Re: [PATCH v1] vfio/common: Separate vfio-pci ranges

[Joao Martins <joao.m.martins@oracle.com>]

On 08/09/2023 10:29, Joao Martins wrote:
> QEMU computes the DMA logging ranges for two predefined ranges: 32-bit
> and 64-bit. In the OVMF case, when the dynamic MMIO window is enabled,
> QEMU includes in the 64-bit range the RAM regions at the lower part
> and vfio-pci device RAM regions which are at the top of the address
> space. This range contains a large gap and the size can be bigger than
> the dirty tracking HW limits of some devices (MLX5 has a 2^42 limit).
> 
> To avoid such large ranges, introduce a new PCI range covering the
> vfio-pci device RAM regions, this only if the addresses are above 4GB
> to avoid breaking potential SeaBIOS guests.
> 
> Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
> [ clg: - wrote commit log
>        - fixed overlapping 32-bit and PCI ranges when using SeaBIOS ]
> Signed-off-by: Cédric Le Goater <clg@redhat.com>

This should have a:

Fixes: 5255bbf4ec16 ("vfio/common: Add device dirty page tracking start/stop")

> ---
> v2:
> * s/minpci/minpci64/
> * s/maxpci/maxpci64/
> * Expand comment to cover the pci-hole64 and why we don't do special
>   handling of pci-hole32.
> ---
>  hw/vfio/common.c     | 71 +++++++++++++++++++++++++++++++++++++-------
>  hw/vfio/trace-events |  2 +-
>  2 files changed, 61 insertions(+), 12 deletions(-)
> 
> diff --git a/hw/vfio/common.c b/hw/vfio/common.c
> index 237101d03844..134649226d43 100644
> --- a/hw/vfio/common.c
> +++ b/hw/vfio/common.c
> @@ -27,6 +27,7 @@
>  
>  #include "hw/vfio/vfio-common.h"
>  #include "hw/vfio/vfio.h"
> +#include "hw/vfio/pci.h"
>  #include "exec/address-spaces.h"
>  #include "exec/memory.h"
>  #include "exec/ram_addr.h"
> @@ -1400,6 +1401,8 @@ typedef struct VFIODirtyRanges {
>      hwaddr max32;
>      hwaddr min64;
>      hwaddr max64;
> +    hwaddr minpci64;
> +    hwaddr maxpci64;
>  } VFIODirtyRanges;
>  
>  typedef struct VFIODirtyRangesListener {
> @@ -1408,6 +1411,31 @@ typedef struct VFIODirtyRangesListener {
>      MemoryListener listener;
>  } VFIODirtyRangesListener;
>  
> +static bool vfio_section_is_vfio_pci(MemoryRegionSection *section,
> +                                     VFIOContainer *container)
> +{
> +    VFIOPCIDevice *pcidev;
> +    VFIODevice *vbasedev;
> +    VFIOGroup *group;
> +    Object *owner;
> +
> +    owner = memory_region_owner(section->mr);
> +
> +    QLIST_FOREACH(group, &container->group_list, container_next) {
> +        QLIST_FOREACH(vbasedev, &group->device_list, next) {
> +            if (vbasedev->type != VFIO_DEVICE_TYPE_PCI) {
> +                continue;
> +            }
> +            pcidev = container_of(vbasedev, VFIOPCIDevice, vbasedev);
> +            if (OBJECT(pcidev) == owner) {
> +                return true;
> +            }
> +        }
> +    }
> +
> +    return false;
> +}
> +
>  static void vfio_dirty_tracking_update(MemoryListener *listener,
>                                         MemoryRegionSection *section)
>  {
> @@ -1424,19 +1452,32 @@ static void vfio_dirty_tracking_update(MemoryListener *listener,
>      }
>  
>      /*
> -     * The address space passed to the dirty tracker is reduced to two ranges:
> -     * one for 32-bit DMA ranges, and another one for 64-bit DMA ranges.
> +     * The address space passed to the dirty tracker is reduced to three ranges:
> +     * one for 32-bit DMA ranges, one for 64-bit DMA ranges and one for the
> +     * PCI 64-bit hole.
> +     *
>       * The underlying reports of dirty will query a sub-interval of each of
>       * these ranges.
>       *
> -     * The purpose of the dual range handling is to handle known cases of big
> -     * holes in the address space, like the x86 AMD 1T hole. The alternative
> -     * would be an IOVATree but that has a much bigger runtime overhead and
> -     * unnecessary complexity.
> +     * The purpose of the three range handling is to handle known cases of big
> +     * holes in the address space, like the x86 AMD 1T hole, and firmware (like
> +     * OVMF) which may relocate the pci-hole64 to the end of the address space.
> +     * The latter would otherwise generate large ranges for tracking, stressing
> +     * the limits of supported hardware. The pci-hole32 will always be below 4G
> +     * (overlapping or not) so it doesn't need special handling and is part of
> +     * the 32-bit range.
> +     *
> +     * The alternative would be an IOVATree but that has a much bigger runtime
> +     * overhead and unnecessary complexity.
>       */
> -    min = (end <= UINT32_MAX) ? &range->min32 : &range->min64;
> -    max = (end <= UINT32_MAX) ? &range->max32 : &range->max64;
> -
> +    if (vfio_section_is_vfio_pci(section, dirty->container) &&
> +        iova >= UINT32_MAX) {
> +        min = &range->minpci64;
> +        max = &range->maxpci64;
> +    } else {
> +        min = (end <= UINT32_MAX) ? &range->min32 : &range->min64;
> +        max = (end <= UINT32_MAX) ? &range->max32 : &range->max64;
> +    }
>      if (*min > iova) {
>          *min = iova;
>      }
> @@ -1461,6 +1502,7 @@ static void vfio_dirty_tracking_init(VFIOContainer *container,
>      memset(&dirty, 0, sizeof(dirty));
>      dirty.ranges.min32 = UINT32_MAX;
>      dirty.ranges.min64 = UINT64_MAX;
> +    dirty.ranges.minpci64 = UINT64_MAX;
>      dirty.listener = vfio_dirty_tracking_listener;
>      dirty.container = container;
>  
> @@ -1531,7 +1573,8 @@ vfio_device_feature_dma_logging_start_create(VFIOContainer *container,
>       * DMA logging uAPI guarantees to support at least a number of ranges that
>       * fits into a single host kernel base page.
>       */
> -    control->num_ranges = !!tracking->max32 + !!tracking->max64;
> +    control->num_ranges = !!tracking->max32 + !!tracking->max64 +
> +        !!tracking->maxpci64;
>      ranges = g_try_new0(struct vfio_device_feature_dma_logging_range,
>                          control->num_ranges);
>      if (!ranges) {
> @@ -1550,11 +1593,17 @@ vfio_device_feature_dma_logging_start_create(VFIOContainer *container,
>      if (tracking->max64) {
>          ranges->iova = tracking->min64;
>          ranges->length = (tracking->max64 - tracking->min64) + 1;
> +        ranges++;
> +    }
> +    if (tracking->maxpci64) {
> +        ranges->iova = tracking->minpci64;
> +        ranges->length = (tracking->maxpci64 - tracking->minpci64) + 1;
>      }
>  
>      trace_vfio_device_dirty_tracking_start(control->num_ranges,
>                                             tracking->min32, tracking->max32,
> -                                           tracking->min64, tracking->max64);
> +                                           tracking->min64, tracking->max64,
> +                                           tracking->minpci64, tracking->maxpci64);
>  
>      return feature;
>  }
> diff --git a/hw/vfio/trace-events b/hw/vfio/trace-events
> index ce61b10827b6..cc7c21365c92 100644
> --- a/hw/vfio/trace-events
> +++ b/hw/vfio/trace-events
> @@ -104,7 +104,7 @@ vfio_known_safe_misalignment(const char *name, uint64_t iova, uint64_t offset_wi
>  vfio_listener_region_add_no_dma_map(const char *name, uint64_t iova, uint64_t size, uint64_t page_size) "Region \"%s\" 0x%"PRIx64" size=0x%"PRIx64" is not aligned to 0x%"PRIx64" and cannot be mapped for DMA"
>  vfio_listener_region_del(uint64_t start, uint64_t end) "region_del 0x%"PRIx64" - 0x%"PRIx64
>  vfio_device_dirty_tracking_update(uint64_t start, uint64_t end, uint64_t min, uint64_t max) "section 0x%"PRIx64" - 0x%"PRIx64" -> update [0x%"PRIx64" - 0x%"PRIx64"]"
> -vfio_device_dirty_tracking_start(int nr_ranges, uint64_t min32, uint64_t max32, uint64_t min64, uint64_t max64) "nr_ranges %d 32:[0x%"PRIx64" - 0x%"PRIx64"], 64:[0x%"PRIx64" - 0x%"PRIx64"]"
> +vfio_device_dirty_tracking_start(int nr_ranges, uint64_t min32, uint64_t max32, uint64_t min64, uint64_t max64, uint64_t minpci, uint64_t maxpci) "nr_ranges %d 32:[0x%"PRIx64" - 0x%"PRIx64"], 64:[0x%"PRIx64" - 0x%"PRIx64"], pci64:[0x%"PRIx64" - 0x%"PRIx64"]"
>  vfio_disconnect_container(int fd) "close container->fd=%d"
>  vfio_put_group(int fd) "close group->fd=%d"
>  vfio_get_device(const char * name, unsigned int flags, unsigned int num_regions, unsigned int num_irqs) "Device %s flags: %u, regions: %u, irqs: %u"