Re: [Qemu-devel] [RFC PATCH] migration: do not use atomic__nocheck() functions directly

[Philippe Mathieu-Daudé]

On 01/19/2018 03:01 PM, Dr. David Alan Gilbert wrote:
> * Philippe Mathieu-Daudé (address@hidden) wrote:
>> (incorrectly use in 3be98be4e9f)
>>
>> Signed-off-by: Philippe Mathieu-Daudé <address@hidden>
> 
> I'm a bit confused; isnt the only difference between the nocheck
> versions that it'll fail at compile time instead of link?

You are right, this isn't the right approach.

Peter gave a clearer explanation here:
https://lists.gnu.org/archive/html/qemu-devel/2018-01/msg02103.html

'''
because atomic operations on types larger than the host pointer
size are not portable to all architectures. This should probably
use the atomic_cmpxchg(), not __nocheck, because the latter
bypasses the build time assert for this problem and turns a
"fails on any 32-bit host" into "fails obscurely on some
architectures only".
'''

Maybe we should remove the __nocheck() functions and inline them in the
'checked' functions?
There is no performance cost doing this, right?

> 
> Dave
> 
>> ---
>> currently on ppc32 the linking fails:
>>
>>   CC      migration/postcopy-ram.o
>> ...
>>   LINK    microblaze-softmmu/qemu-system-microblaze
>> ../migration/postcopy-ram.o: In function `mark_postcopy_blocktime_end':
>> migration/postcopy-ram.c:717: undefined reference to `__atomic_fetch_add_8'
>> migration/postcopy-ram.c:738: undefined reference to `__atomic_fetch_add_8'
>> ../migration/postcopy-ram.o: In function `mark_postcopy_blocktime_begin':
>> migration/postcopy-ram.c:651: undefined reference to `__atomic_exchange_8'
>> migration/postcopy-ram.c:652: undefined reference to `__atomic_exchange_8'
>> migration/postcopy-ram.c:661: undefined reference to `__atomic_exchange_8'
>> collect2: error: ld returned 1 exit status
>> Makefile:193: recipe for target 'qemu-system-microblaze' failed
>> make[1]: *** [qemu-system-microblaze] Error 1
>>
>> with this patch the compilation fails:
>>
>>   CC      migration/postcopy-ram.o
>> In file included from include/qemu/osdep.h:36:0,
>>                  from migration/postcopy-ram.c:19:
>> migration/postcopy-ram.c: In function 'mark_postcopy_blocktime_begin':
>> include/qemu/compiler.h:86:30: error: static assertion failed: "not 
>> expecting: sizeof(*&dc->last_begin) > ATOMIC_REG_SIZE"
>>  #define QEMU_BUILD_BUG_ON(x) _Static_assert(!(x), "not expecting: " #x)
>>                               ^
>> include/qemu/atomic.h:183:5: note: in expansion of macro 'QEMU_BUILD_BUG_ON'
>>      QEMU_BUILD_BUG_ON(sizeof(*ptr) > ATOMIC_REG_SIZE);      \
>>      ^
>> migration/postcopy-ram.c:651:5: note: in expansion of macro 'atomic_xchg'
>>      atomic_xchg(&dc->last_begin, now_ms);
>>      ^
>>
>>  migration/postcopy-ram.c | 12 ++++++------
>>  1 file changed, 6 insertions(+), 6 deletions(-)
>>
>> diff --git a/migration/postcopy-ram.c b/migration/postcopy-ram.c
>> index 7814da5b4b..6ecc1aa820 100644
>> --- a/migration/postcopy-ram.c
>> +++ b/migration/postcopy-ram.c
>> @@ -648,17 +648,17 @@ static void mark_postcopy_blocktime_begin(uintptr_t 
>> addr, uint32_t ptid,
>>          atomic_inc(&dc->smp_cpus_down);
>>      }
>>  
>> -    atomic_xchg__nocheck(&dc->last_begin, now_ms);
>> -    atomic_xchg__nocheck(&dc->page_fault_vcpu_time[cpu], now_ms);
>> -    atomic_xchg__nocheck(&dc->vcpu_addr[cpu], addr);
>> +    atomic_xchg(&dc->last_begin, now_ms);
>> +    atomic_xchg(&dc->page_fault_vcpu_time[cpu], now_ms);
>> +    atomic_xchg(&dc->vcpu_addr[cpu], addr);
>>  
>>      /* check it here, not at the begining of the function,
>>       * due to, check could accur early than bitmap_set in
>>       * qemu_ufd_copy_ioctl */
>>      already_received = ramblock_recv_bitmap_test(rb, (void *)addr);
>>      if (already_received) {
>> -        atomic_xchg__nocheck(&dc->vcpu_addr[cpu], 0);
>> -        atomic_xchg__nocheck(&dc->page_fault_vcpu_time[cpu], 0);
>> +        atomic_xchg(&dc->vcpu_addr[cpu], 0);
>> +        atomic_xchg(&dc->page_fault_vcpu_time[cpu], 0);
>>          atomic_dec(&dc->smp_cpus_down);
>>      }
>>      trace_mark_postcopy_blocktime_begin(addr, dc, 
>> dc->page_fault_vcpu_time[cpu],
>> @@ -719,7 +719,7 @@ static void mark_postcopy_blocktime_end(uintptr_t addr)
>>              read_vcpu_time == 0) {
>>              continue;
>>          }
>> -        atomic_xchg__nocheck(&dc->vcpu_addr[i], 0);
>> +        atomic_xchg(&dc->vcpu_addr[i], 0);
>>          vcpu_blocktime = now_ms - read_vcpu_time;
>>          affected_cpu += 1;
>>          /* we need to know is that mark_postcopy_end was due to
>> -- 
>> 2.15.1
>>
> --
> Dr. David Alan Gilbert / address@hidden / Manchester, UK
>