3 files changed, 45 insertions, 30 deletions

diff --git a/arch/tile/lib/memchr_32.c b/arch/tile/lib/memchr_32.c
index 6235283b4859..cc3d9badf030 100644
--- a/arch/tile/lib/memchr_32.c
+++ b/arch/tile/lib/memchr_32.c
@@ -18,12 +18,24 @@
 
 void *memchr(const void *s, int c, size_t n)
 {
+	const uint32_t *last_word_ptr;
+	const uint32_t *p;
+	const char *last_byte_ptr;
+	uintptr_t s_int;
+	uint32_t goal, before_mask, v, bits;
+	char *ret;
+
+	if (__builtin_expect(n == 0, 0)) {
+		/* Don't dereference any memory if the array is empty. */
+		return NULL;
+	}
+
 	/* Get an aligned pointer. */
-	const uintptr_t s_int = (uintptr_t) s;
-	const uint32_t *p = (const uint32_t *)(s_int & -4);
+	s_int = (uintptr_t) s;
+	p = (const uint32_t *)(s_int & -4);
 
 	/* Create four copies of the byte for which we are looking. */
-	const uint32_t goal = 0x01010101 * (uint8_t) c;
+	goal = 0x01010101 * (uint8_t) c;
 
 	/* Read the first word, but munge it so that bytes before the array
 	 * will not match goal.
@@ -31,23 +43,14 @@ void *memchr(const void *s, int c, size_t n)
 	 * Note that this shift count expression works because we know
 	 * shift counts are taken mod 32.
 	 */
-	const uint32_t before_mask = (1 << (s_int << 3)) - 1;
-	uint32_t v = (*p | before_mask) ^ (goal & before_mask);
+	before_mask = (1 << (s_int << 3)) - 1;
+	v = (*p | before_mask) ^ (goal & before_mask);
 
 	/* Compute the address of the last byte. */
-	const char *const last_byte_ptr = (const char *)s + n - 1;
+	last_byte_ptr = (const char *)s + n - 1;
 
 	/* Compute the address of the word containing the last byte. */
-	const uint32_t *const last_word_ptr =
-	    (const uint32_t *)((uintptr_t) last_byte_ptr & -4);
-
-	uint32_t bits;
-	char *ret;
-
-	if (__builtin_expect(n == 0, 0)) {
-		/* Don't dereference any memory if the array is empty. */
-		return NULL;
-	}
+	last_word_ptr = (const uint32_t *)((uintptr_t) last_byte_ptr & -4);
 
 	while ((bits = __insn_seqb(v, goal)) == 0) {
 		if (__builtin_expect(p == last_word_ptr, 0)) {
diff --git a/arch/tile/lib/memcpy_tile64.c b/arch/tile/lib/memcpy_tile64.c
index dfedea7b266b..f7d4a6ad61e8 100644
--- a/arch/tile/lib/memcpy_tile64.c
+++ b/arch/tile/lib/memcpy_tile64.c
@@ -54,7 +54,7 @@ typedef unsigned long (*memcpy_t)(void *, const void *, unsigned long);
  * we must run with interrupts disabled to avoid the risk of some
  * other code seeing the incoherent data in our cache.  (Recall that
  * our cache is indexed by PA, so even if the other code doesn't use
- * our KM_MEMCPY virtual addresses, they'll still hit in cache using
+ * our kmap_atomic virtual addresses, they'll still hit in cache using
  * the normal VAs that aren't supposed to hit in cache.)
  */
 static void memcpy_multicache(void *dest, const void *source,
@@ -64,6 +64,7 @@ static void memcpy_multicache(void *dest, const void *source,
 	unsigned long flags, newsrc, newdst;
 	pmd_t *pmdp;
 	pte_t *ptep;
+	int type0, type1;
 	int cpu = get_cpu();
 
 	/*
@@ -77,7 +78,8 @@ static void memcpy_multicache(void *dest, const void *source,
 	sim_allow_multiple_caching(1);
 
 	/* Set up the new dest mapping */
-	idx = FIX_KMAP_BEGIN + (KM_TYPE_NR * cpu) + KM_MEMCPY0;
+	type0 = kmap_atomic_idx_push();
+	idx = FIX_KMAP_BEGIN + (KM_TYPE_NR * cpu) + type0;
 	newdst = __fix_to_virt(idx) + ((unsigned long)dest & (PAGE_SIZE-1));
 	pmdp = pmd_offset(pud_offset(pgd_offset_k(newdst), newdst), newdst);
 	ptep = pte_offset_kernel(pmdp, newdst);
@@ -87,7 +89,8 @@ static void memcpy_multicache(void *dest, const void *source,
 	}
 
 	/* Set up the new source mapping */
-	idx += (KM_MEMCPY0 - KM_MEMCPY1);
+	type1 = kmap_atomic_idx_push();
+	idx += (type0 - type1);
 	src_pte = hv_pte_set_nc(src_pte);
 	src_pte = hv_pte_clear_writable(src_pte);  /* be paranoid */
 	newsrc = __fix_to_virt(idx) + ((unsigned long)source & (PAGE_SIZE-1));
@@ -119,6 +122,8 @@ static void memcpy_multicache(void *dest, const void *source,
 	 * We're done: notify the simulator that all is back to normal,
 	 * and re-enable interrupts and pre-emption.
 	 */
+	kmap_atomic_idx_pop();
+	kmap_atomic_idx_pop();
 	sim_allow_multiple_caching(0);
 	local_irq_restore(flags);
 	put_cpu();
diff --git a/arch/tile/lib/spinlock_32.c b/arch/tile/lib/spinlock_32.c
index 485e24d62c6b..5cd1c4004eca 100644
--- a/arch/tile/lib/spinlock_32.c
+++ b/arch/tile/lib/spinlock_32.c
@@ -167,23 +167,30 @@ void arch_write_lock_slow(arch_rwlock_t *rwlock, u32 val)
 	 * when we compare them.
 	 */
 	u32 my_ticket_;
+	u32 iterations = 0;
 
-	/* Take out the next ticket; this will also stop would-be readers. */
-	if (val & 1)
-		val = get_rwlock(rwlock);
-	rwlock->lock = __insn_addb(val, 1 << WR_NEXT_SHIFT);
+	/*
+	 * Wait until there are no readers, then bump up the next
+	 * field and capture the ticket value.
+	 */
+	for (;;) {
+		if (!(val & 1)) {
+			if ((val >> RD_COUNT_SHIFT) == 0)
+				break;
+			rwlock->lock = val;
+		}
+		delay_backoff(iterations++);
+		val = __insn_tns((int *)&rwlock->lock);
+	}
 
-	/* Extract my ticket value from the original word. */
+	/* Take out the next ticket and extract my ticket value. */
+	rwlock->lock = __insn_addb(val, 1 << WR_NEXT_SHIFT);
 	my_ticket_ = val >> WR_NEXT_SHIFT;
 
-	/*
-	 * Wait until the "current" field matches our ticket, and
-	 * there are no remaining readers.
-	 */
+	/* Wait until the "current" field matches our ticket. */
 	for (;;) {
 		u32 curr_ = val >> WR_CURR_SHIFT;
-		u32 readers = val >> RD_COUNT_SHIFT;
-		u32 delta = ((my_ticket_ - curr_) & WR_MASK) + !!readers;
+		u32 delta = ((my_ticket_ - curr_) & WR_MASK);
 		if (likely(delta == 0))
 			break;