diff options
author | David S. Miller <davem@davemloft.net> | 2005-06-27 18:42:04 -0400 |
---|---|---|
committer | David S. Miller <davem@davemloft.net> | 2005-06-27 18:42:04 -0400 |
commit | b445e26cbf784cdba10f2b6c3e2cd3ee7bab360a (patch) | |
tree | 8c8c377a7e5cbf608d730ec45e091e4f2b826a82 /arch/sparc64/lib/atomic.S | |
parent | 020f46a39eb7b99a575b9f4d105fce2b142acdf1 (diff) |
[SPARC64]: Avoid membar instructions in delay slots.
In particular, avoid membar instructions in the delay
slot of a jmpl instruction.
UltraSPARC-I, II, IIi, and IIe have a bug, documented in
the UltraSPARC-IIi User's Manual, Appendix K, Erratum 51
The long and short of it is that if the IMU unit misses
on a branch or jmpl, and there is a store buffer synchronizing
membar in the delay slot, the chip can stop fetching instructions.
If interrupts are enabled or some other trap is enabled, the
chip will unwedge itself, but performance will suffer.
We already had a workaround for this bug in a few spots, but
it's better to have the entire tree sanitized for this rule.
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'arch/sparc64/lib/atomic.S')
-rw-r--r-- | arch/sparc64/lib/atomic.S | 42 |
1 files changed, 26 insertions, 16 deletions
diff --git a/arch/sparc64/lib/atomic.S b/arch/sparc64/lib/atomic.S index e528b8d1a3e6..faf87c31598b 100644 --- a/arch/sparc64/lib/atomic.S +++ b/arch/sparc64/lib/atomic.S | |||
@@ -7,18 +7,6 @@ | |||
7 | #include <linux/config.h> | 7 | #include <linux/config.h> |
8 | #include <asm/asi.h> | 8 | #include <asm/asi.h> |
9 | 9 | ||
10 | /* On SMP we need to use memory barriers to ensure | ||
11 | * correct memory operation ordering, nop these out | ||
12 | * for uniprocessor. | ||
13 | */ | ||
14 | #ifdef CONFIG_SMP | ||
15 | #define ATOMIC_PRE_BARRIER membar #StoreLoad | #LoadLoad | ||
16 | #define ATOMIC_POST_BARRIER membar #StoreLoad | #StoreStore | ||
17 | #else | ||
18 | #define ATOMIC_PRE_BARRIER nop | ||
19 | #define ATOMIC_POST_BARRIER nop | ||
20 | #endif | ||
21 | |||
22 | .text | 10 | .text |
23 | 11 | ||
24 | /* Two versions of the atomic routines, one that | 12 | /* Two versions of the atomic routines, one that |
@@ -52,6 +40,24 @@ atomic_sub: /* %o0 = decrement, %o1 = atomic_ptr */ | |||
52 | nop | 40 | nop |
53 | .size atomic_sub, .-atomic_sub | 41 | .size atomic_sub, .-atomic_sub |
54 | 42 | ||
43 | /* On SMP we need to use memory barriers to ensure | ||
44 | * correct memory operation ordering, nop these out | ||
45 | * for uniprocessor. | ||
46 | */ | ||
47 | #ifdef CONFIG_SMP | ||
48 | |||
49 | #define ATOMIC_PRE_BARRIER membar #StoreLoad | #LoadLoad; | ||
50 | #define ATOMIC_POST_BARRIER \ | ||
51 | ba,pt %xcc, 80b; \ | ||
52 | membar #StoreLoad | #StoreStore | ||
53 | |||
54 | 80: retl | ||
55 | nop | ||
56 | #else | ||
57 | #define ATOMIC_PRE_BARRIER | ||
58 | #define ATOMIC_POST_BARRIER | ||
59 | #endif | ||
60 | |||
55 | .globl atomic_add_ret | 61 | .globl atomic_add_ret |
56 | .type atomic_add_ret,#function | 62 | .type atomic_add_ret,#function |
57 | atomic_add_ret: /* %o0 = increment, %o1 = atomic_ptr */ | 63 | atomic_add_ret: /* %o0 = increment, %o1 = atomic_ptr */ |
@@ -62,9 +68,10 @@ atomic_add_ret: /* %o0 = increment, %o1 = atomic_ptr */ | |||
62 | cmp %g1, %g7 | 68 | cmp %g1, %g7 |
63 | bne,pn %icc, 1b | 69 | bne,pn %icc, 1b |
64 | add %g7, %o0, %g7 | 70 | add %g7, %o0, %g7 |
71 | sra %g7, 0, %o0 | ||
65 | ATOMIC_POST_BARRIER | 72 | ATOMIC_POST_BARRIER |
66 | retl | 73 | retl |
67 | sra %g7, 0, %o0 | 74 | nop |
68 | .size atomic_add_ret, .-atomic_add_ret | 75 | .size atomic_add_ret, .-atomic_add_ret |
69 | 76 | ||
70 | .globl atomic_sub_ret | 77 | .globl atomic_sub_ret |
@@ -77,9 +84,10 @@ atomic_sub_ret: /* %o0 = decrement, %o1 = atomic_ptr */ | |||
77 | cmp %g1, %g7 | 84 | cmp %g1, %g7 |
78 | bne,pn %icc, 1b | 85 | bne,pn %icc, 1b |
79 | sub %g7, %o0, %g7 | 86 | sub %g7, %o0, %g7 |
87 | sra %g7, 0, %o0 | ||
80 | ATOMIC_POST_BARRIER | 88 | ATOMIC_POST_BARRIER |
81 | retl | 89 | retl |
82 | sra %g7, 0, %o0 | 90 | nop |
83 | .size atomic_sub_ret, .-atomic_sub_ret | 91 | .size atomic_sub_ret, .-atomic_sub_ret |
84 | 92 | ||
85 | .globl atomic64_add | 93 | .globl atomic64_add |
@@ -118,9 +126,10 @@ atomic64_add_ret: /* %o0 = increment, %o1 = atomic_ptr */ | |||
118 | cmp %g1, %g7 | 126 | cmp %g1, %g7 |
119 | bne,pn %xcc, 1b | 127 | bne,pn %xcc, 1b |
120 | add %g7, %o0, %g7 | 128 | add %g7, %o0, %g7 |
129 | mov %g7, %o0 | ||
121 | ATOMIC_POST_BARRIER | 130 | ATOMIC_POST_BARRIER |
122 | retl | 131 | retl |
123 | mov %g7, %o0 | 132 | nop |
124 | .size atomic64_add_ret, .-atomic64_add_ret | 133 | .size atomic64_add_ret, .-atomic64_add_ret |
125 | 134 | ||
126 | .globl atomic64_sub_ret | 135 | .globl atomic64_sub_ret |
@@ -133,7 +142,8 @@ atomic64_sub_ret: /* %o0 = decrement, %o1 = atomic_ptr */ | |||
133 | cmp %g1, %g7 | 142 | cmp %g1, %g7 |
134 | bne,pn %xcc, 1b | 143 | bne,pn %xcc, 1b |
135 | sub %g7, %o0, %g7 | 144 | sub %g7, %o0, %g7 |
145 | mov %g7, %o0 | ||
136 | ATOMIC_POST_BARRIER | 146 | ATOMIC_POST_BARRIER |
137 | retl | 147 | retl |
138 | mov %g7, %o0 | 148 | nop |
139 | .size atomic64_sub_ret, .-atomic64_sub_ret | 149 | .size atomic64_sub_ret, .-atomic64_sub_ret |