10 files changed, 71 insertions, 207 deletions

diff --git a/arch/tile/lib/atomic_32.c b/arch/tile/lib/atomic_32.c
index 771b251b409d..f5cada70c3c8 100644
--- a/arch/tile/lib/atomic_32.c
+++ b/arch/tile/lib/atomic_32.c
@@ -18,7 +18,6 @@
 #include <linux/module.h>
 #include <linux/mm.h>
 #include <linux/atomic.h>
-#include <asm/futex.h>
 #include <arch/chip.h>
 
 /* See <asm/atomic_32.h> */
@@ -50,7 +49,7 @@ int atomic_locks[PAGE_SIZE / sizeof(int)] __page_aligned_bss;
 
 #endif /* ATOMIC_LOCKS_FOUND_VIA_TABLE() */
 
-static inline int *__atomic_hashed_lock(volatile void *v)
+int *__atomic_hashed_lock(volatile void *v)
 {
         /* NOTE: this code must match "sys_cmpxchg" in kernel/intvec_32.S */
 #if ATOMIC_LOCKS_FOUND_VIA_TABLE()
@@ -191,47 +190,6 @@ u64 _atomic64_cmpxchg(atomic64_t *v, u64 o, u64 n)
 EXPORT_SYMBOL(_atomic64_cmpxchg);
 
 
-static inline int *__futex_setup(int __user *v)
-{
-        /*
-         * Issue a prefetch to the counter to bring it into cache.
-         * As for __atomic_setup, but we can't do a read into the L1
-         * since it might fault; instead we do a prefetch into the L2.
-         */
-        __insn_prefetch(v);
-        return __atomic_hashed_lock((int __force *)v);
-}
-
-struct __get_user futex_set(u32 __user *v, int i)
-{
-        return __atomic_xchg((int __force *)v, __futex_setup(v), i);
-}
-
-struct __get_user futex_add(u32 __user *v, int n)
-{
-        return __atomic_xchg_add((int __force *)v, __futex_setup(v), n);
-}
-
-struct __get_user futex_or(u32 __user *v, int n)
-{
-        return __atomic_or((int __force *)v, __futex_setup(v), n);
-}
-
-struct __get_user futex_andn(u32 __user *v, int n)
-{
-        return __atomic_andn((int __force *)v, __futex_setup(v), n);
-}
-
-struct __get_user futex_xor(u32 __user *v, int n)
-{
-        return __atomic_xor((int __force *)v, __futex_setup(v), n);
-}
-
-struct __get_user futex_cmpxchg(u32 __user *v, int o, int n)
-{
-        return __atomic_cmpxchg((int __force *)v, __futex_setup(v), o, n);
-}
-
 /*
  * If any of the atomic or futex routines hit a bad address (not in
  * the page tables at kernel PL) this routine is called.  The futex
@@ -323,7 +281,4 @@ void __init __init_atomic_per_cpu(void)
         BUILD_BUG_ON((PAGE_SIZE >> 3) > ATOMIC_HASH_SIZE);
 
 #endif /* ATOMIC_LOCKS_FOUND_VIA_TABLE() */
-
-        /* The futex code makes this assumption, so we validate it here. */
-        BUILD_BUG_ON(sizeof(atomic_t) != sizeof(int));
 }
diff --git a/arch/tile/lib/exports.c b/arch/tile/lib/exports.c
index 2a81d32de0da..dd5f0a33fdaf 100644
--- a/arch/tile/lib/exports.c
+++ b/arch/tile/lib/exports.c
@@ -18,14 +18,6 @@
 
 /* arch/tile/lib/usercopy.S */
 #include <linux/uaccess.h>
-EXPORT_SYMBOL(__get_user_1);
-EXPORT_SYMBOL(__get_user_2);
-EXPORT_SYMBOL(__get_user_4);
-EXPORT_SYMBOL(__get_user_8);
-EXPORT_SYMBOL(__put_user_1);
-EXPORT_SYMBOL(__put_user_2);
-EXPORT_SYMBOL(__put_user_4);
-EXPORT_SYMBOL(__put_user_8);
 EXPORT_SYMBOL(strnlen_user_asm);
 EXPORT_SYMBOL(strncpy_from_user_asm);
 EXPORT_SYMBOL(clear_user_asm);
diff --git a/arch/tile/lib/memchr_64.c b/arch/tile/lib/memchr_64.c
index 84fdc8d8e735..6f867dbf7c56 100644
--- a/arch/tile/lib/memchr_64.c
+++ b/arch/tile/lib/memchr_64.c
@@ -15,6 +15,7 @@
 #include <linux/types.h>
 #include <linux/string.h>
 #include <linux/module.h>
+#include "string-endian.h"
 
 void *memchr(const void *s, int c, size_t n)
 {
@@ -39,11 +40,8 @@ void *memchr(const void *s, int c, size_t n)
 
         /* Read the first word, but munge it so that bytes before the array
          * will not match goal.
-         *
-         * Note that this shift count expression works because we know
-         * shift counts are taken mod 64.
          */
-        before_mask = (1ULL << (s_int << 3)) - 1;
+        before_mask = MASK(s_int);
         v = (*p | before_mask) ^ (goal & before_mask);
 
         /* Compute the address of the last byte. */
@@ -65,7 +63,7 @@ void *memchr(const void *s, int c, size_t n)
         /* We found a match, but it might be in a byte past the end
          * of the array.
          */
-        ret = ((char *)p) + (__insn_ctz(bits) >> 3);
+        ret = ((char *)p) + (CFZ(bits) >> 3);
         return (ret <= last_byte_ptr) ? ret : NULL;
 }
 EXPORT_SYMBOL(memchr);
diff --git a/arch/tile/lib/memcpy_64.c b/arch/tile/lib/memcpy_64.c
index 3fab9a6a2bbe..c79b8e7c6828 100644
--- a/arch/tile/lib/memcpy_64.c
+++ b/arch/tile/lib/memcpy_64.c
@@ -15,7 +15,6 @@
 #include <linux/types.h>
 #include <linux/string.h>
 #include <linux/module.h>
-#define __memcpy memcpy
 /* EXPORT_SYMBOL() is in arch/tile/lib/exports.c since this should be asm. */
 
 /* Must be 8 bytes in size. */
@@ -188,6 +187,7 @@ int USERCOPY_FUNC(void *__restrict dstv, const void *__restrict srcv, size_t n)
 
         /* n != 0 if we get here.  Write out any trailing bytes. */
         dst1 = (char *)dst8;
+#ifndef __BIG_ENDIAN__
         if (n & 4) {
                 ST4((uint32_t *)dst1, final);
                 dst1 += 4;
@@ -202,11 +202,30 @@ int USERCOPY_FUNC(void *__restrict dstv, const void *__restrict srcv, size_t n)
         }
         if (n)
                 ST1((uint8_t *)dst1, final);
+#else
+        if (n & 4) {
+                ST4((uint32_t *)dst1, final >> 32);
+                dst1 += 4;
+        }
+        else
+        {
+                final >>= 32;
+        }
+        if (n & 2) {
+                ST2((uint16_t *)dst1, final >> 16);
+                dst1 += 2;
+        }
+        else
+        {
+                final >>= 16;
+        }
+        if (n & 1)
+                ST1((uint8_t *)dst1, final >> 8);
+#endif
 
         return RETVAL;
 }
 
-
 #ifdef USERCOPY_FUNC
 #undef ST1
 #undef ST2
diff --git a/arch/tile/lib/memcpy_tile64.c b/arch/tile/lib/memcpy_tile64.c
index b2fe15e01075..3bc4b4e40d93 100644
--- a/arch/tile/lib/memcpy_tile64.c
+++ b/arch/tile/lib/memcpy_tile64.c
@@ -160,7 +160,7 @@ retry_source:
                         break;
                 if (get_remote_cache_cpu(src_pte) == smp_processor_id())
                         break;
-                src_page = pfn_to_page(hv_pte_get_pfn(src_pte));
+                src_page = pfn_to_page(pte_pfn(src_pte));
                 get_page(src_page);
                 if (pte_val(src_pte) != pte_val(*src_ptep)) {
                         put_page(src_page);
@@ -168,7 +168,7 @@ retry_source:
                 }
                 if (pte_huge(src_pte)) {
                         /* Adjust the PTE to correspond to a small page */
-                        int pfn = hv_pte_get_pfn(src_pte);
+                        int pfn = pte_pfn(src_pte);
                         pfn += (((unsigned long)source & (HPAGE_SIZE-1))
                                 >> PAGE_SHIFT);
                         src_pte = pfn_pte(pfn, src_pte);
@@ -188,7 +188,7 @@ retry_dest:
                         put_page(src_page);
                         break;
                 }
-                dst_page = pfn_to_page(hv_pte_get_pfn(dst_pte));
+                dst_page = pfn_to_page(pte_pfn(dst_pte));
                 if (dst_page == src_page) {
                         /*
                          * Source and dest are on the same page; this
@@ -206,7 +206,7 @@ retry_dest:
                 }
                 if (pte_huge(dst_pte)) {
                         /* Adjust the PTE to correspond to a small page */
-                        int pfn = hv_pte_get_pfn(dst_pte);
+                        int pfn = pte_pfn(dst_pte);
                         pfn += (((unsigned long)dest & (HPAGE_SIZE-1))
                                 >> PAGE_SHIFT);
                         dst_pte = pfn_pte(pfn, dst_pte);
diff --git a/arch/tile/lib/strchr_64.c b/arch/tile/lib/strchr_64.c
index 617a9273aaa8..f39f9dc422b0 100644
--- a/arch/tile/lib/strchr_64.c
+++ b/arch/tile/lib/strchr_64.c
@@ -15,8 +15,7 @@
 #include <linux/types.h>
 #include <linux/string.h>
 #include <linux/module.h>
-
-#undef strchr
+#include "string-endian.h"
 
 char *strchr(const char *s, int c)
 {
@@ -33,13 +32,9 @@ char *strchr(const char *s, int c)
          * match neither zero nor goal (we make sure the high bit of each
          * byte is 1, and the low 7 bits are all the opposite of the goal
          * byte).
-         *
-         * Note that this shift count expression works because we know shift
-         * counts are taken mod 64.
          */
-        const uint64_t before_mask = (1ULL << (s_int << 3)) - 1;
-        uint64_t v = (*p | before_mask) ^
-                (goal & __insn_v1shrsi(before_mask, 1));
+        const uint64_t before_mask = MASK(s_int);
+        uint64_t v = (*p | before_mask) ^ (goal & __insn_v1shrui(before_mask, 1));
 
         uint64_t zero_matches, goal_matches;
         while (1) {
@@ -55,8 +50,8 @@ char *strchr(const char *s, int c)
                 v = *++p;
         }
 
-        z = __insn_ctz(zero_matches);
-        g = __insn_ctz(goal_matches);
+        z = CFZ(zero_matches);
+        g = CFZ(goal_matches);
 
         /* If we found c before '\0' we got a match. Note that if c == '\0'
          * then g == z, and we correctly return the address of the '\0'
diff --git a/arch/tile/lib/string-endian.h b/arch/tile/lib/string-endian.h
new file mode 100644
index 000000000000..c0eed7ce69c3
--- /dev/null
+++ b/arch/tile/lib/string-endian.h
@@ -0,0 +1,33 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * Provide a mask based on the pointer alignment that
+ * sets up non-zero bytes before the beginning of the string.
+ * The MASK expression works because shift counts are taken mod 64.
+ * Also, specify how to count "first" and "last" bits
+ * when the bits have been read as a word.
+ */
+
+#include <asm/byteorder.h>
+
+#ifdef __LITTLE_ENDIAN
+#define MASK(x) (__insn_shl(1ULL, (x << 3)) - 1)
+#define NULMASK(x) ((2ULL << x) - 1)
+#define CFZ(x) __insn_ctz(x)
+#define REVCZ(x) __insn_clz(x)
+#else
+#define MASK(x) (__insn_shl(-2LL, ((-x << 3) - 1)))
+#define NULMASK(x) (-2LL << (63 - x))
+#define CFZ(x) __insn_clz(x)
+#define REVCZ(x) __insn_ctz(x)
+#endif
diff --git a/arch/tile/lib/strlen_64.c b/arch/tile/lib/strlen_64.c
index 1c92d46202a8..9583fc3361fa 100644
--- a/arch/tile/lib/strlen_64.c
+++ b/arch/tile/lib/strlen_64.c
@@ -15,8 +15,7 @@
 #include <linux/types.h>
 #include <linux/string.h>
 #include <linux/module.h>
-
-#undef strlen
+#include "string-endian.h"
 
 size_t strlen(const char *s)
 {
@@ -24,15 +23,13 @@ size_t strlen(const char *s)
         const uintptr_t s_int = (uintptr_t) s;
         const uint64_t *p = (const uint64_t *)(s_int & -8);
 
-        /* Read the first word, but force bytes before the string to be nonzero.
-         * This expression works because we know shift counts are taken mod 64.
-         */
-        uint64_t v = *p | ((1ULL << (s_int << 3)) - 1);
+        /* Read and MASK the first word. */
+        uint64_t v = *p | MASK(s_int);
 
         uint64_t bits;
         while ((bits = __insn_v1cmpeqi(v, 0)) == 0)
                 v = *++p;
 
-        return ((const char *)p) + (__insn_ctz(bits) >> 3) - s;
+        return ((const char *)p) + (CFZ(bits) >> 3) - s;
 }
 EXPORT_SYMBOL(strlen);
diff --git a/arch/tile/lib/usercopy_32.S b/arch/tile/lib/usercopy_32.S
index 979f76d83746..b62d002af009 100644
--- a/arch/tile/lib/usercopy_32.S
+++ b/arch/tile/lib/usercopy_32.S
@@ -19,82 +19,6 @@
 
 /* Access user memory, but use MMU to avoid propagating kernel exceptions. */
 
-        .pushsection .fixup,"ax"
-
-get_user_fault:
-        { move r0, zero; move r1, zero }
-        { movei r2, -EFAULT; jrp lr }
-        ENDPROC(get_user_fault)
-
-put_user_fault:
-        { movei r0, -EFAULT; jrp lr }
-        ENDPROC(put_user_fault)
-
-        .popsection
-
-/*
- * __get_user_N functions take a pointer in r0, and return 0 in r2
- * on success, with the value in r0; or else -EFAULT in r2.
- */
-#define __get_user_N(bytes, LOAD) \
-        STD_ENTRY(__get_user_##bytes); \
-1:      { LOAD r0, r0; move r1, zero; move r2, zero }; \
-        jrp lr; \
-        STD_ENDPROC(__get_user_##bytes); \
-        .pushsection __ex_table,"a"; \
-        .word 1b, get_user_fault; \
-        .popsection
-
-__get_user_N(1, lb_u)
-__get_user_N(2, lh_u)
-__get_user_N(4, lw)
-
-/*
- * __get_user_8 takes a pointer in r0, and returns 0 in r2
- * on success, with the value in r0/r1; or else -EFAULT in r2.
- */
-        STD_ENTRY(__get_user_8);
-1:      { lw r0, r0; addi r1, r0, 4 };
-2:      { lw r1, r1; move r2, zero };
-        jrp lr;
-        STD_ENDPROC(__get_user_8);
-        .pushsection __ex_table,"a";
-        .word 1b, get_user_fault;
-        .word 2b, get_user_fault;
-        .popsection
-
-/*
- * __put_user_N functions take a value in r0 and a pointer in r1,
- * and return 0 in r0 on success or -EFAULT on failure.
- */
-#define __put_user_N(bytes, STORE) \
-        STD_ENTRY(__put_user_##bytes); \
-1:      { STORE r1, r0; move r0, zero }; \
-        jrp lr; \
-        STD_ENDPROC(__put_user_##bytes); \
-        .pushsection __ex_table,"a"; \
-        .word 1b, put_user_fault; \
-        .popsection
-
-__put_user_N(1, sb)
-__put_user_N(2, sh)
-__put_user_N(4, sw)
-
-/*
- * __put_user_8 takes a value in r0/r1 and a pointer in r2,
- * and returns 0 in r0 on success or -EFAULT on failure.
- */
-STD_ENTRY(__put_user_8)
-1:      { sw r2, r0; addi r2, r2, 4 }
-2:      { sw r2, r1; move r0, zero }
-        jrp lr
-        STD_ENDPROC(__put_user_8)
-        .pushsection __ex_table,"a"
-        .word 1b, put_user_fault
-        .word 2b, put_user_fault
-        .popsection
-
-
 /*
  * strnlen_user_asm takes the pointer in r0, and the length bound in r1.
  * It returns the length, including the terminating NUL, or zero on exception.
diff --git a/arch/tile/lib/usercopy_64.S b/arch/tile/lib/usercopy_64.S
index 2ff44f87b78e..adb2dbbc70cd 100644
--- a/arch/tile/lib/usercopy_64.S
+++ b/arch/tile/lib/usercopy_64.S
@@ -19,55 +19,6 @@
 
 /* Access user memory, but use MMU to avoid propagating kernel exceptions. */
 
-        .pushsection .fixup,"ax"
-
-get_user_fault:
-        { movei r1, -EFAULT; move r0, zero }
-        jrp lr
-        ENDPROC(get_user_fault)
-
-put_user_fault:
-        { movei r0, -EFAULT; jrp lr }
-        ENDPROC(put_user_fault)
-
-        .popsection
-
-/*
- * __get_user_N functions take a pointer in r0, and return 0 in r1
- * on success, with the value in r0; or else -EFAULT in r1.
- */
-#define __get_user_N(bytes, LOAD) \
-        STD_ENTRY(__get_user_##bytes); \
-1:      { LOAD r0, r0; move r1, zero }; \
-        jrp lr; \
-        STD_ENDPROC(__get_user_##bytes); \
-        .pushsection __ex_table,"a"; \
-        .quad 1b, get_user_fault; \
-        .popsection
-
-__get_user_N(1, ld1u)
-__get_user_N(2, ld2u)
-__get_user_N(4, ld4u)
-__get_user_N(8, ld)
-
-/*
- * __put_user_N functions take a value in r0 and a pointer in r1,
- * and return 0 in r0 on success or -EFAULT on failure.
- */
-#define __put_user_N(bytes, STORE) \
-        STD_ENTRY(__put_user_##bytes); \
-1:      { STORE r1, r0; move r0, zero }; \
-        jrp lr; \
-        STD_ENDPROC(__put_user_##bytes); \
-        .pushsection __ex_table,"a"; \
-        .quad 1b, put_user_fault; \
-        .popsection
-
-__put_user_N(1, st1)
-__put_user_N(2, st2)
-__put_user_N(4, st4)
-__put_user_N(8, st)
-
 /*
  * strnlen_user_asm takes the pointer in r0, and the length bound in r1.
  * It returns the length, including the terminating NUL, or zero on exception.