hwpmc module does not work on armv6/armv7

[Arnaud YSMAL]

----- Le 6 Juin 19, à 17:36, Ian Lepore ian at freebsd.org a écrit :
> On Thu, 2019-06-06 at 17:32 +0200, Arnaud YSMAL wrote:
>> ----- Le 6 Juin 19, à 17:22, Ian Lepore ian at freebsd.org a écrit :
>> 
>> > On Thu, 2019-06-06 at 15:56 +0200, Arnaud YSMAL wrote:
>> > > Hi,
>> > > 
>> > > On an armv6/armv7 box, trying to load the hwpmc module lead to an
>> > > error message stating that the pmc_armv7_finalize symbol is
>> > > undefined.
>> > > 
>> > > This can be fixed by adding this in sys/modules/hwpmc/Makefile :
>> > > 
>> > > .if ${MACHINE_ARCH:Marmv[67]*} != ""
>> > > SRCS+= hwpmc_armv7.c
>> > > .endif
>> > > 
>> > > Does anyone see any reason for not having this?
>> > > 
>> > > 
>> > 
>> > No reason, that was just a bug.  I've committed your fix as
>> > r348741,
>> > thanks!
>> 
>> Thank you.
>> 
>> Sorry for not spotting it earlier but, could you please fix the armv7
>> invariant build with this patch?
>> 
>> 
>> diff --git a/sys/dev/hwpmc/hwpmc_armv7.c
>> b/sys/dev/hwpmc/hwpmc_armv7.c
>> index 0a601145f8d..d89b5ce302f 100644
>> --- a/sys/dev/hwpmc/hwpmc_armv7.c
>> +++ b/sys/dev/hwpmc/hwpmc_armv7.c
>> @@ -315,11 +315,12 @@ armv7_intr(struct trapframe *tf)
>>         int error;
>>         int reg, cpu;
>>  
>> +       cpu = curcpu;
>> +
>>         KASSERT(cpu >= 0 && cpu < pmc_cpu_max(),
>>             ("[armv7,%d] CPU %d out of range", __LINE__, cpu));
>>  
>>         retval = 0;
>> -       cpu = curcpu;
>>         pc = armv7_pcpu[cpu];
>>  
>>         for (ri = 0; ri < armv7_npmcs; ri++) {
>> 
>> 
> 
> Already did, right before I committed your change. :)

Thanks. :)
 
> I've just heard on irc that there may be other trouble lurking in armv*
> hwpmc, some kind of fallout from the changes in r334827.  Manu is
> looking into it, and will probably post something on the arm@ list when
> he gets it figured out.

Thank you for the information.

> 
> -- Ian