On 2020-09-27 3:45 p.m., Helge Deller wrote:
> On 9/27/20 4:17 PM, John David Anglin wrote:
>> With this patch collection, I haven't seen any stalls on my rp3440 and c8000 running v5.8.x. Further,
>> I haven't had any signing failures on the buildd machine. The rp3440 (mx3210) has also been successful in
>> building several packages that failed multiple time on the c8000 buildd.
>>
>> I think this is due to the three changes at the bottom from Linus.
> Weren't thos added already upstream (or variants of them) ?
Thought so but nothing has been committed to v5.8.x that conflicts with the changes. I also searched the git log
for similar commits but I didn't find anything. So, I'm puzzled as to what happened to these changes. This is the
main reason I threw out my patch list for discussion.
The following packages built successfully on mx3210 after falling on c8000: tigervnc, postgresql-13, verse, bali-phy
and libbpf. I also tried mariadb-10.5 but it had the same testsuite fail as before.
>
>> diff --git a/arch/parisc/include/asm/barrier.h b/arch/parisc/include/asm/barrier.h
>> index 640d46edf32e..c705decf2bed 100644
>> --- a/arch/parisc/include/asm/barrier.h
>> +++ b/arch/parisc/include/asm/barrier.h
>> @@ -2,11 +2,15 @@
>> /* The synchronize caches instruction executes as a nop on systems in
>> which all memory references are performed in order. */
>> -#define synchronize_caches() __asm__ __volatile__ ("sync" : : : "memory")
>> +#define synchronize_caches() asm volatile("sync" \
>> + ALTERNATIVE(ALT_COND_NO_SMP, INSN_NOP) \
>> + : : : "memory")
> I wonder if this has any impact.
> If it's an UP system, won't this "sync" just be a "nop" anyway?
I believe we want this when SMP kernel is run on a UP machine or a SMP machine running with one CPU.
> Did you ran some performance tests, just to check?
No.
>
> Would you mind to send separate patches for each logical change?
>
I'll try to get there but I'm currently focusing on toolchain issues.
Regards,
Dave
--
John David Anglin dave.anglin@xxxxxxxx
