> Tell me, please: what tradeoffs are involved in this patch?
> Obviously it works around a pretty fatal problem, but
> what are we giving away?
The change decreases performance a bit. For well-behaved systems the
loss is fifteen instructions: a local APIC read (uncached but supposedly
cheap), a global memory read (a cache line invalidation and fetch), seven
stack accesses (cached for sure), a taken branch and five ALU. With the
version you have I see gcc is actually doing an extra memory read due to
the volatile APIC access presumably -- this is now fixed.
For misdelivered interrupts the overhead is much, much bigger, involving
acquiring a spinlock and multiple (uncached and possibly slow) I/O APIC
accesses. We may lower the overhead by undefining APIC_LOCKUP_DEBUG,
which we should do after a bit of testing. I think we might leave
APIC_MISMATCH_DEBUG intact -- its cost is a single locked instruction
which is negligible IMO.
Note the original version consisted of two instructions only -- a local
APIC write and "ret", sigh...
Maciej
-- + Maciej W. Rozycki, Technical University of Gdansk, Poland + +--------------------------------------------------------------+ + e-mail: macro@ds2.pg.gda.pl, PGP key available +- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/