Hi Marcelo! Marcelo Tosatti wrote: > On Fri, Oct 02, 2009 at 04:19:27PM -0400, Gregory Haskins wrote: >> What: xinterface is a mechanism that allows kernel modules external to >> the kvm.ko proper to interface with a running guest. It accomplishes >> this by creating an abstracted interface which does not expose any >> private details of the guest or its related KVM structures, and provides >> a mechanism to find and bind to this interface at run-time. >> >> Why: There are various subsystems that would like to interact with a KVM >> guest which are ideally suited to exist outside the domain of the kvm.ko >> core logic. For instance, external pci-passthrough, virtual-bus, and >> virtio-net modules are currently under development. In order for these >> modules to successfully interact with the guest, they need, at the very >> least, various interfaces for signaling IO events, pointer translation, >> and possibly memory mapping. >> >> The signaling case is covered by the recent introduction of the >> irqfd/ioeventfd mechanisms. This patch provides a mechanism to cover the >> other cases. Note that today we only expose pointer-translation related >> functions, but more could be added at a future date as needs arise. >> >> Example usage: QEMU instantiates a guest, and an external module "foo" >> that desires the ability to interface with the guest (say via >> open("/dev/foo")). QEMU may then pass the kvmfd to foo via an >> ioctl, such as: ioctl(foofd, FOO_SET_VMID, &kvmfd). Upon receipt, the >> foo module can issue kvm_xinterface_bind(kvmfd) to acquire >> the proper context. Internally, the struct kvm* and associated >> struct module* will remain pinned at least until the foo module calls >> kvm_xinterface_put(). > >> --- /dev/null >> +++ b/virt/kvm/xinterface.c >> @@ -0,0 +1,409 @@ >> +/* >> + * KVM module interface - Allows external modules to interface with a guest >> + * >> + * Copyright 2009 Novell. All Rights Reserved. >> + * >> + * Author: >> + * Gregory Haskins <ghaskins@xxxxxxxxxx> >> + * >> + * This file is free software; you can redistribute it and/or modify >> + * it under the terms of version 2 of the GNU General Public License >> + * as published by the Free Software Foundation. >> + * >> + * This program is distributed in the hope that it will be useful, >> + * but WITHOUT ANY WARRANTY; without even the implied warranty of >> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the >> + * GNU General Public License for more details. >> + * >> + * You should have received a copy of the GNU General Public License >> + * along with this program; if not, write to the Free Software Foundation, >> + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. >> + */ >> + >> +#include <linux/mm.h> >> +#include <linux/vmalloc.h> >> +#include <linux/highmem.h> >> +#include <linux/module.h> >> +#include <linux/mmu_context.h> >> +#include <linux/kvm_host.h> >> +#include <linux/kvm_xinterface.h> >> + >> +struct _xinterface { >> + struct kvm *kvm; >> + struct task_struct *task; >> + struct mm_struct *mm; >> + struct kvm_xinterface intf; >> + struct kvm_memory_slot *slotcache[NR_CPUS]; >> +}; >> + >> +struct _xvmap { >> + struct kvm_memory_slot *memslot; >> + unsigned long npages; >> + struct kvm_xvmap vmap; >> +}; >> + >> +static struct _xinterface * >> +to_intf(struct kvm_xinterface *intf) >> +{ >> + return container_of(intf, struct _xinterface, intf); >> +} >> + >> +#define _gfn_to_hva(gfn, memslot) \ >> + (memslot->userspace_addr + (gfn - memslot->base_gfn) * PAGE_SIZE) >> + >> +/* >> + * gpa_to_hva() - translate a guest-physical to host-virtual using >> + * a per-cpu cache of the memslot. >> + * >> + * The gfn_to_memslot() call is relatively expensive, and the gpa access >> + * patterns exhibit a high degree of locality. Therefore, lets cache >> + * the last slot used on a per-cpu basis to optimize the lookup >> + * >> + * assumes slots_lock held for read >> + */ >> +static unsigned long >> +gpa_to_hva(struct _xinterface *_intf, unsigned long gpa) >> +{ >> + int cpu = get_cpu(); >> + unsigned long gfn = gpa >> PAGE_SHIFT; >> + struct kvm_memory_slot *memslot = _intf->slotcache[cpu]; >> + unsigned long addr = 0; >> + >> + if (!memslot >> + || gfn < memslot->base_gfn >> + || gfn >= memslot->base_gfn + memslot->npages) { >> + >> + memslot = gfn_to_memslot(_intf->kvm, gfn); >> + if (!memslot) >> + goto out; >> + >> + _intf->slotcache[cpu] = memslot; >> + } >> + >> + addr = _gfn_to_hva(gfn, memslot) + offset_in_page(gpa); >> + >> +out: >> + put_cpu(); >> + >> + return addr; > > Please optimize gfn_to_memslot() instead, so everybody benefits. It > shows very often on profiles. Yeah, its not a bad idea. The reason why I did it here is because the requirements for sync (kvm-vcpu) vs async (xinterface) access is slightly different. Sync is probably optimal with per-vcpu caching, whereas async is optimal with per-cpu. That said, we could probably build the entire algorithm to be per-cpu as a compromise and still gain benefits. Perhaps I will split this out as a separate patch for v3. > >> + >> + page_list = (struct page **) __get_free_page(GFP_KERNEL); >> + if (!page_list) >> + return NULL; >> + >> + down_write(&mm->mmap_sem); >> + >> + ret = get_user_pages(p, mm, addr, npages, 1, 0, page_list, NULL); >> + if (ret < 0) >> + goto out; >> + >> + ptr = vmap(page_list, npages, VM_MAP, PAGE_KERNEL); >> + if (ptr) >> + mm->locked_vm += npages; > > Why don't you use gfn_to_page (here and elsewhere in the patch). Primarily ignorance, I suspect ;) The truth is I ported this from one of our other connectors, which was more userspace oriented and thus gup() made sense and gtp() was not an option. That said, it probably doesn't matter a ton in the vmap case, because that is slow-path. However, I will definitely look to change over to the gtp() variant, especially if it affects any fast path code. Thanks Marcelo, -Greg
Attachment:
signature.asc
Description: OpenPGP digital signature