Re: [PATCH] Enable non page boundary BAR device assignment

["Michael S. Tsirkin" <mst@redhat.com>]

On Tue, Dec 15, 2009 at 07:16:13PM +0100, Alexander Graf wrote:
> Michael S. Tsirkin wrote:
> > On Fri, Dec 11, 2009 at 12:06:25AM +0100, Alexander Graf wrote:
> >   
> >> While trying to get device passthrough working with an emulex hba, kvm
> >> refused to pass it through because it has a BAR of 256 bytes:
> >>
> >>         Region 0: Memory at d2100000 (64-bit, non-prefetchable) [size=4K]
> >>         Region 2: Memory at d2101000 (64-bit, non-prefetchable) [size=256]
> >>         Region 4: I/O ports at b100 [size=256]
> >>
> >> Since the page boundary is an arbitrary optimization to allow 1:1 mapping of
> >> physical to virtual addresses, we can still take the old MMIO callback route.
> >>
> >> So let's add a second code path that allows for size & 0xFFF != 0 sized regions
> >> by looping it through userspace.
> >>
> >> I verified that it works by passing through an e1000 with this additional patch
> >> applied and the card acted the same way it did without this patch:
> >>
> >>              map_func = assigned_dev_iomem_map;
> >> -            if (cur_region->size & 0xFFF) {
> >> +            if (i != PCI_ROM_SLOT){
> >>                  fprintf(stderr, "PCI region %d at address 0x%llx "
> >>     
> >
> > The patch is pretty clean. However, I think I see a bug below.
> > Did you also try a device with sub-page BAR?
> > If yes, did the bar get non page aligned value?
> >
> >   
> >> Signed-off-by: Alexander Graf <agraf@suse.de>
> >>
> >> ---
> >>
> >> v1 -> v2:
> >>
> >>   - don't use map_func function pointer
> >>   - use the same code for mmap on fast and slow path
> >> ---
> >>  hw/device-assignment.c |  123 +++++++++++++++++++++++++++++++++++++++++++++---
> >>  1 files changed, 116 insertions(+), 7 deletions(-)
> >>
> >> diff --git a/hw/device-assignment.c b/hw/device-assignment.c
> >> index 13a86bb..5cee929 100644
> >> --- a/hw/device-assignment.c
> >> +++ b/hw/device-assignment.c
> >> @@ -148,6 +148,105 @@ static uint32_t assigned_dev_ioport_readl(void *opaque, uint32_t addr)
> >>      return value;
> >>  }
> >>  
> >> +static uint32_t slow_bar_readb(void *opaque, target_phys_addr_t addr)
> >> +{
> >> +    AssignedDevRegion *d = opaque;
> >> +    uint8_t *in = (uint8_t*)(d->u.r_virtbase + addr);
> >>     
> >
> > this is a void* pointer, no need to cast
> >
> >   
> >> +    uint32_t r = -1;
> >>     
> >
> > don't initialize r here as you override 1 line below.
> >   
> >> +
> >> +    r = *in;
> >> +    DEBUG("slow_bar_readl addr=0x" TARGET_FMT_plx " val=0x%08x\n", addr, r);
> >> +
> >> +    return r;
> >> +}
> >> +
> >> +static uint32_t slow_bar_readw(void *opaque, target_phys_addr_t addr)
> >> +{
> >> +    AssignedDevRegion *d = opaque;
> >> +    uint16_t *in = (uint16_t*)(d->u.r_virtbase + addr);
> >> +    uint32_t r = -1;
> >> +
> >> +    r = *in;
> >> +    DEBUG("slow_bar_readl addr=0x" TARGET_FMT_plx " val=0x%08x\n", addr, r);
> >> +
> >> +    return r;
> >>     
> >
> > same as above
> >
> >   
> >> +}
> >> +
> >> +static uint32_t slow_bar_readl(void *opaque, target_phys_addr_t addr)
> >> +{
> >> +    AssignedDevRegion *d = opaque;
> >> +    uint32_t *in = (uint32_t*)(d->u.r_virtbase + addr);
> >> +    uint32_t r = -1;
> >> +
> >> +    r = *in;
> >> +    DEBUG("slow_bar_readl addr=0x" TARGET_FMT_plx " val=0x%08x\n", addr, r);
> >> +
> >> +    return r;
> >>     
> >
> > same as above
> >
> >   
> >> +}
> >> +
> >> +static void slow_bar_writeb(void *opaque, target_phys_addr_t addr, uint32_t val)
> >> +{
> >> +    AssignedDevRegion *d = opaque;
> >> +    uint8_t *out = (uint8_t*)(d->u.r_virtbase + addr);
> >>     
> >
> > no need for cast as r_virtbase is a void pointer.
> >
> >   
> >> +
> >> +    DEBUG("slow_bar_writeb addr=0x" TARGET_FMT_plx " val=0x%02x\n", addr, val);
> >> +    *out = val;
> >> +}
> >> +
> >> +static void slow_bar_writew(void *opaque, target_phys_addr_t addr, uint32_t val)
> >> +{
> >> +    AssignedDevRegion *d = opaque;
> >> +    uint16_t *out = (uint16_t*)(d->u.r_virtbase + addr);
> >> +
> >> +    DEBUG("slow_bar_writew addr=0x" TARGET_FMT_plx " val=0x%04x\n", addr, val);
> >> +    *out = val;
> >> +}
> >> +
> >> +static void slow_bar_writel(void *opaque, target_phys_addr_t addr, uint32_t val)
> >> +{
> >> +    AssignedDevRegion *d = opaque;
> >> +    uint32_t *out = (uint32_t*)(d->u.r_virtbase + addr);
> >>     
> >
> > same as above
> >
> >   
> >> +
> >> +    DEBUG("slow_bar_writel addr=0x" TARGET_FMT_plx " val=0x%08x\n", addr, val);
> >> +    *out = val;
> >> +}
> >> +
> >> +static CPUWriteMemoryFunc * const slow_bar_write[] = {
> >> +    &slow_bar_writeb,
> >> +    &slow_bar_writew,
> >> +    &slow_bar_writel
> >> +};
> >> +
> >> +static CPUReadMemoryFunc * const slow_bar_read[] = {
> >> +    &slow_bar_readb,
> >> +    &slow_bar_readw,
> >> +    &slow_bar_readl
> >> +};
> >> +
> >> +static void assigned_dev_iomem_map_slow(PCIDevice *pci_dev, int region_num,
> >> +                                        pcibus_t e_phys, pcibus_t e_size,
> >> +                                        int type)
> >> +{
> >> +    AssignedDevice *r_dev = container_of(pci_dev, AssignedDevice, dev);
> >> +    AssignedDevRegion *region = &r_dev->v_addrs[region_num];
> >> +    PCIRegion *real_region = &r_dev->real_device.regions[region_num];
> >> +    int m;
> >> +
> >> +    DEBUG("slow map\n");
> >> +    m = cpu_register_io_memory(slow_bar_read, slow_bar_write, region);
> >> +    cpu_register_physical_memory(e_phys, e_size, m);
> >> +
> >> +    /* MSI-X MMIO page */
> >>     
> >
> > some code duplication  ... use a common function?
> >   
> 
> I'm not sure. It's pretty little duplication. If you feel like this
> should be merged, feel free to make a cleanup patch afterwards :-).
> 
> >   
> >> +    if ((e_size > 0) &&
> >> +        real_region->base_addr <= r_dev->msix_table_addr &&
> >> +        real_region->base_addr + real_region->size >= r_dev->msix_table_addr) {
> >> +        int offset = r_dev->msix_table_addr - real_region->base_addr;
> >> +
> >> +        cpu_register_physical_memory(e_phys + offset,
> >> +                TARGET_PAGE_SIZE, r_dev->mmio_index);
> >>     
> >
> > How does this work? Does the last registered callback win
> > or are both called for MSIX page?
> >   
> 
> In qemu memory allocations the last one always wins.
> 
> >   
> >> +    }
> >> +}
> >> +
> >>  static void assigned_dev_iomem_map(PCIDevice *pci_dev, int region_num,
> >>                                     pcibus_t e_phys, pcibus_t e_size, int type)
> >>  {
> >> @@ -381,15 +480,22 @@ static int assigned_dev_register_regions(PCIRegion *io_regions,
> >>  
> >>          /* handle memory io regions */
> >>          if (cur_region->type & IORESOURCE_MEM) {
> >> +            int slow_map = 0;
> >>              int t = cur_region->type & IORESOURCE_PREFETCH
> >>                  ? PCI_BASE_ADDRESS_MEM_PREFETCH
> >>                  : PCI_BASE_ADDRESS_SPACE_MEMORY;
> >> +
> >>              if (cur_region->size & 0xFFF) {
> >> -                fprintf(stderr, "Unable to assign device: PCI region %d "
> >> -                        "at address 0x%llx has size 0x%x, "
> >> -                        " which is not a multiple of 4K\n",
> >> +                fprintf(stderr, "PCI region %d at address 0x%llx "
> >> +                        "has size 0x%x, which is not a multiple of 4K. "
> >> +                        "Using slow map\n",
> >>     
> >
> > I still think "Using slow map" tells the user nothing.
> > How about "Disabling direct guest access, device will be slow"?
> >   
> 
> Changed.
> 
> >   
> >>                          i, (unsigned long long)cur_region->base_addr,
> >>                          cur_region->size);
> >> +                slow_map = 1;
> >> +            }
> >> +
> >> +            if (slow_map && (i == PCI_ROM_SLOT)) {
> >> +                fprintf(stderr, "ROM not aligned - can't continue\n");
> >>                  return -1;
> >>              }
> >>  
> >> @@ -405,7 +511,7 @@ static int assigned_dev_register_regions(PCIRegion *io_regions,
> >>              } else {
> >>                  pci_dev->v_addrs[i].u.r_virtbase =
> >>                      mmap(NULL,
> >> -                         (cur_region->size + 0xFFF) & 0xFFFFF000,
> >> +                         cur_region->size,
> >>                           PROT_WRITE | PROT_READ, MAP_SHARED,
> >>                           cur_region->resource_fd, (off_t) 0);
> >>              }
> >> @@ -429,12 +535,15 @@ static int assigned_dev_register_regions(PCIRegion *io_regions,
> >>              pci_dev->v_addrs[i].e_size = 0;
> >>  
> >>              /* add offset */
> >> -            pci_dev->v_addrs[i].u.r_virtbase +=
> >> -                (cur_region->base_addr & 0xFFF);
> >> +            if (!slow_map) {
> >> +                pci_dev->v_addrs[i].u.r_virtbase +=
> >> +                    (cur_region->base_addr & 0xFFF);
> >> +            }
> >>     
> >
> > This looks wrong. I think mmap returns a page aligned address,
> > that's why we did this offset math. No?
> > And if not, there was no reason for this code in the first place.
> >   
> 
> Yeah, I guess you're right.
> 
> Alex

I guess this means you'll have to find a device with a sub-page BAR
to test this on, instead of hacking driver for a device with full
page BAR. Did anyone ever try doing passthrough on an emulated
device in nested virt?

-- 
MST