From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 0276DA0C43; Tue, 16 Nov 2021 11:21:15 +0100 (CET) Received: from [217.70.189.124] (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id A631D410E7; Tue, 16 Nov 2021 11:21:15 +0100 (CET) Received: from us-smtp-delivery-124.mimecast.com (us-smtp-delivery-124.mimecast.com [170.10.133.124]) by mails.dpdk.org (Postfix) with ESMTP id 0674A40040 for ; Tue, 16 Nov 2021 11:21:13 +0100 (CET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=redhat.com; s=mimecast20190719; t=1637058073; 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: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=XN3ipfS5AAjVKVpNYOcbQaudLYZNoUEfLhugbgPEtik=; b=J7p2v/0cyZ+rtDMHFgQYrICrHKJStEM2rgsHLbl+tiPtdkHqhLWLiUL+1Xw/g5geOBc2O0 ys2JjO6X2gS1J0SvEhzwwfcuH6LgfHXrxMSAYo/wGCtA8o4IDaegvNZ2I1FTKHakWfo632 NqUNSZK1T/iy6EZ7mPl/UEyW0mjm3GM= 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-224-GiQ0PTwrMVWr7D-mxj3FhQ-1; Tue, 16 Nov 2021 05:21:08 -0500 X-MC-Unique: GiQ0PTwrMVWr7D-mxj3FhQ-1 Received: from smtp.corp.redhat.com (int-mx01.intmail.prod.int.phx2.redhat.com [10.5.11.11]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mimecast-mx01.redhat.com (Postfix) with ESMTPS id DE00A18125C3; Tue, 16 Nov 2021 10:21:06 +0000 (UTC) Received: from [10.39.208.9] (unknown [10.39.208.9]) by smtp.corp.redhat.com (Postfix) with ESMTPS id 0760160BD8; Tue, 16 Nov 2021 10:21:04 +0000 (UTC) Message-ID: Date: Tue, 16 Nov 2021 11:21:03 +0100 MIME-Version: 1.0 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101 Thunderbird/91.2.0 Subject: Re: [PATCH] vhost: fix get hpa fail from guest pages To: "Ding, Xuan" , "Xia, Chenbo" , "Wang, YuanX" Cc: "dev@dpdk.org" , "Hu, Jiayu" , "Ma, WenwuX" , "He, Xingguang" , "Yang, YvonneX" References: <20211111062725.108297-1-yuanx.wang@intel.com> From: Maxime Coquelin In-Reply-To: X-Scanned-By: MIMEDefang 2.79 on 10.5.11.11 Authentication-Results: relay.mimecast.com; auth=pass smtp.auth=CUSA124A263 smtp.mailfrom=maxime.coquelin@redhat.com X-Mimecast-Spam-Score: 0 X-Mimecast-Originator: redhat.com Content-Language: en-US Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 8bit X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org On 11/16/21 11:08, Ding, Xuan wrote: > Hi, > >> -----Original Message----- >> From: Xia, Chenbo >> Sent: 2021年11月16日 17:01 >> To: Ding, Xuan ; Wang, YuanX ; >> maxime.coquelin@redhat.com >> Cc: dev@dpdk.org; Hu, Jiayu ; Ma, WenwuX >> ; He, Xingguang ; Yang, >> YvonneX >> Subject: RE: [PATCH] vhost: fix get hpa fail from guest pages >> >>> -----Original Message----- >>> From: Ding, Xuan >>> Sent: Tuesday, November 16, 2021 4:47 PM >>> To: Xia, Chenbo ; Wang, YuanX >>> ; maxime.coquelin@redhat.com >>> Cc: dev@dpdk.org; Hu, Jiayu ; Ma, WenwuX >>> ; He, Xingguang ; Yang, >>> YvonneX >>> Subject: RE: [PATCH] vhost: fix get hpa fail from guest pages >>> >>> Hi Chenbo, >>> >>>> -----Original Message----- >>>> From: Xia, Chenbo >>>> Sent: 2021年11月16日 16:07 >>>> To: Xia, Chenbo ; Wang, YuanX >>>> ; maxime.coquelin@redhat.com >>>> Cc: dev@dpdk.org; Hu, Jiayu ; Ding, Xuan >>>> ; Ma, WenwuX ; He, >>>> Xingguang ; Yang, YvonneX >>>> >>>> Subject: RE: [PATCH] vhost: fix get hpa fail from guest pages >>>> >>>>> -----Original Message----- >>>>> From: Xia, Chenbo >>>>> Sent: Monday, November 15, 2021 4:04 PM >>>>> To: Wang, YuanX ; >> maxime.coquelin@redhat.com >>>>> Cc: dev@dpdk.org; Hu, Jiayu ; Ding, Xuan >>>>> ; Ma, WenwuX ; He, >>>> Xingguang >>>>> ; Yang, YvonneX >>>>> Subject: RE: [PATCH] vhost: fix get hpa fail from guest pages >>>>> >>>>>> -----Original Message----- >>>>>> From: Wang, YuanX >>>>>> Sent: Thursday, November 11, 2021 2:27 PM >>>>>> To: maxime.coquelin@redhat.com; Xia, Chenbo >>>>>> >>>>>> Cc: dev@dpdk.org; Hu, Jiayu ; Ding, Xuan >>>>>> ; Ma, WenwuX ; He, >>>>>> Xingguang ; Yang, YvonneX >>>>>> ; Wang, >>>>> YuanX >>>>>> >>>>>> Subject: [PATCH] vhost: fix get hpa fail from guest pages >>>>>> >>>>>> When processing front-end memory regions messages, vhost saves >>>>>> the guest/host physical address mappings to guest pages and >>>>>> merges adjacent contiguous pages if hpa is contiguous, however >>>>>> gpa is likely not contiguous in PA mode and merging will cause >>>>>> the gpa range to change. >>>>>> This patch distinguishes the case of discontinuous gpa and does >>>>>> a range lookup on gpa when doing a binary search. >>>>>> >>>>>> Fixes: e246896178e("vhost: get guest/host physical address >>>>>> mappings") >>>>>> Fixes: 6563cf92380 ("vhost: fix async copy on multi-page >>>>>> buffers") >>>>>> >>>>>> Signed-off-by: Yuan Wang >>>>>> --- >>>>>> lib/vhost/vhost.h | 18 ++++++++++++++++-- >>>>>> lib/vhost/vhost_user.c | 15 +++++++++++---- >>>>>> 2 files changed, 27 insertions(+), 6 deletions(-) >>>>>> >>>>>> diff --git a/lib/vhost/vhost.h b/lib/vhost/vhost.h index >>>>>> 7085e0885c..b3f0c1d07c 100644 >>>>>> --- a/lib/vhost/vhost.h >>>>>> +++ b/lib/vhost/vhost.h >>>>>> @@ -587,6 +587,20 @@ static __rte_always_inline int >>>>>> guest_page_addrcmp(const void *p1, >>>>>> return 0; >>>>>> } >>>>>> >>>>>> +static __rte_always_inline int guest_page_rangecmp(const void >>>>>> +*p1, const >>>>> void >>>>>> *p2) >>>>>> +{ >>>>>> + const struct guest_page *page1 = (const struct guest_page *)p1; >>>>>> + const struct guest_page *page2 = (const struct guest_page >>>>>> +*)p2; >>>>>> + >>>>>> + if (page1->guest_phys_addr >= page2->guest_phys_addr) { >>>>>> + if (page1->guest_phys_addr < page2->guest_phys_addr >> + >>>> page2->size) >>>>>> + return 0; >>>>>> + else >>>>>> + return 1; >>>>>> + } else >>>>>> + return -1; >>>>>> +} >>>>>> + >>>>>> static __rte_always_inline rte_iova_t gpa_to_first_hpa(struct >>>>>> virtio_net *dev, uint64_t gpa, >>>>>> uint64_t gpa_size, uint64_t *hpa_size) @@ -597,9 +611,9 @@ >>>>>> gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa, >>>>>> >>>>>> *hpa_size = gpa_size; >>>>>> if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) { >>>>>> - key.guest_phys_addr = gpa & ~(dev->guest_pages[0].size - 1); >>>>>> + key.guest_phys_addr = gpa; >>>>>> page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages, >>>>>> - sizeof(struct guest_page), guest_page_addrcmp); >>>>>> + sizeof(struct guest_page), >> guest_page_rangecmp); >>>>>> if (page) { >>>>>> if (gpa + gpa_size <= >>>>>> page->guest_phys_addr + page->size) >>>> { diff --git >>>>>> a/lib/vhost/vhost_user.c b/lib/vhost/vhost_user.c index >>>>>> a781346c4d..7d58fde458 100644 >>>>>> --- a/lib/vhost/vhost_user.c >>>>>> +++ b/lib/vhost/vhost_user.c >>>>>> @@ -999,10 +999,17 @@ add_one_guest_page(struct virtio_net *dev, >>>>>> uint64_t guest_phys_addr, >>>>>> if (dev->nr_guest_pages > 0) { >>>>>> last_page = &dev->guest_pages[dev->nr_guest_pages - 1]; >>>>>> /* merge if the two pages are continuous */ >>>>>> - if (host_phys_addr == last_page->host_phys_addr + >>>>>> - last_page->size) { >>>>>> - last_page->size += size; >>>>>> - return 0; >>>>>> + if (host_phys_addr == last_page->host_phys_addr + >>>>>> +last_page->size) >>>>>> { >>>>>> + if (rte_eal_iova_mode() == RTE_IOVA_VA) { >>>>>> + last_page->size += size; >>>>>> + return 0; >>>>>> + } >>>>> >>>>> This makes me think about a question: In IOVA_VA mode, what >>>>> ensures HPA and GPA are both contiguous? >>>> >>>> When I wrote this email, I thought host_phys_addr is HPA but in VA >>>> mode, >>> it's >>>> Host IOVA, aka VA in DPDK's case. So in most cases GPA and VA will >>>> be both contiguous when the contiguous pages are all in one memory >>>> region. But I >>> think >>>> we should not make such assumption as some vhost master may send >>>> GPA- contiguous pages in different memory region, then the VA after >>>> mmap may not be contiguous. >>> >>> Now we do async vfio mapping at page granularity. >>> For 4K/2M pages, without this assumption, pages not be merged may >>> exceed the IOMMU's capability. This limits only use 1G hugepage for DMA >> device. >> >> I was not saying merging should be done, but merging with correct logic. >> >> That is to say, the pages that should be merged will still be merged, but make >> the Logic to handle all cases. > > Sorry, I misunderstood your point. I was to add a supplement, > PA mode will have lots of pages to map for 4K/2M cases, > it may exceed IOMMU's capability. But it is rational, that's what I hope > to explain in the doc. We are running out of time for v21.11 release. Do you agree to postpone the fix to new release? It can be backported to v21.11 LTS later. > Thanks, > Xuan > >> >> Thanks, >> Chenbo >> >>> >>> Thanks, >>> Xuan >>> >>>> >>>> Maxime, What do you think? >>>> >>>> Thanks, >>>> Chenbo >>>> >>>>> >>>>> Maxime & Yuan, any thought? >>>>> >>>>> Thanks, >>>>> Chenbo >>>>> >>>>>> + >>>>>> + if (rte_eal_iova_mode() == RTE_IOVA_PA && >>>>>> + guest_phys_addr == last_page- >>>>> guest_phys_addr + >>>>>> last_page->size) { >>>>>> + last_page->size += size; >>>>>> + return 0; >>>>>> + } >>>>>> } >>>>>> } >>>>>> >>>>>> -- >>>>>> 2.25.1 >