[PATCH v2] powerpc/booke64: wrap tlb lock and search in htw miss with FTR_SMT

Scott Wood scottwood at freescale.com
Sat May 31 08:45:46 EST 2014

From: Laurentiu Tudor <Laurentiu.Tudor at freescale.com>

Virtualized environments may expose a e6500 dual-threaded core
as two single-threaded e6500 cores. Take advantage of this
and get rid of the tlb lock and the trap-causing tlbsx in
the htw miss handler by guarding with CPU_FTR_SMT, as it's
already being done in the bolted tlb1 miss handler.

As seen in the results below, measurements done with lmbench
random memory access latency test running under Freescale's
Embedded Hypervisor, there is a ~34% improvement.

Memory latencies in nanoseconds - smaller is better
    (WARNING - may not be correct, check graphs)
Host       Mhz   L1 $   L2 $    Main mem    Rand mem
---------  ---   ----   ----    --------    --------
smt       1665 1.8020   13.2    83.0         1149.7
nosmt     1665 1.8020   13.2    83.0          758.1

Signed-off-by: Laurentiu Tudor <Laurentiu.Tudor at freescale.com>
Cc: Scott Wood <scottwood at freescale.com>
[scottwood at freescale.com: commit message tweak]
Signed-off-by: Scott Wood <scottwood at freescale.com>
 - s/expose/may expose/ in commit message
 - rebased onto my patch queue to resolve conflict
 - resent since the original didn't make it to the list archives
   or patchwork.

 arch/powerpc/mm/tlb_low_64e.S | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/arch/powerpc/mm/tlb_low_64e.S b/arch/powerpc/mm/tlb_low_64e.S
index 131f1f4..57c4d66 100644
--- a/arch/powerpc/mm/tlb_low_64e.S
+++ b/arch/powerpc/mm/tlb_low_64e.S
@@ -299,6 +299,7 @@ itlb_miss_fault_bolted:
  * r10 = crap (free to use)
 	 * Search if we already have an indirect entry for that virtual
 	 * address, and if we do, bail out.
@@ -333,6 +334,7 @@ tlb_miss_common_e6500:
 	andis.	r10,r10,MAS1_VALID at h
 	bne	tlb_miss_done_e6500
 	/* Now, we need to walk the page tables. First check if we are in
 	 * range.
@@ -393,11 +395,13 @@ tlb_miss_common_e6500:
 	.macro	tlb_unlock_e6500
 	beq	cr1,1f		/* no unlock if lock was recursively grabbed */
 	li	r15,0
 	stb	r15,0(r11)

More information about the Linuxppc-dev mailing list