system/physmem: Silence warning from ubsan

[PATCH] system/physmem: Silence warning from ubsan

Posted by Thomas Huth 3 months, 2 weeks ago

From: Thomas Huth <thuth@redhat.com>

When compiling QEMU with --enable-ubsan there is a undefined behavior
warning when running the bios-tables-test for example:

 .../system/physmem.c:3243:13: runtime error: applying non-zero offset 262144 to null pointer
    #0 0x55ac1df5fbc4 in address_space_write_rom_internal .../system/physmem.c:3243:13

The problem is that buf is indeed NULL if the function is e.g. called
with type == FLUSH_CACHE. Add a check to fix the issue.

Signed-off-by: Thomas Huth <thuth@redhat.com>
---
 system/physmem.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/system/physmem.c b/system/physmem.c
index 130c148ffb5..00333ffa7f7 100644
--- a/system/physmem.c
+++ b/system/physmem.c
@@ -3240,8 +3240,10 @@ static inline MemTxResult address_space_write_rom_internal(AddressSpace *as,
             }
         }
         len -= l;
-        buf += l;
         addr += l;
+        if (buf) {
+            buf += l;
+        }
     }
     return MEMTX_OK;
 }
-- 
2.50.1

Re: [PATCH] system/physmem: Silence warning from ubsan

Posted by David Hildenbrand 3 months, 2 weeks ago

On 28.07.25 19:25, Thomas Huth wrote:
> From: Thomas Huth <thuth@redhat.com>
> 
> When compiling QEMU with --enable-ubsan there is a undefined behavior
> warning when running the bios-tables-test for example:
> 
>   .../system/physmem.c:3243:13: runtime error: applying non-zero offset 262144 to null pointer
>      #0 0x55ac1df5fbc4 in address_space_write_rom_internal .../system/physmem.c:3243:13
> 
> The problem is that buf is indeed NULL if the function is e.g. called
> with type == FLUSH_CACHE. Add a check to fix the issue.
> 
> Signed-off-by: Thomas Huth <thuth@redhat.com>

Reviewed-by: David Hildenbrand <david@redhat.com>

-- 
Cheers,

David / dhildenb

Re: [PATCH] system/physmem: Silence warning from ubsan

Posted by Philippe Mathieu-Daudé 3 months, 2 weeks ago

Cc'ing Patrick & Peter for similar patch:
https://lore.kernel.org/qemu-devel/aCUDxEQVACn5CY8f@x1.local/

On 28/7/25 19:25, Thomas Huth wrote:
> From: Thomas Huth <thuth@redhat.com>
> 
> When compiling QEMU with --enable-ubsan there is a undefined behavior
> warning when running the bios-tables-test for example:
> 
>   .../system/physmem.c:3243:13: runtime error: applying non-zero offset 262144 to null pointer
>      #0 0x55ac1df5fbc4 in address_space_write_rom_internal .../system/physmem.c:3243:13
> 
> The problem is that buf is indeed NULL if the function is e.g. called
> with type == FLUSH_CACHE. Add a check to fix the issue.
> 
> Signed-off-by: Thomas Huth <thuth@redhat.com>
> ---
>   system/physmem.c | 4 +++-
>   1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/system/physmem.c b/system/physmem.c
> index 130c148ffb5..00333ffa7f7 100644
> --- a/system/physmem.c
> +++ b/system/physmem.c
> @@ -3240,8 +3240,10 @@ static inline MemTxResult address_space_write_rom_internal(AddressSpace *as,
>               }
>           }
>           len -= l;
> -        buf += l;
>           addr += l;
> +        if (buf) {
> +            buf += l;
> +        }
>       }
>       return MEMTX_OK;
>   }

Re: [PATCH] system/physmem: Silence warning from ubsan

Posted by Peter Xu 3 months, 2 weeks ago

On Mon, Jul 28, 2025 at 08:43:47PM +0200, Philippe Mathieu-Daudé wrote:
> Cc'ing Patrick & Peter for similar patch:
> https://lore.kernel.org/qemu-devel/aCUDxEQVACn5CY8f@x1.local/

This version seems to work.

> 
> On 28/7/25 19:25, Thomas Huth wrote:
> > From: Thomas Huth <thuth@redhat.com>
> > 
> > When compiling QEMU with --enable-ubsan there is a undefined behavior
> > warning when running the bios-tables-test for example:
> > 
> >   .../system/physmem.c:3243:13: runtime error: applying non-zero offset 262144 to null pointer
> >      #0 0x55ac1df5fbc4 in address_space_write_rom_internal .../system/physmem.c:3243:13

I can trigger this by as simple as "./qemu-system-x86_64 -accel kvm".  I
queued it, thanks.

-- 
Peter Xu