How long am I allowed to disable preemption? (was How to make use of SPE instructions?)

Von: Scott Wood [scottwood@freescale.com]
Gesendet: Dienstag, 20. Januar 2015 08:38
An: Markus Stockhausen
Cc: Michael Ellerman; linuxppc-dev@lists.ozlabs.org
Betreff: Re: AW: How to make use of SPE instructions?
...

quoted

I did some tests with the tcrypt module and I get a hashing speed of
~ 46MByte/s for 2K data chunks. Stock module gives 29MByte/s. In
other words ~22,000 hashes per second. Overhead of the tcrypt data
feeder of around 10% included. That are worst case 46us per hash and
therefore 46us inside a non preemptive task.

...
It's OK if the worst case is really 46 us, but if you can find a way to
break it up a bit without affecting throughput too much, I'd do so.

Hi Scott (and all),

with your answer I'm a bit confused about disabling preemption inside the 
kernel right now. I understand that the interval between preempt_disable()
and preempt_enable() should not take too long. But when I look at other
kernel hash crypto modules I see that nobody really cares about that. Take
sha256_ssse3_update() in sha256_ssse3_glue.c for example. It runs

  kernel_fpu_begin();
  res = __sha256_ssse3_update(desc, data, len, partial);
  kernel_fpu_end();

The kernel_fpu_xxx() enables/disables preemption like I need to do. 
Nevertheless parameter len can be any number of bytes. Just take 
reasonable parameters of 256K of input data, a 3GHz core and 13 
cycles/byte of SHA256 throughput. That will be a 1ms timeframe.

Can I rely on that implementation, or do I have to take special care 
because I'm only programming for a single core CPU? 

With your advice I would place a enable/disable preemption call after 
1K of processed data. But wil that be sufficient if I only reeanble it
for a short timeframe like this:

  do {
    disable_preemption()
    ... calc hashes for 1K of data with 16.000 CPU cycles (or 20us) ...
    enable_preemption()
  while (dataleft>0);

Markus

On Tue, 2015-01-20 at 14:53 +0000, Markus Stockhausen wrote:

quoted

Von: Scott Wood [scottwood@freescale.com]
Gesendet: Dienstag, 20. Januar 2015 08:38
An: Markus Stockhausen
Cc: Michael Ellerman; linuxppc-dev@lists.ozlabs.org
Betreff: Re: AW: How to make use of SPE instructions?
...

quoted

I did some tests with the tcrypt module and I get a hashing speed of
~ 46MByte/s for 2K data chunks. Stock module gives 29MByte/s. In
other words ~22,000 hashes per second. Overhead of the tcrypt data
feeder of around 10% included. That are worst case 46us per hash and
therefore 46us inside a non preemptive task.

...
It's OK if the worst case is really 46 us, but if you can find a way to
break it up a bit without affecting throughput too much, I'd do so.

Hi Scott (and all),

with your answer I'm a bit confused about disabling preemption inside the 
kernel right now. I understand that the interval between preempt_disable()
and preempt_enable() should not take too long. But when I look at other
kernel hash crypto modules I see that nobody really cares about that.

"Some people don't care about that" doesn't imply "nobody cares about
that".

 Take sha256_ssse3_update() in sha256_ssse3_glue.c for example. It runs

  kernel_fpu_begin();
  res = __sha256_ssse3_update(desc, data, len, partial);
  kernel_fpu_end();

The kernel_fpu_xxx() enables/disables preemption like I need to do. 

Nevertheless parameter len can be any number of bytes. Just take 
reasonable parameters of 256K of input data, a 3GHz core and 13 
cycles/byte of SHA256 throughput. That will be a 1ms timeframe.

That's bad.  Don't create more bad stuff. :-)

Can I rely on that implementation, or do I have to take special care 
because I'm only programming for a single core CPU? 

Not sure what you mean here.

With your advice I would place a enable/disable preemption call after 
1K of processed data. But wil that be sufficient if I only reeanble it
for a short timeframe like this:

  do {
    disable_preemption()
    ... calc hashes for 1K of data with 16.000 CPU cycles (or 20us) ...
    enable_preemption()
  while (dataleft>0);

Yes, it's sufficient.  When you enable preemption it will check to see
whether there is a pending reschedule.

-Scott