Re: [Qemu-devel] [PATCH qemu v16 14/19] vfio: spapr: Add DMA memory preregistering (SPAPR IOMMU v2)

[David Gibson]

On Mon, May 16, 2016 at 02:20:33PM -0600, Alex Williamson wrote:
> On Mon, 16 May 2016 11:10:05 +1000
> Alexey Kardashevskiy <address@hidden> wrote:
> 
> > On 05/14/2016 08:25 AM, Alex Williamson wrote:
> > > On Wed,  4 May 2016 16:52:26 +1000
> > > Alexey Kardashevskiy <address@hidden> wrote:
> > >  
> > >> This makes use of the new "memory registering" feature. The idea is
> > >> to provide the userspace ability to notify the host kernel about pages
> > >> which are going to be used for DMA. Having this information, the host
> > >> kernel can pin them all once per user process, do locked pages
> > >> accounting (once) and not spent time on doing that in real time with
> > >> possible failures which cannot be handled nicely in some cases.
> > >>
> > >> This adds a prereg memory listener which listens on address_space_memory
> > >> and notifies a VFIO container about memory which needs to be
> > >> pinned/unpinned. VFIO MMIO regions (i.e. "skip dump" regions) are 
> > >> skipped.
> > >>
> > >> As there is no per-IOMMU-type release() callback anymore, this stores
> > >> the IOMMU type in the container so vfio_listener_release() can determine
> > >> if it needs to unregister @prereg_listener.
> > >>
> > >> The feature is only enabled for SPAPR IOMMU v2. The host kernel changes
> > >> are required. Since v2 does not need/support VFIO_IOMMU_ENABLE, this does
> > >> not call it when v2 is detected and enabled.
> > >>
> > >> This enforces guest RAM blocks to be host page size aligned; however
> > >> this is not new as KVM already requires memory slots to be host page
> > >> size aligned.
> > >>
> > >> Signed-off-by: Alexey Kardashevskiy <address@hidden>
> > >> ---
> > >> Changes:
> > >> v16:
> > >> * switched to 64bit math everywhere as there is no chance to see
> > >> region_add on RAM blocks even remotely close to 1<<64bytes.
> > >>
> > >> v15:
> > >> * banned unaligned sections
> > >> * added an vfio_prereg_gpa_to_ua() helper
> > >>
> > >> v14:
> > >> * s/free_container_exit/listener_release_exit/g
> > >> * added "if memory_region_is_iommu()" to 
> > >> vfio_prereg_listener_skipped_section
> > >> ---
> > >>  hw/vfio/Makefile.objs         |   1 +
> > >>  hw/vfio/common.c              |  38 +++++++++---
> > >>  hw/vfio/prereg.c              | 137 
> > >> ++++++++++++++++++++++++++++++++++++++++++
> > >>  include/hw/vfio/vfio-common.h |   4 ++
> > >>  trace-events                  |   2 +
> > >>  5 files changed, 172 insertions(+), 10 deletions(-)
> > >>  create mode 100644 hw/vfio/prereg.c
> > >>
> > >> diff --git a/hw/vfio/Makefile.objs b/hw/vfio/Makefile.objs
> > >> index ceddbb8..5800e0e 100644
> > >> --- a/hw/vfio/Makefile.objs
> > >> +++ b/hw/vfio/Makefile.objs
> > >> @@ -4,4 +4,5 @@ obj-$(CONFIG_PCI) += pci.o pci-quirks.o
> > >>  obj-$(CONFIG_SOFTMMU) += platform.o
> > >>  obj-$(CONFIG_SOFTMMU) += calxeda-xgmac.o
> > >>  obj-$(CONFIG_SOFTMMU) += amd-xgbe.o
> > >> +obj-$(CONFIG_SOFTMMU) += prereg.o
> > >>  endif
> > >> diff --git a/hw/vfio/common.c b/hw/vfio/common.c
> > >> index 2050040..496eb82 100644
> > >> --- a/hw/vfio/common.c
> > >> +++ b/hw/vfio/common.c
> > >> @@ -501,6 +501,9 @@ static const MemoryListener vfio_memory_listener = {
> > >>  static void vfio_listener_release(VFIOContainer *container)
> > >>  {
> > >>      memory_listener_unregister(&container->listener);
> > >> +    if (container->iommu_type == VFIO_SPAPR_TCE_v2_IOMMU) {
> > >> +        memory_listener_unregister(&container->prereg_listener);
> > >> +    }
> > >>  }
> > >>
> > >>  int vfio_region_setup(Object *obj, VFIODevice *vbasedev, VFIORegion 
> > >> *region,
> > >> @@ -808,8 +811,8 @@ static int vfio_connect_container(VFIOGroup *group, 
> > >> AddressSpace *as)
> > >>              goto free_container_exit;
> > >>          }
> > >>
> > >> -        ret = ioctl(fd, VFIO_SET_IOMMU,
> > >> -                    v2 ? VFIO_TYPE1v2_IOMMU : VFIO_TYPE1_IOMMU);
> > >> +        container->iommu_type = v2 ? VFIO_TYPE1v2_IOMMU : 
> > >> VFIO_TYPE1_IOMMU;
> > >> +        ret = ioctl(fd, VFIO_SET_IOMMU, container->iommu_type);
> > >>          if (ret) {
> > >>              error_report("vfio: failed to set iommu for container: %m");
> > >>              ret = -errno;
> > >> @@ -834,8 +837,10 @@ static int vfio_connect_container(VFIOGroup *group, 
> > >> AddressSpace *as)
> > >>          if ((ret == 0) && (info.flags & VFIO_IOMMU_INFO_PGSIZES)) {
> > >>              container->iova_pgsizes = info.iova_pgsizes;
> > >>          }
> > >> -    } else if (ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_SPAPR_TCE_IOMMU)) {
> > >> +    } else if (ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_SPAPR_TCE_IOMMU) ||
> > >> +               ioctl(fd, VFIO_CHECK_EXTENSION, 
> > >> VFIO_SPAPR_TCE_v2_IOMMU)) {
> > >>          struct vfio_iommu_spapr_tce_info info;
> > >> +        bool v2 = !!ioctl(fd, VFIO_CHECK_EXTENSION, 
> > >> VFIO_SPAPR_TCE_v2_IOMMU);
> > >>
> > >>          ret = ioctl(group->fd, VFIO_GROUP_SET_CONTAINER, &fd);
> > >>          if (ret) {
> > >> @@ -843,7 +848,9 @@ static int vfio_connect_container(VFIOGroup *group, 
> > >> AddressSpace *as)
> > >>              ret = -errno;
> > >>              goto free_container_exit;
> > >>          }
> > >> -        ret = ioctl(fd, VFIO_SET_IOMMU, VFIO_SPAPR_TCE_IOMMU);
> > >> +        container->iommu_type =
> > >> +            v2 ? VFIO_SPAPR_TCE_v2_IOMMU : VFIO_SPAPR_TCE_IOMMU;
> > >> +        ret = ioctl(fd, VFIO_SET_IOMMU, container->iommu_type);
> > >>          if (ret) {
> > >>              error_report("vfio: failed to set iommu for container: %m");
> > >>              ret = -errno;
> > >> @@ -855,11 +862,22 @@ static int vfio_connect_container(VFIOGroup 
> > >> *group, AddressSpace *as)
> > >>           * when container fd is closed so we do not call it explicitly
> > >>           * in this file.
> > >>           */
> > >> -        ret = ioctl(fd, VFIO_IOMMU_ENABLE);
> > >> -        if (ret) {
> > >> -            error_report("vfio: failed to enable container: %m");
> > >> -            ret = -errno;
> > >> -            goto free_container_exit;
> > >> +        if (!v2) {
> > >> +            ret = ioctl(fd, VFIO_IOMMU_ENABLE);
> > >> +            if (ret) {
> > >> +                error_report("vfio: failed to enable container: %m");
> > >> +                ret = -errno;
> > >> +                goto free_container_exit;
> > >> +            }
> > >> +        } else {
> > >> +            container->prereg_listener = vfio_prereg_listener;
> > >> +
> > >> +            memory_listener_register(&container->prereg_listener,
> > >> +                                     &address_space_memory);
> > >> +            if (container->error) {
> > >> +                error_report("vfio: RAM memory listener initialization 
> > >> failed for container");
> > >> +                goto listener_release_exit;
> > >> +            }
> > >>          }
> > >>
> > >>          /*
> > >> @@ -872,7 +890,7 @@ static int vfio_connect_container(VFIOGroup *group, 
> > >> AddressSpace *as)
> > >>          if (ret) {
> > >>              error_report("vfio: VFIO_IOMMU_SPAPR_TCE_GET_INFO failed: 
> > >> %m");
> > >>              ret = -errno;
> > >> -            goto free_container_exit;
> > >> +            goto listener_release_exit;
> > >>          }
> > >>          container->min_iova = info.dma32_window_start;
> > >>          container->max_iova = container->min_iova + 
> > >> info.dma32_window_size - 1;
> > >> diff --git a/hw/vfio/prereg.c b/hw/vfio/prereg.c
> > >> new file mode 100644
> > >> index 0000000..d0e4728
> > >> --- /dev/null
> > >> +++ b/hw/vfio/prereg.c
> > >> @@ -0,0 +1,137 @@
> > >> +/*
> > >> + * DMA memory preregistration
> > >> + *
> > >> + * Authors:
> > >> + *  Alexey Kardashevskiy <address@hidden>
> > >> + *
> > >> + * This work is licensed under the terms of the GNU GPL, version 2.  See
> > >> + * the COPYING file in the top-level directory.
> > >> + */
> > >> +
> > >> +#include "qemu/osdep.h"
> > >> +#include <sys/ioctl.h>
> > >> +#include <linux/vfio.h>
> > >> +
> > >> +#include "hw/vfio/vfio-common.h"
> > >> +#include "hw/hw.h"
> > >> +#include "qemu/error-report.h"
> > >> +#include "trace.h"
> > >> +
> > >> +static bool vfio_prereg_listener_skipped_section(MemoryRegionSection 
> > >> *section)
> > >> +{
> > >> +    if (memory_region_is_iommu(section->mr)) {
> > >> +        error_report("Cannot possibly preregister IOMMU memory");  
> > >
> > > What is a user supposed to do with this error_report()?  Is it
> > > continue-able?  How is it possible?  What should they do differently?  
> > 
> > 
> > If I remember correctly, David did have theories where this may be 
> > possible, not today with the existing code though. Could be assert() or 
> > abort(), what is better here?
> 
> If it's a hardware configuration error, then use hw_error(), I prefer
> not to add either assert() or abort() calls to vfio.

Personally I would have gone with assert() - hitting this is _almost_
certainly an indication of a bug in the code (an IOMMU translating
into an AS that itself has IOMMUs).  In theory a combination of a
wierd platform with a multi-layered IOMMU and the wrong config options
could trigger it, but it's very unlikely.

hw_error() is fine by me too.

> > >> +        return true;
> > >> +    }
> > >> +
> > >> +    return !memory_region_is_ram(section->mr) ||
> > >> +            memory_region_is_skip_dump(section->mr);
> > >> +}
> > >> +
> > >> +static void *vfio_prereg_gpa_to_ua(MemoryRegionSection *section, hwaddr 
> > >> gpa)  
> > >
> > > What's "ua"?  
> > 
> > 
> > Userspace address.
> 
> But we use it to set a vaddr below, so let's just call it vaddr.
> 
> > >  
> > >> +{
> > >> +    return memory_region_get_ram_ptr(section->mr) +
> > >> +        section->offset_within_region +
> > >> +        (gpa - section->offset_within_address_space);
> > >> +}
> > >> +
> > >> +static void vfio_prereg_listener_region_add(MemoryListener *listener,
> > >> +                                            MemoryRegionSection 
> > >> *section)
> > >> +{
> > >> +    VFIOContainer *container = container_of(listener, VFIOContainer,
> > >> +                                            prereg_listener);
> > >> +    const hwaddr gpa = section->offset_within_address_space;
> > >> +    hwaddr end;
> > >> +    int ret;
> > >> +    hwaddr page_mask = qemu_real_host_page_mask;
> > >> +    struct vfio_iommu_spapr_register_memory reg = {
> > >> +        .argsz = sizeof(reg),
> > >> +        .flags = 0,
> > >> +    };  
> > >
> > > So we're just pretending that this spapr specific code is some sort of
> > > generic pre-registration interface?  
> > 
> > Yes.
> 
> :-\

It's a bit of an odd mix because the actual caps and ioctls are spapr
specific, but there's nothing inherently spapr specific about the
concept of pre-registration.  It's only implemented for spapr now, but
pre-reg would probably be a good idea for performance on any platform
where the guest expects to actively manage an IOMMU with a reasonably
small window.

> 
> > >> +
> > >> +    if (vfio_prereg_listener_skipped_section(section)) {
> > >> +        trace_vfio_listener_region_add_skip(
> > >> +                section->offset_within_address_space,
> > >> +                section->offset_within_address_space +
> > >> +                int128_get64(int128_sub(section->size, int128_one())));
> > >> +        return;
> > >> +    }
> > >> +
> > >> +    if (unlikely((section->offset_within_address_space & ~page_mask) ||
> > >> +                 (section->offset_within_region & ~page_mask) ||
> > >> +                 (int128_get64(section->size) & ~page_mask))) {
> > >> +        error_report("%s received unaligned region", __func__);
> > >> +        return;
> > >> +    }
> > >> +
> > >> +    end = section->offset_within_address_space + 
> > >> int128_get64(section->size);
> > >> +    g_assert(gpa < end);
> > >> +
> > >> +    memory_region_ref(section->mr);
> > >> +
> > >> +    reg.vaddr = (__u64) vfio_prereg_gpa_to_ua(section, gpa);  
> > >
> > > Hmm, why wasn't that simply gpa_to_vaddr?  
> > 
> > I wanted to keep a prefix in all functions, even if they are static, easier 
> > to grep. Bad idea?
> 
> My question about "ua" means that it's not obvious what we're returning
> based on the name of the function alone, so I would avoid such a name.
> 
> > >> +    reg.size = end - gpa;
> > >> +
> > >> +    ret = ioctl(container->fd, VFIO_IOMMU_SPAPR_REGISTER_MEMORY, &reg);
> > >> +    trace_vfio_ram_register(reg.vaddr, reg.size, ret ? -errno : 0);
> > >> +    if (ret) {
> > >> +        /*
> > >> +         * On the initfn path, store the first error in the container 
> > >> so we
> > >> +         * can gracefully fail.  Runtime, there's not much we can do 
> > >> other
> > >> +         * than throw a hardware error.
> > >> +         */
> > >> +        if (!container->initialized) {
> > >> +            if (!container->error) {
> > >> +                container->error = ret;
> > >> +            }
> > >> +        } else {
> > >> +            hw_error("vfio: Memory registering failed, unable to 
> > >> continue");
> > >> +        }
> > >> +    }
> > >> +}
> > >> +
> > >> +static void vfio_prereg_listener_region_del(MemoryListener *listener,
> > >> +                                            MemoryRegionSection 
> > >> *section)
> > >> +{
> > >> +    VFIOContainer *container = container_of(listener, VFIOContainer,
> > >> +                                            prereg_listener);
> > >> +    const hwaddr gpa = section->offset_within_address_space;
> > >> +    hwaddr end;
> > >> +    int ret;
> > >> +    hwaddr page_mask = qemu_real_host_page_mask;
> > >> +    struct vfio_iommu_spapr_register_memory reg = {
> > >> +        .argsz = sizeof(reg),
> > >> +        .flags = 0,
> > >> +    };
> > >> +
> > >> +    if (vfio_prereg_listener_skipped_section(section)) {
> > >> +        trace_vfio_listener_region_del_skip(
> > >> +                section->offset_within_address_space,
> > >> +                section->offset_within_address_space +
> > >> +                int128_get64(int128_sub(section->size, int128_one())));
> > >> +        return;
> > >> +    }
> > >> +
> > >> +    if (unlikely((section->offset_within_address_space & ~page_mask) ||
> > >> +                 (section->offset_within_region & ~page_mask) ||
> > >> +                 (int128_get64(section->size) & ~page_mask))) {
> > >> +        error_report("%s received unaligned region", __func__);
> > >> +        return;
> > >> +    }
> > >> +
> > >> +    end = section->offset_within_address_space + 
> > >> int128_get64(section->size);
> > >> +    if (gpa >= end) {
> > >> +        return;
> > >> +    }
> > >> +
> > >> +    reg.vaddr = (__u64) vfio_prereg_gpa_to_ua(section, gpa);
> > >> +    reg.size = end - gpa;
> > >> +
> > >> +    ret = ioctl(container->fd, VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY, 
> > >> &reg);
> > >> +    trace_vfio_ram_unregister(reg.vaddr, reg.size, ret ? -errno : 0);
> > >> +}
> > >> +
> > >> +const MemoryListener vfio_prereg_listener = {
> > >> +    .region_add = vfio_prereg_listener_region_add,
> > >> +    .region_del = vfio_prereg_listener_region_del,
> > >> +};
> > >> diff --git a/include/hw/vfio/vfio-common.h 
> > >> b/include/hw/vfio/vfio-common.h
> > >> index c9b6622..c72e45a 100644
> > >> --- a/include/hw/vfio/vfio-common.h
> > >> +++ b/include/hw/vfio/vfio-common.h
> > >> @@ -73,6 +73,8 @@ typedef struct VFIOContainer {
> > >>      VFIOAddressSpace *space;
> > >>      int fd; /* /dev/vfio/vfio, empowered by the attached groups */
> > >>      MemoryListener listener;
> > >> +    MemoryListener prereg_listener;
> > >> +    unsigned iommu_type;
> > >>      int error;
> > >>      bool initialized;
> > >>      /*
> > >> @@ -156,4 +158,6 @@ extern QLIST_HEAD(vfio_as_head, VFIOAddressSpace) 
> > >> vfio_address_spaces;
> > >>  int vfio_get_region_info(VFIODevice *vbasedev, int index,
> > >>                           struct vfio_region_info **info);
> > >>  #endif
> > >> +extern const MemoryListener vfio_prereg_listener;
> > >> +
> > >>  #endif /* !HW_VFIO_VFIO_COMMON_H */
> > >> diff --git a/trace-events b/trace-events
> > >> index dd50005..d0d8615 100644
> > >> --- a/trace-events
> > >> +++ b/trace-events
> > >> @@ -1737,6 +1737,8 @@ vfio_region_mmap(const char *name, unsigned long 
> > >> offset, unsigned long end) "Reg
> > >>  vfio_region_exit(const char *name, int index) "Device %s, region %d"
> > >>  vfio_region_finalize(const char *name, int index) "Device %s, region %d"
> > >>  vfio_region_mmaps_set_enabled(const char *name, bool enabled) "Region 
> > >> %s mmaps enabled: %d"
> > >> +vfio_ram_register(uint64_t va, uint64_t size, int ret) "va=%"PRIx64" 
> > >> size=%"PRIx64" ret=%d"
> > >> +vfio_ram_unregister(uint64_t va, uint64_t size, int ret) "va=%"PRIx64" 
> > >> size=%"PRIx64" ret=%d"
> > >>
> > >>  # hw/vfio/platform.c
> > >>  vfio_platform_base_device_init(char *name, int groupid) "%s belongs to 
> > >> group #%d"  
> > 
> > 
> 

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

signature.asc
Description: PGP signature