Merge branch 'perf/core' into perf/uprobes

Merge in latest upstream (and the latest perf development tree),
to prepare for tooling changes, and also to pick up v3.4 MM
changes that the uprobes code needs to take care of.

Signed-off-by: Ingo Molnar <mingo@kernel.org>

5 files changed, 37 insertions, 6 deletions

diff --git a/arch/tile/lib/Makefile b/arch/tile/lib/Makefile
index 0c26086ecbe..985f5985823 100644
--- a/arch/tile/lib/Makefile
+++ b/arch/tile/lib/Makefile
@@ -7,6 +7,7 @@ lib-y = cacheflush.o checksum.o cpumask.o delay.o uaccess.o \
 	strchr_$(BITS).o strlen_$(BITS).o
 
 ifeq ($(CONFIG_TILEGX),y)
+CFLAGS_REMOVE_memcpy_user_64.o = -fno-omit-frame-pointer
 lib-y += memcpy_user_64.o
 else
 lib-y += atomic_32.o atomic_asm_32.o memcpy_tile64.o
diff --git a/arch/tile/lib/cacheflush.c b/arch/tile/lib/cacheflush.c
index 8928aace7a6..db4fb89e12d 100644
--- a/arch/tile/lib/cacheflush.c
+++ b/arch/tile/lib/cacheflush.c
@@ -39,7 +39,21 @@ void finv_buffer_remote(void *buffer, size_t size, int hfh)
 {
 	char *p, *base;
 	size_t step_size, load_count;
+
+	/*
+	 * On TILEPro the striping granularity is a fixed 8KB; on
+	 * TILE-Gx it is configurable, and we rely on the fact that
+	 * the hypervisor always configures maximum striping, so that
+	 * bits 9 and 10 of the PA are part of the stripe function, so
+	 * every 512 bytes we hit a striping boundary.
+	 *
+	 */
+#ifdef __tilegx__
+	const unsigned long STRIPE_WIDTH = 512;
+#else
 	const unsigned long STRIPE_WIDTH = 8192;
+#endif
+
 #ifdef __tilegx__
 	/*
 	 * On TILE-Gx, we must disable the dstream prefetcher before doing
@@ -74,7 +88,7 @@ void finv_buffer_remote(void *buffer, size_t size, int hfh)
 	 * memory, that one load would be sufficient, but since we may
 	 * be, we also need to back up to the last load issued to
 	 * another memory controller, which would be the point where
-	 * we crossed an 8KB boundary (the granularity of striping
+	 * we crossed a "striping" boundary (the granularity of striping
 	 * across memory controllers).  Keep backing up and doing this
 	 * until we are before the beginning of the buffer, or have
 	 * hit all the controllers.
@@ -88,12 +102,22 @@ void finv_buffer_remote(void *buffer, size_t size, int hfh)
 	 * every cache line on a full memory stripe on each
 	 * controller" that we simply do that, to simplify the logic.
 	 *
-	 * FIXME: See bug 9535 for some issues with this code.
+	 * On TILE-Gx the hash-for-home function is much more complex,
+	 * with the upshot being we can't readily guarantee we have
+	 * hit both entries in the 128-entry AMT that were hit by any
+	 * load in the entire range, so we just re-load them all.
+	 * With larger buffers, we may want to consider using a hypervisor
+	 * trap to issue loads directly to each hash-for-home tile for
+	 * each controller (doing it from Linux would trash the TLB).
 	 */
 	if (hfh) {
 		step_size = L2_CACHE_BYTES;
+#ifdef __tilegx__
+		load_count = (size + L2_CACHE_BYTES - 1) / L2_CACHE_BYTES;
+#else
 		load_count = (STRIPE_WIDTH / L2_CACHE_BYTES) *
 			      (1 << CHIP_LOG_NUM_MSHIMS());
+#endif
 	} else {
 		step_size = STRIPE_WIDTH;
 		load_count = (1 << CHIP_LOG_NUM_MSHIMS());
@@ -109,7 +133,7 @@ void finv_buffer_remote(void *buffer, size_t size, int hfh)
 
 	/* Figure out how far back we need to go. */
 	base = p - (step_size * (load_count - 2));
-	if ((long)base < (long)buffer)
+	if ((unsigned long)base < (unsigned long)buffer)
 		base = buffer;
 
 	/*
diff --git a/arch/tile/lib/memcpy_user_64.c b/arch/tile/lib/memcpy_user_64.c
index 4763b3aff1c..37440caa737 100644
--- a/arch/tile/lib/memcpy_user_64.c
+++ b/arch/tile/lib/memcpy_user_64.c
@@ -14,7 +14,13 @@
  * Do memcpy(), but trap and return "n" when a load or store faults.
  *
  * Note: this idiom only works when memcpy() compiles to a leaf function.
- * If "sp" is updated during memcpy, the "jrp lr" will be incorrect.
+ * Here leaf function not only means it does not have calls, but also
+ * requires no stack operations (sp, stack frame pointer) and no
+ * use of callee-saved registers, else "jrp lr" will be incorrect since
+ * unwinding stack frame is bypassed. Since memcpy() is not complex so
+ * these conditions are satisfied here, but we need to be careful when
+ * modifying this file. This is not a clean solution but is the best
+ * one so far.
  *
  * Also note that we are capturing "n" from the containing scope here.
  */
diff --git a/arch/tile/lib/spinlock_32.c b/arch/tile/lib/spinlock_32.c
index cb0999fb64b..b16ac49a968 100644
--- a/arch/tile/lib/spinlock_32.c
+++ b/arch/tile/lib/spinlock_32.c
@@ -144,7 +144,7 @@ void arch_read_unlock(arch_rwlock_t *rwlock)
 	for (;;) {
 		__insn_mtspr(SPR_INTERRUPT_CRITICAL_SECTION, 1);
 		val = __insn_tns((int *)&rwlock->lock);
-		if (likely(val & 1) == 0) {
+		if (likely((val & 1) == 0)) {
 			rwlock->lock = val - (1 << _RD_COUNT_SHIFT);
 			__insn_mtspr(SPR_INTERRUPT_CRITICAL_SECTION, 0);
 			break;
diff --git a/arch/tile/lib/spinlock_common.h b/arch/tile/lib/spinlock_common.h
index c1010980913..6ac37509fac 100644
--- a/arch/tile/lib/spinlock_common.h
+++ b/arch/tile/lib/spinlock_common.h
@@ -60,5 +60,5 @@ static void delay_backoff(int iterations)
 	loops += __insn_crc32_32(stack_pointer, get_cycles_low()) &
 		(loops - 1);
 
-	relax(1 << exponent);
+	relax(loops);
 }