From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from us-smtp-delivery-1.mimecast.com ([207.211.31.120]:48916 "EHLO us-smtp-1.mimecast.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1728903AbgBQJoJ (ORCPT ); Mon, 17 Feb 2020 04:44:09 -0500 Subject: Re: [PATCH v2 02/42] KVM: s390/interrupt: do not pin adapter interrupt pages References: <20200214222658.12946-1-borntraeger@de.ibm.com> <20200214222658.12946-3-borntraeger@de.ibm.com> From: David Hildenbrand Message-ID: <073d3666-480e-5ba5-a46b-4cbd615f4174@redhat.com> Date: Mon, 17 Feb 2020 10:43:48 +0100 MIME-Version: 1.0 In-Reply-To: <20200214222658.12946-3-borntraeger@de.ibm.com> Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: quoted-printable Sender: linux-s390-owner@vger.kernel.org List-ID: To: Christian Borntraeger , Janosch Frank Cc: KVM , Cornelia Huck , Thomas Huth , Ulrich Weigand , Claudio Imbrenda , linux-s390 , Michael Mueller , Vasily Gorbik On 14.02.20 23:26, Christian Borntraeger wrote: > From: Ulrich Weigand >=20 > The adapter interrupt page containing the indicator bits is currently > pinned. That means that a guest with many devices can pin a lot of > memory pages in the host. This also complicates the reference tracking > which is needed for memory management handling of protected virtual > machines. It might also have some strange side effects for madvise > MADV_DONTNEED and other things. >=20 > We can simply try to get the userspace page set the bits and free the > page. By storing the userspace address in the irq routing entry instead > of the guest address we can actually avoid many lookups and list walks > so that this variant is very likely not slower. >=20 > If userspace messes around with the memory slots the worst thing that > can happen is that we write to some other memory within that process. > As we get the the page with FOLL_WRITE this can also not be used to > write to shared read-only pages. >=20 > Signed-off-by: Ulrich Weigand > [borntraeger@de.ibm.com: patch simplification] > Signed-off-by: Christian Borntraeger > --- > Documentation/virt/kvm/devices/s390_flic.rst | 11 +- > arch/s390/include/asm/kvm_host.h | 3 - > arch/s390/kvm/interrupt.c | 170 ++++++------------- > 3 files changed, 53 insertions(+), 131 deletions(-) >=20 > diff --git a/Documentation/virt/kvm/devices/s390_flic.rst b/Documentati= on/virt/kvm/devices/s390_flic.rst > index 954190da7d04..ea96559ba501 100644 > --- a/Documentation/virt/kvm/devices/s390_flic.rst > +++ b/Documentation/virt/kvm/devices/s390_flic.rst > @@ -108,16 +108,9 @@ Groups: > mask or unmask the adapter, as specified in mask > =20 > KVM_S390_IO_ADAPTER_MAP > - perform a gmap translation for the guest address provided in add= r, > - pin a userspace page for the translated address and add it to th= e > - list of mappings > - > - .. note:: A new mapping will be created unconditionally; therefo= re, > - the calling code should avoid making duplicate mappings. > - > + This is now a no-op. The mapping is purely done by the irq route= . > KVM_S390_IO_ADAPTER_UNMAP > - release a userspace page for the translated address specified in= addr > - from the list of mappings > + This is now a no-op. The mapping is purely done by the irq route= . > =20 The interface should have accepted a hva from the very start and not guest addresses ... [...] > =20 > static int modify_io_adapter(struct kvm_device *dev, > @@ -2456,12 +2378,13 @@ static int modify_io_adapter(struct kvm_device = *dev, > if (ret > 0) > ret =3D 0; > break; > + /* > + * We resolve the gpa to hva when setting the IRQ routing. the set_ir= q > + * code uses get_user_pages_remote to do the actual write. nit: "get_user_pages_remote()" > + */ > case KVM_S390_IO_ADAPTER_MAP: > - ret =3D kvm_s390_adapter_map(dev->kvm, req.id, req.addr); > - break; > case KVM_S390_IO_ADAPTER_UNMAP: > - ret =3D kvm_s390_adapter_unmap(dev->kvm, req.id, req.addr); > - break; > + return 0; > default: > ret =3D -EINVAL; > } > @@ -2699,19 +2622,21 @@ static unsigned long get_ind_bit(__u64 addr, un= signed long bit_nr, bool swap) > return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit; > } > =20 > -static struct s390_map_info *get_map_info(struct s390_io_adapter *adap= ter, > - u64 addr) > +static struct page *get_map_page(struct kvm *kvm, > + struct s390_io_adapter *adapter, > + u64 uaddr) > { > - struct s390_map_info *map; > + struct page *page =3D NULL; > =20 > if (!adapter) > return NULL; AFAIKs, this check is not necessary. > - > - list_for_each_entry(map, &adapter->maps, list) { > - if (map->guest_addr =3D=3D addr) > - return map; > - } > - return NULL; > + if (!uaddr) > + return NULL; I do wonder if that check is necessary. I don't think so but might be missing something. > + down_read(&kvm->mm->mmap_sem); > + get_user_pages_remote(NULL, kvm->mm, uaddr, 1, FOLL_WRITE, > + &page, NULL, NULL); > + up_read(&kvm->mm->mmap_sem); > + return page; > } > =20 > static int adapter_indicators_set(struct kvm *kvm, > @@ -2720,30 +2645,35 @@ static int adapter_indicators_set(struct kvm *k= vm, > { > unsigned long bit; > int summary_set, idx; > - struct s390_map_info *info; > + struct page *ind_page, *summary_page; > void *map; > =20 > - info =3D get_map_info(adapter, adapter_int->ind_addr); > - if (!info) > + ind_page =3D get_map_page(kvm, adapter, adapter_int->ind_addr); > + if (!ind_page) > return -1; > - map =3D page_address(info->page); > - bit =3D get_ind_bit(info->addr, adapter_int->ind_offset, adapter->swa= p); > - set_bit(bit, map); > - idx =3D srcu_read_lock(&kvm->srcu); > - mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT); > - set_page_dirty_lock(info->page); > - info =3D get_map_info(adapter, adapter_int->summary_addr); > - if (!info) { > - srcu_read_unlock(&kvm->srcu, idx); > + summary_page =3D get_map_page(kvm, adapter, adapter_int->summary_addr= ); > + if (!summary_page) { > + put_page(ind_page); > return -1; > } > - map =3D page_address(info->page); > - bit =3D get_ind_bit(info->addr, adapter_int->summary_offset, > - adapter->swap); > + > + idx =3D srcu_read_lock(&kvm->srcu); > + map =3D page_address(ind_page); > + bit =3D get_ind_bit(adapter_int->ind_addr, > + adapter_int->ind_offset, adapter->swap); > + set_bit(bit, map); > + mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT); > + set_page_dirty_lock(ind_page); > + map =3D page_address(summary_page); > + bit =3D get_ind_bit(adapter_int->summary_addr, > + adapter_int->summary_offset, adapter->swap); > summary_set =3D test_and_set_bit(bit, map); > - mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT); > - set_page_dirty_lock(info->page); > + mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT); > + set_page_dirty_lock(summary_page); > srcu_read_unlock(&kvm->srcu, idx); > + > + put_page(ind_page); > + put_page(summary_page); > return summary_set ? 0 : 1; > } > =20 > @@ -2765,9 +2695,7 @@ static int set_adapter_int(struct kvm_kernel_irq_= routing_entry *e, > adapter =3D get_io_adapter(kvm, e->adapter.adapter_id); > if (!adapter) > return -1; > - down_read(&adapter->maps_lock); > ret =3D adapter_indicators_set(kvm, adapter, &e->adapter); > - up_read(&adapter->maps_lock); > if ((ret > 0) && !adapter->masked) { > ret =3D kvm_s390_inject_airq(kvm, adapter); > if (ret =3D=3D 0) > @@ -2818,23 +2746,27 @@ int kvm_set_routing_entry(struct kvm *kvm, > struct kvm_kernel_irq_routing_entry *e, > const struct kvm_irq_routing_entry *ue) > { > - int ret; > + u64 uaddr; > =20 > switch (ue->type) { > + /* we store the userspace addresses instead of the guest addresses */ > case KVM_IRQ_ROUTING_S390_ADAPTER: > e->set =3D set_adapter_int; > - e->adapter.summary_addr =3D ue->u.adapter.summary_addr; > - e->adapter.ind_addr =3D ue->u.adapter.ind_addr; > + uaddr =3D gmap_translate(kvm->arch.gmap, ue->u.adapter.summary_addr= ); > + if (uaddr =3D=3D -EFAULT) > + return -EFAULT; > + e->adapter.summary_addr =3D uaddr; > + uaddr =3D gmap_translate(kvm->arch.gmap, ue->u.adapter.ind_addr); > + if (uaddr =3D=3D -EFAULT) > + return -EFAULT; AFAIK, leaving e->adapter.summary_addr set is not an issue. Interesting, in kvm_s390_adapter_map(), we didn't synchronize again slot updates when doing the gmap_translate(), which looks wrong to me ... It seems to be the same thing here. I do wonder if it is safe to do a gmap_translate() here, looks like this can race with kvm_arch_commit_memory_region(). I would have assumed we need e.g., the slots_lock while doing the gmap_translate() - or a srcu_read_lock(&vcpu->kvm->srcu) or similar ... Apart from that, looks good to me. --=20 Thanks, David / dhildenb