On Wed, Jun 17, 2015 at 7:54 AM, Dan Williams <dan.j.williams@xxxxxxxxx> wrote:
> On Wed, Jun 17, 2015 at 4:31 AM, Christoph Hellwig <hch@xxxxxx> wrote:
>> This mess with arch_ methods and an ops vecor is almost unreadable.
>>
>> What's the problem with having something like:
>>
>> pmem_foo()
>> {
>> if (arch_has_pmem) // or sync_pmem
>> arch_pmem_foo();
>> generic_pmem_foo();
>> }
>>
>> This adds a branch at runtime, but that shoudn't really be any slower
>> than an indirect call on architectures that matter.
>
> No doubt it's premature optimization, but it bothered me that we'll
> end up calling cpuid perhaps multiple times every i/o. If it's just a
> readability concern I could wrap it in helpers. Getting it upstream
> is my primary concern at this point so I have no strong attachment to
> the indirect calls if that's all that is preventing an ack.
A cpuid per i/o would be a killer, but the cpufeature code is way
smarter than that.
You want static_cpu_has, though -- it's even faster, since it gets
patched at boot time.
--Andy
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>