Re: update i386 spinlock for hyperthreading - Mailing list pgsql-patches

Tom Lane wrote:

>Manfred Spraul <manfred@colorfullife.com> writes:
>
>
>>Intel recommends to add a special pause instruction into spinlock busy
>>loops. It's necessary for hyperthreading - without it, the cpu can't
>>figure out that a logical thread does no useful work and incorrectly
>>awards lots of execution resources to that thread. Additionally, it's
>>supposed to reduce the time the cpu needs to recover from the
>>(mispredicted) branch after the spinlock was obtained.
>>
>>
>
>Don't you have to put it in a specific place in the loop to make that
>work?  If not, why not?  I doubt that rep;nop is magic enough to
>recognize the loop that will be generated from s_lock()'s code.
>
>
Rep;nop is just a short delay - that's all. It means that the cpu
pipelines have a chance to drain, and that the other thread gets enough
cpu resources. Below is the full instruction documentation, from the
latest ia32 doc set from Intel:
<<<
Improves the performance of spin-wait loops. When executing a  spin-wait
loop,  a Pentium 4 or Intel Xeon processor suffers a severe performance
penalty when exiting the loop because it detects a possible memory order
violation. The PAUSE instruction provides a hint to the processor that
the code sequence is a spin-wait loop. The processor uses this hint to
avoid the memory order violation in most situations, which greatly
improves processor performance. For this reason, it is recommended that
a PAUSE instruction be placed in all spin-wait loops. An additional
function of the PAUSE instruction is to reduce the power consumed by a
Pentium 4 processor while executing a spin loop. The Pentium 4 processor
can execute a spin-wait loop extremely quickly, causing the processor to
consume a lot of power while it waits for the resource it is spinning on
to become available. Inserting a pause instruction in a spin-wait loop
greatly reduces the processor s power consumption. This instruction was
introduced in the Pentium 4 processors, but is backward compatible with
all IA-32 processors. In earlier IA-32 processors, the PAUSE instruction
operates like a NOP instruction. The Pentium 4 and Intel Xeon processors
implement the PAUSE instruction as a pre-defined delay. The delay is
finite and can be zero for some processors. This instruction does not
change the architectural state of the processor (that is, it performs
essentially a delaying noop operation).
<<<

I think a separate function is better than adding it into TAS: if it's
part of tas, then it would automatically be included by every
SpinLockAcquire call - unnecessary .text bloat. Additionally, there
might be other busy loops, in addition to TAS, that could use a delay
function.

I'll post a new patch that doesn't rely on __inline__ in the i386
independant part.

--
    Manfred