[Qemu-devel] [PATCH] exec: Respect as_tranlsate_internal length clamp

[Qemu-devel] [PATCH] exec: Respect as_tranlsate_internal length clamp

[Peter Crosthwaite]

address_space_translate_internal will clamp the *plen length argument
based on the size of the memory region being queried. The iommu walker
logic in addresss_space_translate was ignoring this by discarding the
post fn call value of *plen. Fix by just always using *plen as the
length argument throughout the fn, removing the len local variable.

This fixes a bootloader bug when a single elf section spans multiple
QEMU memory regions.

Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
---
 exec.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/exec.c b/exec.c
index e97071a..8b922db 100644
--- a/exec.c
+++ b/exec.c
@@ -380,7 +380,6 @@ MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
     IOMMUTLBEntry iotlb;
     MemoryRegionSection *section;
     MemoryRegion *mr;
-    hwaddr len = *plen;
 
     rcu_read_lock();
     for (;;) {
@@ -395,7 +394,7 @@ MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
         iotlb = mr->iommu_ops->translate(mr, addr, is_write);
         addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
                 | (addr & iotlb.addr_mask));
-        len = MIN(len, (addr | iotlb.addr_mask) - addr + 1);
+        *plen = MIN(*plen, (addr | iotlb.addr_mask) - addr + 1);
         if (!(iotlb.perm & (1 << is_write))) {
             mr = &io_mem_unassigned;
             break;
@@ -406,10 +405,9 @@ MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
 
     if (xen_enabled() && memory_access_is_direct(mr, is_write)) {
         hwaddr page = ((addr & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE) - addr;
-        len = MIN(page, len);
+        *plen = MIN(page, *plen);
     }
 
-    *plen = len;
     *xlat = addr;
     rcu_read_unlock();
     return mr;
-- 
2.3.1.2.g90df61e.dirty

Re: [Qemu-devel] [PATCH] exec: Respect as_tranlsate_internal length clamp

[Paolo Bonzini]

On 17/03/2015 06:35, Peter Crosthwaite wrote:
> address_space_translate_internal will clamp the *plen length argument
> based on the size of the memory region being queried. The iommu walker
> logic in addresss_space_translate was ignoring this by discarding the
> post fn call value of *plen. Fix by just always using *plen as the
> length argument throughout the fn, removing the len local variable.
> 
> This fixes a bootloader bug when a single elf section spans multiple
> QEMU memory regions.
> 
> Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
> ---
>  exec.c | 6 ++----
>  1 file changed, 2 insertions(+), 4 deletions(-)
> 
> diff --git a/exec.c b/exec.c
> index e97071a..8b922db 100644
> --- a/exec.c
> +++ b/exec.c
> @@ -380,7 +380,6 @@ MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
>      IOMMUTLBEntry iotlb;
>      MemoryRegionSection *section;
>      MemoryRegion *mr;
> -    hwaddr len = *plen;
>  
>      rcu_read_lock();
>      for (;;) {
> @@ -395,7 +394,7 @@ MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
>          iotlb = mr->iommu_ops->translate(mr, addr, is_write);
>          addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
>                  | (addr & iotlb.addr_mask));
> -        len = MIN(len, (addr | iotlb.addr_mask) - addr + 1);
> +        *plen = MIN(*plen, (addr | iotlb.addr_mask) - addr + 1);
>          if (!(iotlb.perm & (1 << is_write))) {
>              mr = &io_mem_unassigned;
>              break;
> @@ -406,10 +405,9 @@ MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
>  
>      if (xen_enabled() && memory_access_is_direct(mr, is_write)) {
>          hwaddr page = ((addr & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE) - addr;
> -        len = MIN(page, len);
> +        *plen = MIN(page, *plen);
>      }
>  
> -    *plen = len;
>      *xlat = addr;
>      rcu_read_unlock();
>      return mr;
> 

Applied, thanks.

Paolo