sharing virtio expertise and helping Chri= stian > > > > > > > > ! > > > > > > > >=20 > > > > > > > > > > > - The Linux 9p driver restricts descriptor chains to = 128 > > > > > > > > > > > elements > > > > > > > > > > >=20 > > > > > > > > > > > (net/9p/trans_virtio.c:VIRTQUEUE_NUM) > > > > > > > > > >=20 > > > > > > > > > > Yes, that's the limitation that I am about to remove (W= IP); > > > > > > > > > > current > > > > > > > > > > kernel > > > > > > > > > > patches: > > > > > > > > > > https://lore.kernel.org/netdev/cover.1632327421.git.lin= ux_os > > > > > > > > > > s@cr > > > > > > > > > > udeb > > > > > > > > > > yt > > > > > > > > > > e. > > > > > > > > > > com/> > > > > > > > > >=20 > > > > > > > > > I haven't read the patches yet but I'm concerned that tod= ay > > > > > > > > > the > > > > > > > > > driver > > > > > > > > > is pretty well-behaved and this new patch series introduc= es a > > > > > > > > > spec > > > > > > > > > violation. Not fixing existing spec violations is okay, b= ut > > > > > > > > > adding > > > > > > > > > new > > > > > > > > > ones is a red flag. I think we need to figure out a clean > > > > > > > > > solution. > > > > > > >=20 > > > > > > > Nobody has reviewed the kernel patches yet. My main concern > > > > > > > therefore > > > > > > > actually is that the kernel patches are already too complex, > > > > > > > because > > > > > > > the > > > > > > > current situation is that only Dominique is handling 9p patch= es on > > > > > > > kernel > > > > > > > side, and he barely has time for 9p anymore. > > > > > > >=20 > > > > > > > Another reason for me to catch up on reading current kernel c= ode > > > > > > > and > > > > > > > stepping in as reviewer of 9p on kernel side ASAP, independen= t of > > > > > > > this > > > > > > > issue. > > > > > > >=20 > > > > > > > As for current kernel patches' complexity: I can certainly dr= op > > > > > > > patch > > > > > > > 7 > > > > > > > entirely as it is probably just overkill. Patch 4 is then the > > > > > > > biggest > > > > > > > chunk, I have to see if I can simplify it, and whether it wou= ld > > > > > > > make > > > > > > > sense to squash with patch 3. > > > > > > >=20 > > > > > > > > > > > - The QEMU 9pfs code passes iovecs directly to preadv= (2) > > > > > > > > > > > and > > > > > > > > > > > will > > > > > > > > > > > fail > > > > > > > > > > >=20 > > > > > > > > > > > with EINVAL when called with more than IOV_MAX iove= cs > > > > > > > > > > > (hw/9pfs/9p.c:v9fs_read()) > > > > > > > > > >=20 > > > > > > > > > > Hmm, which makes me wonder why I never encountered this > > > > > > > > > > error > > > > > > > > > > during > > > > > > > > > > testing. > > > > > > > > > >=20 > > > > > > > > > > Most people will use the 9p qemu 'local' fs driver back= end > > > > > > > > > > in > > > > > > > > > > practice, > > > > > > > > > > so > > > > > > > > > > that v9fs_read() call would translate for most people t= o > > > > > > > > > > this > > > > > > > > > > implementation on QEMU side (hw/9p/9p-local.c): > > > > > > > > > >=20 > > > > > > > > > > static ssize_t local_preadv(FsContext *ctx, V9fsFidOpen= State > > > > > > > > > > *fs, > > > > > > > > > >=20 > > > > > > > > > > const struct iovec *iov, > > > > > > > > > > int iovcnt, off_t offset) > > > > > > > > > >=20 > > > > > > > > > > { > > > > > > > > > > #ifdef CONFIG_PREADV > > > > > > > > > >=20 > > > > > > > > > > return preadv(fs->fd, iov, iovcnt, offset); > > > > > > > > > >=20 > > > > > > > > > > #else > > > > > > > > > >=20 > > > > > > > > > > int err =3D lseek(fs->fd, offset, SEEK_SET); > > > > > > > > > > if (err =3D=3D -1) { > > > > > > > > > > =20 > > > > > > > > > > return err; > > > > > > > > > > =20 > > > > > > > > > > } else { > > > > > > > > > > =20 > > > > > > > > > > return readv(fs->fd, iov, iovcnt); > > > > > > > > > > =20 > > > > > > > > > > } > > > > > > > > > >=20 > > > > > > > > > > #endif > > > > > > > > > > } > > > > > > > > > >=20 > > > > > > > > > > > Unless I misunderstood the code, neither side can tak= e > > > > > > > > > > > advantage > > > > > > > > > > > of > > > > > > > > > > > the > > > > > > > > > > > new 32k descriptor chain limit? > > > > > > > > > > >=20 > > > > > > > > > > > Thanks, > > > > > > > > > > > Stefan > > > > > > > > > >=20 > > > > > > > > > > I need to check that when I have some more time. One > > > > > > > > > > possible > > > > > > > > > > explanation > > > > > > > > > > might be that preadv() already has this wrapped into a = loop > > > > > > > > > > in > > > > > > > > > > its > > > > > > > > > > implementation to circumvent a limit like IOV_MAX. It m= ight > > > > > > > > > > be > > > > > > > > > > another > > > > > > > > > > "it > > > > > > > > > > works, but not portable" issue, but not sure. > > > > > > > > > >=20 > > > > > > > > > > There are still a bunch of other issues I have to resol= ve. > > > > > > > > > > If > > > > > > > > > > you > > > > > > > > > > look > > > > > > > > > > at > > > > > > > > > > net/9p/client.c on kernel side, you'll notice that it > > > > > > > > > > basically > > > > > > > > > > does > > > > > > > > > > this ATM> > > > > > > > > > > >=20 > > > > > > > > > > kmalloc(msize); > > > > > > > >=20 > > > > > > > > Note that this is done twice : once for the T message (clie= nt > > > > > > > > request) > > > > > > > > and > > > > > > > > once for the R message (server answer). The 9p driver could > > > > > > > > adjust > > > > > > > > the > > > > > > > > size > > > > > > > > of the T message to what's really needed instead of allocat= ing > > > > > > > > the > > > > > > > > full > > > > > > > > msize. R message size is not known though. > > > > > > >=20 > > > > > > > Would it make sense adding a second virtio ring, dedicated to > > > > > > > server > > > > > > > responses to solve this? IIRC 9p server already calculates > > > > > > > appropriate > > > > > > > exact sizes for each response type. So server could just push > > > > > > > space > > > > > > > that's > > > > > > > really needed for its responses. > > > > > > >=20 > > > > > > > > > > for every 9p request. So not only does it allocate much= more > > > > > > > > > > memory > > > > > > > > > > for > > > > > > > > > > every request than actually required (i.e. say 9pfs was > > > > > > > > > > mounted > > > > > > > > > > with > > > > > > > > > > msize=3D8M, then a 9p request that actually would just = need 1k > > > > > > > > > > would > > > > > > > > > > nevertheless allocate 8M), but also it allocates > > > > > > > > > > > PAGE_SIZE, > > > > > > > > > > which > > > > > > > > > > obviously may fail at any time.> > > > > > > > > >=20 > > > > > > > > > The PAGE_SIZE limitation sounds like a kmalloc() vs vmall= oc() > > > > > > > > > situation. > > > > > > >=20 > > > > > > > Hu, I didn't even consider vmalloc(). I just tried the kvmall= oc() > > > > > > > wrapper > > > > > > > as a quick & dirty test, but it crashed in the same way as > > > > > > > kmalloc() > > > > > > > with > > > > > > > large msize values immediately on mounting: > > > > > > >=20 > > > > > > > diff --git a/net/9p/client.c b/net/9p/client.c > > > > > > > index a75034fa249b..cfe300a4b6ca 100644 > > > > > > > --- a/net/9p/client.c > > > > > > > +++ b/net/9p/client.c > > > > > > > @@ -227,15 +227,18 @@ static int parse_opts(char *opts, struc= t > > > > > > > p9_client > > > > > > > *clnt) > > > > > > >=20 > > > > > > > static int p9_fcall_init(struct p9_client *c, struct p9_fcal= l > > > > > > > *fc, > > > > > > > =20 > > > > > > > int alloc_msize) > > > > > > > =20 > > > > > > > { > > > > > > >=20 > > > > > > > - if (likely(c->fcall_cache) && alloc_msize =3D=3D c->m= size) { > > > > > > > + //if (likely(c->fcall_cache) && alloc_msize =3D=3D c-= >msize) { > > > > > > > + if (false) { > > > > > > >=20 > > > > > > > fc->sdata =3D kmem_cache_alloc(c->fcall_cache= , > > > > > > > GFP_NOFS); > > > > > > > fc->cache =3D c->fcall_cache; > > > > > > > =20 > > > > > > > } else { > > > > > > >=20 > > > > > > > - fc->sdata =3D kmalloc(alloc_msize, GFP_NOFS); > > > > > > > + fc->sdata =3D kvmalloc(alloc_msize, GFP_NOFS)= ; > > > > > >=20 > > > > > > Ok, GFP_NOFS -> GFP_KERNEL did the trick. > > > > > >=20 > > > > > > Now I get: > > > > > > virtio: bogus descriptor or out of resources > > > > > >=20 > > > > > > So, still some work ahead on both ends. > > > > >=20 > > > > > Few hacks later (only changes on 9p client side) I got this runni= ng > > > > > stable > > > > > now. The reason for the virtio error above was that kvmalloc() re= turns > > > > > a > > > > > non-logical kernel address for any kvmalloc(>4M), i.e. an address= that > > > > > is > > > > > inaccessible from host side, hence that "bogus descriptor" messag= e by > > > > > QEMU. > > > > > So I had to split those linear 9p client buffers into sparse ones= (set > > > > > of > > > > > individual pages). > > > > >=20 > > > > > I tested this for some days with various virtio transmission size= s and > > > > > it > > > > > works as expected up to 128 MB (more precisely: 128 MB read space= + > > > > > 128 MB > > > > > write space per virtio round trip message). > > > > >=20 > > > > > I did not encounter a show stopper for large virtio transmission = sizes > > > > > (4 MB ... 128 MB) on virtio level, neither as a result of testing= , nor > > > > > after reviewing the existing code. > > > > >=20 > > > > > About IOV_MAX: that's apparently not an issue on virtio level. Mo= st of > > > > > the > > > > > iovec code, both on Linux kernel side and on QEMU side do not hav= e > > > > > this > > > > > limitation. It is apparently however indeed a limitation for user= land > > > > > apps > > > > > calling the Linux kernel's syscalls yet. > > > > >=20 > > > > > Stefan, as it stands now, I am even more convinced that the upper > > > > > virtio > > > > > transmission size limit should not be squeezed into the queue siz= e > > > > > argument of virtio_add_queue(). Not because of the previous argum= ent > > > > > that > > > > > it would waste space (~1MB), but rather because they are two diff= erent > > > > > things. To outline this, just a quick recap of what happens exact= ly > > > > > when > > > > > a bulk message is pushed over the virtio wire (assuming virtio "s= plit" > > > > > layout here): > > > > >=20 > > > > > ---------- [recap-start] ---------- > > > > >=20 > > > > > For each bulk message sent guest <-> host, exactly *one* of the > > > > > pre-allocated descriptors is taken and placed (subsequently) into > > > > > exactly > > > > > *one* position of the two available/used ring buffers. The actual > > > > > descriptor table though, containing all the DMA addresses of the > > > > > message > > > > > bulk data, is allocated just in time for each round trip message.= Say, > > > > > it > > > > > is the first message sent, it yields in the following structure: > > > > >=20 > > > > > Ring Buffer Descriptor Table Bulk Data Pages > > > > >=20 > > > > > +-+ +-+ +-----------------+ > > > > > =20 > > > > > |D|------------->|d|---------->| Bulk data block | > > > > > =20 > > > > > +-+ |d|--------+ +-----------------+ > > > > > =20 > > > > > | | |d|------+ | > > > > > =20 > > > > > +-+ . | | +-----------------+ > > > > > =20 > > > > > | | . | +->| Bulk data block | > > > > > =20 > > > > > . . | +-----------------+ > > > > > . |d|-+ | > > > > > . +-+ | | +-----------------+ > > > > > =20 > > > > > | | | +--->| Bulk data block | > > > > > =20 > > > > > +-+ | +-----------------+ > > > > > =20 > > > > > | | | . > > > > > =20 > > > > > +-+ | . > > > > > =20 > > > > > | . > > > > > | =20 > > > > > | +-----------------+ > > > > > =20 > > > > > +-------->| Bulk data block | > > > > > =20 > > > > > +-----------------+ > > > > >=20 > > > > > Legend: > > > > > D: pre-allocated descriptor > > > > > d: just in time allocated descriptor > > > > > -->: memory pointer (DMA) > > > > >=20 > > > > > The bulk data blocks are allocated by the respective device drive= r > > > > > above > > > > > virtio subsystem level (guest side). > > > > >=20 > > > > > There are exactly as many descriptors pre-allocated (D) as the si= ze of > > > > > a > > > > > ring buffer. > > > > >=20 > > > > > A "descriptor" is more or less just a chainable DMA memory pointe= r; > > > > > defined > > > > > as: > > > > >=20 > > > > > /* Virtio ring descriptors: 16 bytes. These can chain together v= ia > > > > > "next". */ struct vring_desc { > > > > >=20 > > > > > =09/* Address (guest-physical). */ > > > > > =09__virtio64 addr; > > > > > =09/* Length. */ > > > > > =09__virtio32 len; > > > > > =09/* The flags as indicated above. */ > > > > > =09__virtio16 flags; > > > > > =09/* We chain unused descriptors via this, too */ > > > > > =09__virtio16 next; > > > > >=20 > > > > > }; > > > > >=20 > > > > > There are 2 ring buffers; the "available" ring buffer is for send= ing a > > > > > message guest->host (which will transmit DMA addresses of guest > > > > > allocated > > > > > bulk data blocks that are used for data sent to device, and separ= ate > > > > > guest allocated bulk data blocks that will be used by host side t= o > > > > > place > > > > > its response bulk data), and the "used" ring buffer is for sendin= g > > > > > host->guest to let guest know about host's response and that it c= ould > > > > > now > > > > > safely consume and then deallocate the bulk data blocks subsequen= tly. > > > > >=20 > > > > > ---------- [recap-end] ---------- > > > > >=20 > > > > > So the "queue size" actually defines the ringbuffer size. It does= not > > > > > define the maximum amount of descriptors. The "queue size" rather > > > > > defines > > > > > how many pending messages can be pushed into either one ringbuffe= r > > > > > before > > > > > the other side would need to wait until the counter side would st= ep up > > > > > (i.e. ring buffer full). > > > > >=20 > > > > > The maximum amount of descriptors (what VIRTQUEUE_MAX_SIZE actual= ly > > > > > is) > > > > > OTOH defines the max. bulk data size that could be transmitted wi= th > > > > > each > > > > > virtio round trip message. > > > > >=20 > > > > > And in fact, 9p currently handles the virtio "queue size" as dire= ctly > > > > > associative with its maximum amount of active 9p requests the ser= ver > > > > > could > > > > >=20 > > > > > handle simultaniously: > > > > > hw/9pfs/9p.h:#define MAX_REQ 128 > > > > > hw/9pfs/9p.h: V9fsPDU pdus[MAX_REQ]; > > > > > hw/9pfs/virtio-9p-device.c: v->vq =3D virtio_add_queue(vdev, > > > > > MAX_REQ, > > > > > =20 > > > > > handle_9p_output); > > > > >=20 > > > > > So if I would change it like this, just for the purpose to increa= se > > > > > the > > > > > max. virtio transmission size: > > > > >=20 > > > > > --- a/hw/9pfs/virtio-9p-device.c > > > > > +++ b/hw/9pfs/virtio-9p-device.c > > > > > @@ -218,7 +218,7 @@ static void virtio_9p_device_realize(DeviceSt= ate > > > > > *dev, > > > > > Error **errp)> > > > > >=20 > > > > > v->config_size =3D sizeof(struct virtio_9p_config) + > > > > > strlen(s->fsconf.tag); > > > > > virtio_init(vdev, "virtio-9p", VIRTIO_ID_9P, v->config_size, > > > > > =20 > > > > > VIRTQUEUE_MAX_SIZE); > > > > >=20 > > > > > - v->vq =3D virtio_add_queue(vdev, MAX_REQ, handle_9p_output); > > > > > + v->vq =3D virtio_add_queue(vdev, 32*1024, handle_9p_output); > > > > >=20 > > > > > } > > > > >=20 > > > > > Then it would require additional synchronization code on both end= s and > > > > > therefore unnecessary complexity, because it would now be possibl= e > > > > > that > > > > > more requests are pushed into the ringbuffer than server could ha= ndle. > > > > >=20 > > > > > There is one potential issue though that probably did justify the > > > > > "don't > > > > > exceed the queue size" rule: > > > > >=20 > > > > > ATM the descriptor table is allocated (just in time) as *one* > > > > > continuous > > > > > buffer via kmalloc_array(): > > > > > https://github.com/torvalds/linux/blob/2f111a6fd5b5297b4e92f53798= ca086 > > > > > f7c7 > > > > > d33a4/drivers/virtio/virtio_ring.c#L440 > > > > >=20 > > > > > So assuming transmission size of 2 * 128 MB that kmalloc_array() = call > > > > > would > > > > > yield in kmalloc(1M) and the latter might fail if guest had highl= y > > > > > fragmented physical memory. For such kind of error case there is > > > > > currently a fallback path in virtqueue_add_split() that would the= n use > > > > > the required amount of pre-allocated descriptors instead: > > > > > https://github.com/torvalds/linux/blob/2f111a6fd5b5297b4e92f53798= ca086 > > > > > f7c7 > > > > > d33a4/drivers/virtio/virtio_ring.c#L525 > > > > >=20 > > > > > That fallback recovery path would no longer be viable if the queu= e > > > > > size > > > > > was > > > > > exceeded. There would be alternatives though, e.g. by allowing to > > > > > chain > > > > > indirect descriptor tables (currently prohibited by the virtio sp= ecs). > > > >=20 > > > > Making the maximum number of descriptors independent of the queue s= ize > > > > requires a change to the VIRTIO spec since the two values are curre= ntly > > > > explicitly tied together by the spec. > > >=20 > > > Yes, that's what the virtio specs say. But they don't say why, nor di= d I > > > hear a reason in this dicussion. > > >=20 > > > That's why I invested time reviewing current virtio implementation an= d > > > specs, as well as actually testing exceeding that limit. And as I > > > outlined in detail in my previous email, I only found one theoretical > > > issue that could be addressed though. > >=20 > > I agree that there is a limitation in the VIRTIO spec, but violating th= e > > spec isn't an acceptable solution: > >=20 > > 1. QEMU and Linux aren't the only components that implement VIRTIO. You > > cannot make assumptions about their implementations because it may > > break spec-compliant implementations that you haven't looked at. > >=20 > > Your patches weren't able to increase Queue Size because some device > > implementations break when descriptor chains are too long. This show= s > > there is a practical issue even in QEMU. > >=20 > > 2. The specific spec violation that we discussed creates the problem > > that drivers can no longer determine the maximum description chain > > length. This in turn will lead to more implementation-specific > > assumptions being baked into drivers and cause problems with > > interoperability and future changes. > >=20 > > The spec needs to be extended instead. I included an idea for how to do > > that below. >=20 > Sure, I just wanted to see if there was a non-neglectable "hard" show sto= pper > per se that I probably haven't seen yet. I have not questioned aiming a c= lean > solution. >=20 > Thanks for the clarification! >=20 > > > > Before doing that, are there benchmark results showing that 1 MB vs= 128 > > > > MB produces a performance improvement? I'm asking because if perfor= mance > > > > with 1 MB is good then you can probably do that without having to c= hange > > > > VIRTIO and also because it's counter-intuitive that 9p needs 128 MB= for > > > > good performance when it's ultimately implemented on top of disk an= d > > > > network I/O that have lower size limits. > > >=20 > > > First some numbers, linear reading a 12 GB file: > > >=20 > > > msize average notes > > >=20 > > > 8 kB 52.0 MB/s default msize of Linux kernel > > 128 kB 624.8 MB/s default msize of Linux kernel >=3Dv5.15 > > > 512 kB 1961 MB/s current max. msize with any Linux kernel <=3Dv5= .15 > > > 1 MB 2551 MB/s this msize would already violate virtio specs > > > 2 MB 2521 MB/s this msize would already violate virtio specs > > > 4 MB 2628 MB/s planned max. msize of my current kernel patches= [1] > >=20 > > How many descriptors are used? 4 MB can be covered by a single > > descriptor if the data is physically contiguous in memory, so this data > > doesn't demonstrate a need for more descriptors. >=20 > No, in the last couple years there was apparently no kernel version that = used > just one descriptor, nor did my benchmarked version. Even though the Linu= x 9p > client uses (yet) simple linear buffers (contiguous physical memory) on 9= p > client level, these are however split into PAGE_SIZE chunks by function > pack_sg_list() [1] before being fed to virtio level: >=20 > static unsigned int rest_of_page(void *data) > { > =09return PAGE_SIZE - offset_in_page(data); > } > ... > static int pack_sg_list(struct scatterlist *sg, int start, > =09=09=09int limit, char *data, int count) > { > =09int s; > =09int index =3D start; >=20 > =09while (count) { > =09=09s =3D rest_of_page(data); > =09=09... > =09=09sg_set_buf(&sg[index++], data, s); > =09=09count -=3D s; > =09=09data +=3D s; > =09} > =09... > } >=20 > [1] https://github.com/torvalds/linux/blob/19901165d90fdca1e57c9baa0d5b4c= 63d15c476a/net/9p/trans_virtio.c#L171 >=20 > So when sending 4MB over virtio wire, it would yield in 1k descriptors AT= M. >=20 > I have wondered about this before, but did not question it, because due t= o the > cross-platform nature I couldn't say for certain whether that's probably > needed somewhere. I mean for the case virtio-PCI I know for sure that one > descriptor (i.e. >PAGE_SIZE) would be fine, but I don't know if that appl= ies > to all buses and architectures. VIRTIO does not limit descriptor the descriptor len field to PAGE_SIZE, so I don't think there is a limit at the VIRTIO level. If this function coalesces adjacent pages then the descriptor chain length issues could be reduced. > > > But again, this is not just about performance. My conclusion as descr= ibed > > > in my previous email is that virtio currently squeezes > > >=20 > > > =09"max. simultanious amount of bulk messages" > > >=20 > > > vs. > > >=20 > > > =09"max. bulk data transmission size per bulk messaage" > > >=20 > > > into the same configuration parameter, which is IMO inappropriate and > > > hence > > > splitting them into 2 separate parameters when creating a queue makes > > > sense, independent of the performance benchmarks. > > >=20 > > > [1] > > > https://lore.kernel.org/netdev/cover.1632327421.git.linux_oss@crudeby= te.c > > > om/ > > Some devices effectively already have this because the device advertise= s > > a maximum number of descriptors via device-specific mechanisms like the > > struct virtio_blk_config seg_max field. But today these fields can only > > reduce the maximum descriptor chain length because the spec still limit= s > > the length to Queue Size. > >=20 > > We can build on this approach to raise the length above Queue Size. Thi= s > > approach has the advantage that the maximum number of segments isn't pe= r > > device or per virtqueue, it's fine-grained. If the device supports two > > requests types then different max descriptor chain limits could be give= n > > for them by introducing two separate configuration space fields. > >=20 > > Here are the corresponding spec changes: > >=20 > > 1. A new feature bit called VIRTIO_RING_F_LARGE_INDIRECT_DESC is added > > to indicate that indirect descriptor table size and maximum > > descriptor chain length are not limited by Queue Size value. (Maybe > > there still needs to be a limit like 2^15?) >=20 > Sounds good to me! >=20 > AFAIK it is effectively limited to 2^16 because of vring_desc->next: >=20 > /* Virtio ring descriptors: 16 bytes. These can chain together via "next= ". */ > struct vring_desc { > /* Address (guest-physical). */ > __virtio64 addr; > /* Length. */ > __virtio32 len; > /* The flags as indicated above. */ > __virtio16 flags; > /* We chain unused descriptors via this, too */ > __virtio16 next; > }; Yes, Split Virtqueues have a fundamental limit on indirect table size due to the "next" field. Packed Virtqueue descriptors don't have a "next" field so descriptor chains could be longer in theory (currently forbidden by the spec). > > One thing that's messy is that we've been discussing the maximum > > descriptor chain length but 9p has the "msize" concept, which isn't > > aware of contiguous memory. It may be necessary to extend the 9p driver > > code to size requests not just according to their length in bytes but > > also according to the descriptor chain length. That's how the Linux > > block layer deals with queue limits (struct queue_limits max_segments v= s > > max_hw_sectors). >=20 > Hmm, can't follow on that one. For what should that be needed in case of = 9p? > My plan was to limit msize by 9p client simply at session start to whatev= er is > the max. amount virtio descriptors supported by host and using PAGE_SIZE = as > size per descriptor, because that's what 9p client actually does ATM (see > above). So you think that should be changed to e.g. just one descriptor f= or > 4MB, right? Limiting msize to the 9p transport device's maximum number of descriptors is conservative (i.e. 128 descriptors =3D 512 KB msize) because it doesn't take advantage of contiguous memory. I suggest leaving msize alone, adding a separate limit at which requests are split according to the maximum descriptor chain length, and tweaking pack_sg_list() to coalesce adjacent pages. That way msize can be large without necessarily using lots of descriptors (depending on the memory layout). Stefan --M4Sxhk+xVfddXjVp Content-Type: application/pgp-signature; name="signature.asc" -----BEGIN PGP SIGNATURE----- iQEzBAEBCAAdFiEEhpWov9P5fNqsNXdanKSrs4Grc8gFAmGCc40ACgkQnKSrs4Gr c8hYhQgAsDN1v7KcvgQyfEen8CI6V3qJ+z5io48nM2Nh7SN0uAVdWQmptAX2/Q8Z tiipQeIsEgIg9UbU42XttA6YVXzB/ogon7c0j/lv1YE53G9Ky1KolAIngPf0fBj8 yAtUG0ON7ZLnaecVYEfnZxbjQZ4xuw87nFGGynRty8064dUXhtEpJYy8zSd3pBIE 8UXU/Ojs1gaihtC2I2XhqXrLHce8vDE3S09xO0kQlzHGuDUhtjE46h9iqDEcAx8+ VotagROfPYpYQ/pElzzmY4r1gmJ0uZArJHxuX2G56icvQSj9b8mNDg9gJ6/DcFZZ wgDE4rfmcrn5Qlc1JtuDVctmECcV6Q== =HaoU -----END PGP SIGNATURE----- --M4Sxhk+xVfddXjVp-- From mboxrd@z Thu Jan 1 00:00:00 1970 Date: Wed, 3 Nov 2021 11:33:33 +0000 From: Stefan Hajnoczi Message-ID: References: <10221570.6MffRmy8Bz@silver> <136317874.cAzoFBZ3UR@silver> MIME-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha256; protocol="application/pgp-signature"; boundary="M4Sxhk+xVfddXjVp" Content-Disposition: inline In-Reply-To: <136317874.cAzoFBZ3UR@silver> Subject: Re: [Virtio-fs] [PATCH v2 0/3] virtio: increase VIRTQUEUE_MAX_SIZE to 32k List-Id: Development discussions about virtio-fs List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Christian Schoenebeck Cc: Kevin Wolf , Laurent Vivier , qemu-block@nongnu.org, "Michael S. Tsirkin" , Jason Wang , Amit Shah , David Hildenbrand , qemu-devel@nongnu.org, virtio-fs@redhat.com, Eric Auger , Hanna Reitz , "Gonglei (Arei)" , Gerd Hoffmann , =?iso-8859-1?Q?Marc-Andr=E9?= Lureau , Paolo Bonzini , Fam Zheng , Raphael Norwitz --M4Sxhk+xVfddXjVp Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Mon, Nov 01, 2021 at 09:29:26PM +0100, Christian Schoenebeck wrote: > On Donnerstag, 28. Oktober 2021 11:00:48 CET Stefan Hajnoczi wrote: > > On Mon, Oct 25, 2021 at 05:03:25PM +0200, Christian Schoenebeck wrote: > > > On Montag, 25. Oktober 2021 12:30:41 CEST Stefan Hajnoczi wrote: > > > > On Thu, Oct 21, 2021 at 05:39:28PM +0200, Christian Schoenebeck wro= te: > > > > > On Freitag, 8. Oktober 2021 18:08:48 CEST Christian Schoenebeck w= rote: > > > > > > On Freitag, 8. Oktober 2021 16:24:42 CEST Christian Schoenebeck= wrote: > > > > > > > On Freitag, 8. Oktober 2021 09:25:33 CEST Greg Kurz wrote: > > > > > > > > On Thu, 7 Oct 2021 16:42:49 +0100 > > > > > > > >=20 > > > > > > > > Stefan Hajnoczi wrote: > > > > > > > > > On Thu, Oct 07, 2021 at 02:51:55PM +0200, Christian Schoe= nebeck wrote: > > > > > > > > > > On Donnerstag, 7. Oktober 2021 07:23:59 CEST Stefan Haj= noczi wrote: > > > > > > > > > > > On Mon, Oct 04, 2021 at 09:38:00PM +0200, Christian > > > > > > > > > > > Schoenebeck > > > > > >=20 > > > > > > wrote: > > > > > > > > > > > > At the moment the maximum transfer size with virtio= is > > > > > > > > > > > > limited > > > > > > > > > > > > to > > > > > > > > > > > > 4M > > > > > > > > > > > > (1024 * PAGE_SIZE). This series raises this limit t= o its > > > > > > > > > > > > maximum > > > > > > > > > > > > theoretical possible transfer size of 128M (32k pag= es) > > > > > > > > > > > > according > > > > > > > > > > > > to > > > > > > > > > > > > the > > > > > > > > > > > > virtio specs: > > > > > > > > > > > >=20 > > > > > > > > > > > > https://docs.oasis-open.org/virtio/virtio/v1.1/cs01= /virt > > > > > > > > > > > > io-v > > > > > > > > > > > > 1.1- > > > > > > > > > > > > cs > > > > > > > > > > > > 01 > > > > > > > > > > > > .html# > > > > > > > > > > > > x1-240006 > > > > > > > > > > >=20 > > > > > > > > > > > Hi Christian, > > > > > > > >=20 > > > > > > > > > > > I took a quick look at the code: > > > > > > > > Hi, > > > > > > > >=20 > > > > > > > > Thanks Stefan for sharing virtio expertise and helping Chri= stian > > > > > > > > ! > > > > > > > >=20 > > > > > > > > > > > - The Linux 9p driver restricts descriptor chains to = 128 > > > > > > > > > > > elements > > > > > > > > > > >=20 > > > > > > > > > > > (net/9p/trans_virtio.c:VIRTQUEUE_NUM) > > > > > > > > > >=20 > > > > > > > > > > Yes, that's the limitation that I am about to remove (W= IP); > > > > > > > > > > current > > > > > > > > > > kernel > > > > > > > > > > patches: > > > > > > > > > > https://lore.kernel.org/netdev/cover.1632327421.git.lin= ux_os > > > > > > > > > > s@cr > > > > > > > > > > udeb > > > > > > > > > > yt > > > > > > > > > > e. > > > > > > > > > > com/> > > > > > > > > >=20 > > > > > > > > > I haven't read the patches yet but I'm concerned that tod= ay > > > > > > > > > the > > > > > > > > > driver > > > > > > > > > is pretty well-behaved and this new patch series introduc= es a > > > > > > > > > spec > > > > > > > > > violation. Not fixing existing spec violations is okay, b= ut > > > > > > > > > adding > > > > > > > > > new > > > > > > > > > ones is a red flag. I think we need to figure out a clean > > > > > > > > > solution. > > > > > > >=20 > > > > > > > Nobody has reviewed the kernel patches yet. My main concern > > > > > > > therefore > > > > > > > actually is that the kernel patches are already too complex, > > > > > > > because > > > > > > > the > > > > > > > current situation is that only Dominique is handling 9p patch= es on > > > > > > > kernel > > > > > > > side, and he barely has time for 9p anymore. > > > > > > >=20 > > > > > > > Another reason for me to catch up on reading current kernel c= ode > > > > > > > and > > > > > > > stepping in as reviewer of 9p on kernel side ASAP, independen= t of > > > > > > > this > > > > > > > issue. > > > > > > >=20 > > > > > > > As for current kernel patches' complexity: I can certainly dr= op > > > > > > > patch > > > > > > > 7 > > > > > > > entirely as it is probably just overkill. Patch 4 is then the > > > > > > > biggest > > > > > > > chunk, I have to see if I can simplify it, and whether it wou= ld > > > > > > > make > > > > > > > sense to squash with patch 3. > > > > > > >=20 > > > > > > > > > > > - The QEMU 9pfs code passes iovecs directly to preadv= (2) > > > > > > > > > > > and > > > > > > > > > > > will > > > > > > > > > > > fail > > > > > > > > > > >=20 > > > > > > > > > > > with EINVAL when called with more than IOV_MAX iove= cs > > > > > > > > > > > (hw/9pfs/9p.c:v9fs_read()) > > > > > > > > > >=20 > > > > > > > > > > Hmm, which makes me wonder why I never encountered this > > > > > > > > > > error > > > > > > > > > > during > > > > > > > > > > testing. > > > > > > > > > >=20 > > > > > > > > > > Most people will use the 9p qemu 'local' fs driver back= end > > > > > > > > > > in > > > > > > > > > > practice, > > > > > > > > > > so > > > > > > > > > > that v9fs_read() call would translate for most people to > > > > > > > > > > this > > > > > > > > > > implementation on QEMU side (hw/9p/9p-local.c): > > > > > > > > > >=20 > > > > > > > > > > static ssize_t local_preadv(FsContext *ctx, V9fsFidOpen= State > > > > > > > > > > *fs, > > > > > > > > > >=20 > > > > > > > > > > const struct iovec *iov, > > > > > > > > > > int iovcnt, off_t offset) > > > > > > > > > >=20 > > > > > > > > > > { > > > > > > > > > > #ifdef CONFIG_PREADV > > > > > > > > > >=20 > > > > > > > > > > return preadv(fs->fd, iov, iovcnt, offset); > > > > > > > > > >=20 > > > > > > > > > > #else > > > > > > > > > >=20 > > > > > > > > > > int err =3D lseek(fs->fd, offset, SEEK_SET); > > > > > > > > > > if (err =3D=3D -1) { > > > > > > > > > > =20 > > > > > > > > > > return err; > > > > > > > > > > =20 > > > > > > > > > > } else { > > > > > > > > > > =20 > > > > > > > > > > return readv(fs->fd, iov, iovcnt); > > > > > > > > > > =20 > > > > > > > > > > } > > > > > > > > > >=20 > > > > > > > > > > #endif > > > > > > > > > > } > > > > > > > > > >=20 > > > > > > > > > > > Unless I misunderstood the code, neither side can take > > > > > > > > > > > advantage > > > > > > > > > > > of > > > > > > > > > > > the > > > > > > > > > > > new 32k descriptor chain limit? > > > > > > > > > > >=20 > > > > > > > > > > > Thanks, > > > > > > > > > > > Stefan > > > > > > > > > >=20 > > > > > > > > > > I need to check that when I have some more time. One > > > > > > > > > > possible > > > > > > > > > > explanation > > > > > > > > > > might be that preadv() already has this wrapped into a = loop > > > > > > > > > > in > > > > > > > > > > its > > > > > > > > > > implementation to circumvent a limit like IOV_MAX. It m= ight > > > > > > > > > > be > > > > > > > > > > another > > > > > > > > > > "it > > > > > > > > > > works, but not portable" issue, but not sure. > > > > > > > > > >=20 > > > > > > > > > > There are still a bunch of other issues I have to resol= ve. > > > > > > > > > > If > > > > > > > > > > you > > > > > > > > > > look > > > > > > > > > > at > > > > > > > > > > net/9p/client.c on kernel side, you'll notice that it > > > > > > > > > > basically > > > > > > > > > > does > > > > > > > > > > this ATM> > > > > > > > > > > >=20 > > > > > > > > > > kmalloc(msize); > > > > > > > >=20 > > > > > > > > Note that this is done twice : once for the T message (clie= nt > > > > > > > > request) > > > > > > > > and > > > > > > > > once for the R message (server answer). The 9p driver could > > > > > > > > adjust > > > > > > > > the > > > > > > > > size > > > > > > > > of the T message to what's really needed instead of allocat= ing > > > > > > > > the > > > > > > > > full > > > > > > > > msize. R message size is not known though. > > > > > > >=20 > > > > > > > Would it make sense adding a second virtio ring, dedicated to > > > > > > > server > > > > > > > responses to solve this? IIRC 9p server already calculates > > > > > > > appropriate > > > > > > > exact sizes for each response type. So server could just push > > > > > > > space > > > > > > > that's > > > > > > > really needed for its responses. > > > > > > >=20 > > > > > > > > > > for every 9p request. So not only does it allocate much= more > > > > > > > > > > memory > > > > > > > > > > for > > > > > > > > > > every request than actually required (i.e. say 9pfs was > > > > > > > > > > mounted > > > > > > > > > > with > > > > > > > > > > msize=3D8M, then a 9p request that actually would just = need 1k > > > > > > > > > > would > > > > > > > > > > nevertheless allocate 8M), but also it allocates > > > > > > > > > > > PAGE_SIZE, > > > > > > > > > > which > > > > > > > > > > obviously may fail at any time.> > > > > > > > > >=20 > > > > > > > > > The PAGE_SIZE limitation sounds like a kmalloc() vs vmall= oc() > > > > > > > > > situation. > > > > > > >=20 > > > > > > > Hu, I didn't even consider vmalloc(). I just tried the kvmall= oc() > > > > > > > wrapper > > > > > > > as a quick & dirty test, but it crashed in the same way as > > > > > > > kmalloc() > > > > > > > with > > > > > > > large msize values immediately on mounting: > > > > > > >=20 > > > > > > > diff --git a/net/9p/client.c b/net/9p/client.c > > > > > > > index a75034fa249b..cfe300a4b6ca 100644 > > > > > > > --- a/net/9p/client.c > > > > > > > +++ b/net/9p/client.c > > > > > > > @@ -227,15 +227,18 @@ static int parse_opts(char *opts, struct > > > > > > > p9_client > > > > > > > *clnt) > > > > > > >=20 > > > > > > > static int p9_fcall_init(struct p9_client *c, struct p9_fcall > > > > > > > *fc, > > > > > > > =20 > > > > > > > int alloc_msize) > > > > > > > =20 > > > > > > > { > > > > > > >=20 > > > > > > > - if (likely(c->fcall_cache) && alloc_msize =3D=3D c->m= size) { > > > > > > > + //if (likely(c->fcall_cache) && alloc_msize =3D=3D c-= >msize) { > > > > > > > + if (false) { > > > > > > >=20 > > > > > > > fc->sdata =3D kmem_cache_alloc(c->fcall_cache, > > > > > > > GFP_NOFS); > > > > > > > fc->cache =3D c->fcall_cache; > > > > > > > =20 > > > > > > > } else { > > > > > > >=20 > > > > > > > - fc->sdata =3D kmalloc(alloc_msize, GFP_NOFS); > > > > > > > + fc->sdata =3D kvmalloc(alloc_msize, GFP_NOFS); > > > > > >=20 > > > > > > Ok, GFP_NOFS -> GFP_KERNEL did the trick. > > > > > >=20 > > > > > > Now I get: > > > > > > virtio: bogus descriptor or out of resources > > > > > >=20 > > > > > > So, still some work ahead on both ends. > > > > >=20 > > > > > Few hacks later (only changes on 9p client side) I got this runni= ng > > > > > stable > > > > > now. The reason for the virtio error above was that kvmalloc() re= turns > > > > > a > > > > > non-logical kernel address for any kvmalloc(>4M), i.e. an address= that > > > > > is > > > > > inaccessible from host side, hence that "bogus descriptor" messag= e by > > > > > QEMU. > > > > > So I had to split those linear 9p client buffers into sparse ones= (set > > > > > of > > > > > individual pages). > > > > >=20 > > > > > I tested this for some days with various virtio transmission size= s and > > > > > it > > > > > works as expected up to 128 MB (more precisely: 128 MB read space= + > > > > > 128 MB > > > > > write space per virtio round trip message). > > > > >=20 > > > > > I did not encounter a show stopper for large virtio transmission = sizes > > > > > (4 MB ... 128 MB) on virtio level, neither as a result of testing= , nor > > > > > after reviewing the existing code. > > > > >=20 > > > > > About IOV_MAX: that's apparently not an issue on virtio level. Mo= st of > > > > > the > > > > > iovec code, both on Linux kernel side and on QEMU side do not have > > > > > this > > > > > limitation. It is apparently however indeed a limitation for user= land > > > > > apps > > > > > calling the Linux kernel's syscalls yet. > > > > >=20 > > > > > Stefan, as it stands now, I am even more convinced that the upper > > > > > virtio > > > > > transmission size limit should not be squeezed into the queue size > > > > > argument of virtio_add_queue(). Not because of the previous argum= ent > > > > > that > > > > > it would waste space (~1MB), but rather because they are two diff= erent > > > > > things. To outline this, just a quick recap of what happens exact= ly > > > > > when > > > > > a bulk message is pushed over the virtio wire (assuming virtio "s= plit" > > > > > layout here): > > > > >=20 > > > > > ---------- [recap-start] ---------- > > > > >=20 > > > > > For each bulk message sent guest <-> host, exactly *one* of the > > > > > pre-allocated descriptors is taken and placed (subsequently) into > > > > > exactly > > > > > *one* position of the two available/used ring buffers. The actual > > > > > descriptor table though, containing all the DMA addresses of the > > > > > message > > > > > bulk data, is allocated just in time for each round trip message.= Say, > > > > > it > > > > > is the first message sent, it yields in the following structure: > > > > >=20 > > > > > Ring Buffer Descriptor Table Bulk Data Pages > > > > >=20 > > > > > +-+ +-+ +-----------------+ > > > > > =20 > > > > > |D|------------->|d|---------->| Bulk data block | > > > > > =20 > > > > > +-+ |d|--------+ +-----------------+ > > > > > =20 > > > > > | | |d|------+ | > > > > > =20 > > > > > +-+ . | | +-----------------+ > > > > > =20 > > > > > | | . | +->| Bulk data block | > > > > > =20 > > > > > . . | +-----------------+ > > > > > . |d|-+ | > > > > > . +-+ | | +-----------------+ > > > > > =20 > > > > > | | | +--->| Bulk data block | > > > > > =20 > > > > > +-+ | +-----------------+ > > > > > =20 > > > > > | | | . > > > > > =20 > > > > > +-+ | . > > > > > =20 > > > > > | . > > > > > | =20 > > > > > | +-----------------+ > > > > > =20 > > > > > +-------->| Bulk data block | > > > > > =20 > > > > > +-----------------+ > > > > >=20 > > > > > Legend: > > > > > D: pre-allocated descriptor > > > > > d: just in time allocated descriptor > > > > > -->: memory pointer (DMA) > > > > >=20 > > > > > The bulk data blocks are allocated by the respective device driver > > > > > above > > > > > virtio subsystem level (guest side). > > > > >=20 > > > > > There are exactly as many descriptors pre-allocated (D) as the si= ze of > > > > > a > > > > > ring buffer. > > > > >=20 > > > > > A "descriptor" is more or less just a chainable DMA memory pointe= r; > > > > > defined > > > > > as: > > > > >=20 > > > > > /* Virtio ring descriptors: 16 bytes. These can chain together v= ia > > > > > "next". */ struct vring_desc { > > > > >=20 > > > > > /* Address (guest-physical). */ > > > > > __virtio64 addr; > > > > > /* Length. */ > > > > > __virtio32 len; > > > > > /* The flags as indicated above. */ > > > > > __virtio16 flags; > > > > > /* We chain unused descriptors via this, too */ > > > > > __virtio16 next; > > > > >=20 > > > > > }; > > > > >=20 > > > > > There are 2 ring buffers; the "available" ring buffer is for send= ing a > > > > > message guest->host (which will transmit DMA addresses of guest > > > > > allocated > > > > > bulk data blocks that are used for data sent to device, and separ= ate > > > > > guest allocated bulk data blocks that will be used by host side to > > > > > place > > > > > its response bulk data), and the "used" ring buffer is for sending > > > > > host->guest to let guest know about host's response and that it c= ould > > > > > now > > > > > safely consume and then deallocate the bulk data blocks subsequen= tly. > > > > >=20 > > > > > ---------- [recap-end] ---------- > > > > >=20 > > > > > So the "queue size" actually defines the ringbuffer size. It does= not > > > > > define the maximum amount of descriptors. The "queue size" rather > > > > > defines > > > > > how many pending messages can be pushed into either one ringbuffer > > > > > before > > > > > the other side would need to wait until the counter side would st= ep up > > > > > (i.e. ring buffer full). > > > > >=20 > > > > > The maximum amount of descriptors (what VIRTQUEUE_MAX_SIZE actual= ly > > > > > is) > > > > > OTOH defines the max. bulk data size that could be transmitted wi= th > > > > > each > > > > > virtio round trip message. > > > > >=20 > > > > > And in fact, 9p currently handles the virtio "queue size" as dire= ctly > > > > > associative with its maximum amount of active 9p requests the ser= ver > > > > > could > > > > >=20 > > > > > handle simultaniously: > > > > > hw/9pfs/9p.h:#define MAX_REQ 128 > > > > > hw/9pfs/9p.h: V9fsPDU pdus[MAX_REQ]; > > > > > hw/9pfs/virtio-9p-device.c: v->vq =3D virtio_add_queue(vdev, > > > > > MAX_REQ, > > > > > =20 > > > > > handle_9p_output); > > > > >=20 > > > > > So if I would change it like this, just for the purpose to increa= se > > > > > the > > > > > max. virtio transmission size: > > > > >=20 > > > > > --- a/hw/9pfs/virtio-9p-device.c > > > > > +++ b/hw/9pfs/virtio-9p-device.c > > > > > @@ -218,7 +218,7 @@ static void virtio_9p_device_realize(DeviceSt= ate > > > > > *dev, > > > > > Error **errp)> > > > > >=20 > > > > > v->config_size =3D sizeof(struct virtio_9p_config) + > > > > > strlen(s->fsconf.tag); > > > > > virtio_init(vdev, "virtio-9p", VIRTIO_ID_9P, v->config_size, > > > > > =20 > > > > > VIRTQUEUE_MAX_SIZE); > > > > >=20 > > > > > - v->vq =3D virtio_add_queue(vdev, MAX_REQ, handle_9p_output); > > > > > + v->vq =3D virtio_add_queue(vdev, 32*1024, handle_9p_output); > > > > >=20 > > > > > } > > > > >=20 > > > > > Then it would require additional synchronization code on both end= s and > > > > > therefore unnecessary complexity, because it would now be possible > > > > > that > > > > > more requests are pushed into the ringbuffer than server could ha= ndle. > > > > >=20 > > > > > There is one potential issue though that probably did justify the > > > > > "don't > > > > > exceed the queue size" rule: > > > > >=20 > > > > > ATM the descriptor table is allocated (just in time) as *one* > > > > > continuous > > > > > buffer via kmalloc_array(): > > > > > https://github.com/torvalds/linux/blob/2f111a6fd5b5297b4e92f53798= ca086 > > > > > f7c7 > > > > > d33a4/drivers/virtio/virtio_ring.c#L440 > > > > >=20 > > > > > So assuming transmission size of 2 * 128 MB that kmalloc_array() = call > > > > > would > > > > > yield in kmalloc(1M) and the latter might fail if guest had highly > > > > > fragmented physical memory. For such kind of error case there is > > > > > currently a fallback path in virtqueue_add_split() that would the= n use > > > > > the required amount of pre-allocated descriptors instead: > > > > > https://github.com/torvalds/linux/blob/2f111a6fd5b5297b4e92f53798= ca086 > > > > > f7c7 > > > > > d33a4/drivers/virtio/virtio_ring.c#L525 > > > > >=20 > > > > > That fallback recovery path would no longer be viable if the queue > > > > > size > > > > > was > > > > > exceeded. There would be alternatives though, e.g. by allowing to > > > > > chain > > > > > indirect descriptor tables (currently prohibited by the virtio sp= ecs). > > > >=20 > > > > Making the maximum number of descriptors independent of the queue s= ize > > > > requires a change to the VIRTIO spec since the two values are curre= ntly > > > > explicitly tied together by the spec. > > >=20 > > > Yes, that's what the virtio specs say. But they don't say why, nor di= d I > > > hear a reason in this dicussion. > > >=20 > > > That's why I invested time reviewing current virtio implementation and > > > specs, as well as actually testing exceeding that limit. And as I > > > outlined in detail in my previous email, I only found one theoretical > > > issue that could be addressed though. > >=20 > > I agree that there is a limitation in the VIRTIO spec, but violating the > > spec isn't an acceptable solution: > >=20 > > 1. QEMU and Linux aren't the only components that implement VIRTIO. You > > cannot make assumptions about their implementations because it may > > break spec-compliant implementations that you haven't looked at. > >=20 > > Your patches weren't able to increase Queue Size because some device > > implementations break when descriptor chains are too long. This shows > > there is a practical issue even in QEMU. > >=20 > > 2. The specific spec violation that we discussed creates the problem > > that drivers can no longer determine the maximum description chain > > length. This in turn will lead to more implementation-specific > > assumptions being baked into drivers and cause problems with > > interoperability and future changes. > >=20 > > The spec needs to be extended instead. I included an idea for how to do > > that below. >=20 > Sure, I just wanted to see if there was a non-neglectable "hard" show sto= pper > per se that I probably haven't seen yet. I have not questioned aiming a c= lean > solution. >=20 > Thanks for the clarification! >=20 > > > > Before doing that, are there benchmark results showing that 1 MB vs= 128 > > > > MB produces a performance improvement? I'm asking because if perfor= mance > > > > with 1 MB is good then you can probably do that without having to c= hange > > > > VIRTIO and also because it's counter-intuitive that 9p needs 128 MB= for > > > > good performance when it's ultimately implemented on top of disk and > > > > network I/O that have lower size limits. > > >=20 > > > First some numbers, linear reading a 12 GB file: > > >=20 > > > msize average notes > > >=20 > > > 8 kB 52.0 MB/s default msize of Linux kernel > > 128 kB 624.8 MB/s default msize of Linux kernel >=3Dv5.15 > > > 512 kB 1961 MB/s current max. msize with any Linux kernel <=3Dv5= =2E15 > > > 1 MB 2551 MB/s this msize would already violate virtio specs > > > 2 MB 2521 MB/s this msize would already violate virtio specs > > > 4 MB 2628 MB/s planned max. msize of my current kernel patches= [1] > >=20 > > How many descriptors are used? 4 MB can be covered by a single > > descriptor if the data is physically contiguous in memory, so this data > > doesn't demonstrate a need for more descriptors. >=20 > No, in the last couple years there was apparently no kernel version that = used > just one descriptor, nor did my benchmarked version. Even though the Linu= x 9p > client uses (yet) simple linear buffers (contiguous physical memory) on 9p > client level, these are however split into PAGE_SIZE chunks by function > pack_sg_list() [1] before being fed to virtio level: >=20 > static unsigned int rest_of_page(void *data) > { > return PAGE_SIZE - offset_in_page(data); > } > ... > static int pack_sg_list(struct scatterlist *sg, int start, > int limit, char *data, int count) > { > int s; > int index =3D start; >=20 > while (count) { > s =3D rest_of_page(data); > ... > sg_set_buf(&sg[index++], data, s); > count -=3D s; > data +=3D s; > } > ... > } >=20 > [1] https://github.com/torvalds/linux/blob/19901165d90fdca1e57c9baa0d5b4c= 63d15c476a/net/9p/trans_virtio.c#L171 >=20 > So when sending 4MB over virtio wire, it would yield in 1k descriptors AT= M. >=20 > I have wondered about this before, but did not question it, because due t= o the > cross-platform nature I couldn't say for certain whether that's probably > needed somewhere. I mean for the case virtio-PCI I know for sure that one > descriptor (i.e. >PAGE_SIZE) would be fine, but I don't know if that appl= ies > to all buses and architectures. VIRTIO does not limit descriptor the descriptor len field to PAGE_SIZE, so I don't think there is a limit at the VIRTIO level. If this function coalesces adjacent pages then the descriptor chain length issues could be reduced. > > > But again, this is not just about performance. My conclusion as descr= ibed > > > in my previous email is that virtio currently squeezes > > >=20 > > > "max. simultanious amount of bulk messages" > > >=20 > > > vs. > > >=20 > > > "max. bulk data transmission size per bulk messaage" > > >=20 > > > into the same configuration parameter, which is IMO inappropriate and > > > hence > > > splitting them into 2 separate parameters when creating a queue makes > > > sense, independent of the performance benchmarks. > > >=20 > > > [1] > > > https://lore.kernel.org/netdev/cover.1632327421.git.linux_oss@crudeby= te.c > > > om/ > > Some devices effectively already have this because the device advertises > > a maximum number of descriptors via device-specific mechanisms like the > > struct virtio_blk_config seg_max field. But today these fields can only > > reduce the maximum descriptor chain length because the spec still limits > > the length to Queue Size. > >=20 > > We can build on this approach to raise the length above Queue Size. This > > approach has the advantage that the maximum number of segments isn't per > > device or per virtqueue, it's fine-grained. If the device supports two > > requests types then different max descriptor chain limits could be given > > for them by introducing two separate configuration space fields. > >=20 > > Here are the corresponding spec changes: > >=20 > > 1. A new feature bit called VIRTIO_RING_F_LARGE_INDIRECT_DESC is added > > to indicate that indirect descriptor table size and maximum > > descriptor chain length are not limited by Queue Size value. (Maybe > > there still needs to be a limit like 2^15?) >=20 > Sounds good to me! >=20 > AFAIK it is effectively limited to 2^16 because of vring_desc->next: >=20 > /* Virtio ring descriptors: 16 bytes. These can chain together via "next= ". */ > struct vring_desc { > /* Address (guest-physical). */ > __virtio64 addr; > /* Length. */ > __virtio32 len; > /* The flags as indicated above. */ > __virtio16 flags; > /* We chain unused descriptors via this, too */ > __virtio16 next; > }; Yes, Split Virtqueues have a fundamental limit on indirect table size due to the "next" field. Packed Virtqueue descriptors don't have a "next" field so descriptor chains could be longer in theory (currently forbidden by the spec). > > One thing that's messy is that we've been discussing the maximum > > descriptor chain length but 9p has the "msize" concept, which isn't > > aware of contiguous memory. It may be necessary to extend the 9p driver > > code to size requests not just according to their length in bytes but > > also according to the descriptor chain length. That's how the Linux > > block layer deals with queue limits (struct queue_limits max_segments vs > > max_hw_sectors). >=20 > Hmm, can't follow on that one. For what should that be needed in case of = 9p? > My plan was to limit msize by 9p client simply at session start to whatev= er is > the max. amount virtio descriptors supported by host and using PAGE_SIZE = as > size per descriptor, because that's what 9p client actually does ATM (see > above). So you think that should be changed to e.g. just one descriptor f= or > 4MB, right? Limiting msize to the 9p transport device's maximum number of descriptors is conservative (i.e. 128 descriptors =3D 512 KB msize) because it doesn't take advantage of contiguous memory. I suggest leaving msize alone, adding a separate limit at which requests are split according to the maximum descriptor chain length, and tweaking pack_sg_list() to coalesce adjacent pages. That way msize can be large without necessarily using lots of descriptors (depending on the memory layout). Stefan --M4Sxhk+xVfddXjVp Content-Type: application/pgp-signature; name="signature.asc" -----BEGIN PGP SIGNATURE----- iQEzBAEBCAAdFiEEhpWov9P5fNqsNXdanKSrs4Grc8gFAmGCc40ACgkQnKSrs4Gr c8hYhQgAsDN1v7KcvgQyfEen8CI6V3qJ+z5io48nM2Nh7SN0uAVdWQmptAX2/Q8Z tiipQeIsEgIg9UbU42XttA6YVXzB/ogon7c0j/lv1YE53G9Ky1KolAIngPf0fBj8 yAtUG0ON7ZLnaecVYEfnZxbjQZ4xuw87nFGGynRty8064dUXhtEpJYy8zSd3pBIE 8UXU/Ojs1gaihtC2I2XhqXrLHce8vDE3S09xO0kQlzHGuDUhtjE46h9iqDEcAx8+ VotagROfPYpYQ/pElzzmY4r1gmJ0uZArJHxuX2G56icvQSj9b8mNDg9gJ6/DcFZZ wgDE4rfmcrn5Qlc1JtuDVctmECcV6Q== =HaoU -----END PGP SIGNATURE----- --M4Sxhk+xVfddXjVp--