summaryrefslogtreecommitdiffstats
path: root/arch/powerpc/kernel/entry_64.S
diff options
context:
space:
mode:
authorAnton Blanchard <anton@samba.org>2010-08-11 03:40:27 +0200
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>2010-09-02 06:07:30 +0200
commitf89451fbd2b9f28f5ff156154989599ec062354b (patch)
tree1722b247079fb80f972449066c9f5c67a5564d4c /arch/powerpc/kernel/entry_64.S
parentpowerpc: Add 64bit csum_and_copy_to_user (diff)
downloadlinux-f89451fbd2b9f28f5ff156154989599ec062354b.tar.xz
linux-f89451fbd2b9f28f5ff156154989599ec062354b.zip
powerpc: Feature nop out reservation clear when stcx checks address
The POWER architecture does not require stcx to check that it is operating on the same address as the larx. This means it is possible for an an exception handler to execute a larx, get a reservation, decide not to do the stcx and then return back with an active reservation. If the interrupted code was in the middle of a larx/stcx sequence the stcx could incorrectly succeed. All recent POWER CPUs check the address before letting the stcx succeed so we can create a CPU feature and nop it out. As Ben suggested, we can only do this in our syscall path because there is a remote possibility some kernel code gets interrupted by an exception that ends up operating on the same cacheline. Thanks to Paul Mackerras and Derek Williams for the idea. To test this I used a very simple null syscall (actually getppid) testcase at http://ozlabs.org/~anton/junkcode/null_syscall.c I tested against 2.6.35-git10 with the following changes against the pseries_defconfig: CONFIG_VIRT_CPU_ACCOUNTING=n CONFIG_AUDIT=n CONFIG_PPC_4K_PAGES=n CONFIG_PPC_64K_PAGES=y CONFIG_FORCE_MAX_ZONEORDER=9 CONFIG_PPC_SUBPAGE_PROT=n CONFIG_FUNCTION_TRACER=n CONFIG_FUNCTION_GRAPH_TRACER=n CONFIG_IRQSOFF_TRACER=n CONFIG_STACK_TRACER=n to remove the overhead of virtual CPU accounting, syscall auditing and the ftrace mcount tracers. 64kB pages were enabled to minimise TLB misses. POWER6: +8.2% POWER7: +7.0% Another suggestion was to use a larx to something in the L1 instead of a stcx. This was almost as fast as removing the larx on POWER6, but only 3.5% faster on POWER7. We can use this to speed up the reservation clear in our exception exit code. Signed-off-by: Anton Blanchard <anton@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Diffstat (limited to 'arch/powerpc/kernel/entry_64.S')
-rw-r--r--arch/powerpc/kernel/entry_64.S22
1 files changed, 22 insertions, 0 deletions
diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S
index 42e9d908914a..4d5fa12ca6e8 100644
--- a/arch/powerpc/kernel/entry_64.S
+++ b/arch/powerpc/kernel/entry_64.S
@@ -202,7 +202,9 @@ syscall_exit:
bge- syscall_error
syscall_error_cont:
ld r7,_NIP(r1)
+BEGIN_FTR_SECTION
stdcx. r0,0,r1 /* to clear the reservation */
+END_FTR_SECTION_IFCLR(CPU_FTR_STCX_CHECKS_ADDRESS)
andi. r6,r8,MSR_PR
ld r4,_LINK(r1)
/*
@@ -419,6 +421,17 @@ END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
sync
#endif /* CONFIG_SMP */
+ /*
+ * If we optimise away the clear of the reservation in system
+ * calls because we know the CPU tracks the address of the
+ * reservation, then we need to clear it here to cover the
+ * case that the kernel context switch path has no larx
+ * instructions.
+ */
+BEGIN_FTR_SECTION
+ ldarx r6,0,r1
+END_FTR_SECTION_IFSET(CPU_FTR_STCX_CHECKS_ADDRESS)
+
addi r6,r4,-THREAD /* Convert THREAD to 'current' */
std r6,PACACURRENT(r13) /* Set new 'current' */
@@ -576,7 +589,16 @@ ALT_FW_FTR_SECTION_END_IFCLR(FW_FEATURE_ISERIES)
andi. r0,r3,MSR_RI
beq- unrecov_restore
+ /*
+ * Clear the reservation. If we know the CPU tracks the address of
+ * the reservation then we can potentially save some cycles and use
+ * a larx. On POWER6 and POWER7 this is significantly faster.
+ */
+BEGIN_FTR_SECTION
stdcx. r0,0,r1 /* to clear the reservation */
+FTR_SECTION_ELSE
+ ldarx r4,0,r1
+ALT_FTR_SECTION_END_IFCLR(CPU_FTR_STCX_CHECKS_ADDRESS)
/*
* Clear RI before restoring r13. If we are returning to