Re: intel_iommu unit test is also failing

[Thomas Huth <thuth@redhat.com>]

On 05/05/2026 14.33, Michael S. Tsirkin wrote:
> On Tue, May 05, 2026 at 01:38:26PM +0200, Thomas Huth wrote:
>> On 05/05/2026 12.53, Michael S. Tsirkin wrote:
>>> On Tue, May 05, 2026 at 12:34:41PM +0200, Thomas Huth wrote:
>>>> On 05/05/2026 12.23, Michael S. Tsirkin wrote:
>>>>> On Tue, May 05, 2026 at 11:45:17AM +0200, Thomas Huth wrote:
>>>>>> On 05/05/2026 11.27, Clément MATHIEU--DRIF wrote:
>>>>>>> I had a bit more time to hook into qemu to check the root cause.
>>>>>>>
>>>>>>> It seems that testb issues a single byte read (out of the valid size range), as we can see on the following breakpoint:
>>>>>>>
>>>>>>> ```
>>>>>>> Thread 6 "CPU 0/TCG" hit Breakpoint 2, memory_region_dispatch_read (mr=0x55d72883cb30, addr=152, pval=0x7f62d25f4590, op=MO_BSWAP, attrs=...) at ../system/memory.c:1473
>>>>>>> 1473       unsigned size = memop_size(op);
>>>>>>> (gdb) n
>>>>>>> 1474       MemTxResult r;
>>>>>>> (gdb) p size
>>>>>>> $1 = 1
>>>>>>> (gdb)
>>>>>>> ```
>>>>>>
>>>>>> Ouch! That's an excellent finding, Clément ... so GCC 16 is "smart" enough
>>>>>> to see that we only want to test the lowest bit here, so it optimizes the
>>>>>> code to access only one byte of memory instead of 4 bytes... which would be
>>>>>> ok for normal memory, but not for an MMIO register :-/
>>>>>>
>>>>>> Ugly work-around, to force GCC to read 32 bits:
>>>>>>
>>>>>> diff --git a/lib/asm-generic/io.h b/lib/asm-generic/io.h
>>>>>> --- a/lib/asm-generic/io.h
>>>>>> +++ b/lib/asm-generic/io.h
>>>>>> @@ -38,7 +38,9 @@ static inline u16 __raw_readw(const volatile void *addr)
>>>>>>     #ifndef __raw_readl
>>>>>>     static inline u32 __raw_readl(const volatile void *addr)
>>>>>>     {
>>>>>> -       return *(const volatile u32 *)addr;
>>>>>> +       u32 val = *(const volatile u32 *)addr;
>>>>>> +       asm volatile ("\n" : : "r"(addr));
>>>>>> +       return val;
>>>>>>     }
>>>>>>     #endif
>>>>>>
>>>>>> ... but I wonder whether this should rather be treated as a bug in GCC
>>>>>> instead, since it should IMHO really not change the access size for a
>>>>>> volatile memory access?
>>>>>>
>>>>>>     Thomas
>>>>>
>>>>> Wouldn't this break linux generally?
>>>>>
>>>>> #ifndef __READ_ONCE
>>>>> #define __READ_ONCE(x)  (*(const volatile __unqual_scalar_typeof(x) *)&(x))
>>>>> #endif
>>>>
>>>> I asked myself the very same question, but after googling for "GCC 16 linux
>>>> kernel" issues, I did not find anything related... there is likely something
>>>> specific to kvm-unit-tests in here...
>>>>
>>>>    Thomas
>>>
>>>
>>> This seems to be pertinent:
>>> https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html
>>>
>>> -ffuse-ops-with-volatile-access
>>> Allow limited optimization of operations with volatile memory access when doing so does not change the semantics outlined in See When is a Volatile Object Accessed?.
>>>
>>> The default is -ffuse-ops-with-volatile-access
>>>
>>> implemented here:
>>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122343
>>>
>>> Try disabling? -fno-fuse-ops-with-volatile-access
>>
>> Thanks, this seems to fix the issue, indeed!
>>
>> Would you like to send a patch for it?
> 
> So then it's this bug apparently:
> 
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125180

Ah, thanks, that looks like the bug, indeed!

> will likely be in 16.2?
> 
> Maybe just wait?

I guess we could drop the failing test from the CI for the time being...

  Thomas