diff options
Diffstat (limited to 'arch/alpha/lib/ev67-strlen_user.S')
-rw-r--r-- | arch/alpha/lib/ev67-strlen_user.S | 107 |
1 files changed, 107 insertions, 0 deletions
diff --git a/arch/alpha/lib/ev67-strlen_user.S b/arch/alpha/lib/ev67-strlen_user.S new file mode 100644 index 000000000000..57e0d77b81a6 --- /dev/null +++ b/arch/alpha/lib/ev67-strlen_user.S | |||
@@ -0,0 +1,107 @@ | |||
1 | /* | ||
2 | * arch/alpha/lib/ev67-strlen_user.S | ||
3 | * 21264 version contributed by Rick Gorton <rick.gorton@api-networks.com> | ||
4 | * | ||
5 | * Return the length of the string including the NULL terminator | ||
6 | * (strlen+1) or zero if an error occurred. | ||
7 | * | ||
8 | * In places where it is critical to limit the processing time, | ||
9 | * and the data is not trusted, strnlen_user() should be used. | ||
10 | * It will return a value greater than its second argument if | ||
11 | * that limit would be exceeded. This implementation is allowed | ||
12 | * to access memory beyond the limit, but will not cross a page | ||
13 | * boundary when doing so. | ||
14 | * | ||
15 | * Much of the information about 21264 scheduling/coding comes from: | ||
16 | * Compiler Writer's Guide for the Alpha 21264 | ||
17 | * abbreviated as 'CWG' in other comments here | ||
18 | * ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html | ||
19 | * Scheduling notation: | ||
20 | * E - either cluster | ||
21 | * U - upper subcluster; U0 - subcluster U0; U1 - subcluster U1 | ||
22 | * L - lower subcluster; L0 - subcluster L0; L1 - subcluster L1 | ||
23 | * Try not to change the actual algorithm if possible for consistency. | ||
24 | */ | ||
25 | |||
26 | #include <asm/regdef.h> | ||
27 | |||
28 | |||
29 | /* Allow an exception for an insn; exit if we get one. */ | ||
30 | #define EX(x,y...) \ | ||
31 | 99: x,##y; \ | ||
32 | .section __ex_table,"a"; \ | ||
33 | .long 99b - .; \ | ||
34 | lda v0, $exception-99b(zero); \ | ||
35 | .previous | ||
36 | |||
37 | |||
38 | .set noreorder | ||
39 | .set noat | ||
40 | .text | ||
41 | |||
42 | .globl __strlen_user | ||
43 | .ent __strlen_user | ||
44 | .frame sp, 0, ra | ||
45 | |||
46 | .align 4 | ||
47 | __strlen_user: | ||
48 | ldah a1, 32767(zero) # do not use plain strlen_user() for strings | ||
49 | # that might be almost 2 GB long; you should | ||
50 | # be using strnlen_user() instead | ||
51 | nop | ||
52 | nop | ||
53 | nop | ||
54 | |||
55 | .globl __strnlen_user | ||
56 | |||
57 | .align 4 | ||
58 | __strnlen_user: | ||
59 | .prologue 0 | ||
60 | EX( ldq_u t0, 0(a0) ) # L : load first quadword (a0 may be misaligned) | ||
61 | lda t1, -1(zero) # E : | ||
62 | |||
63 | insqh t1, a0, t1 # U : | ||
64 | andnot a0, 7, v0 # E : | ||
65 | or t1, t0, t0 # E : | ||
66 | subq a0, 1, a0 # E : get our +1 for the return | ||
67 | |||
68 | cmpbge zero, t0, t1 # E : t1 <- bitmask: bit i == 1 <==> i-th byte == 0 | ||
69 | subq a1, 7, t2 # E : | ||
70 | subq a0, v0, t0 # E : | ||
71 | bne t1, $found # U : | ||
72 | |||
73 | addq t2, t0, t2 # E : | ||
74 | addq a1, 1, a1 # E : | ||
75 | nop # E : | ||
76 | nop # E : | ||
77 | |||
78 | .align 4 | ||
79 | $loop: ble t2, $limit # U : | ||
80 | EX( ldq t0, 8(v0) ) # L : | ||
81 | nop # E : | ||
82 | nop # E : | ||
83 | |||
84 | cmpbge zero, t0, t1 # E : | ||
85 | subq t2, 8, t2 # E : | ||
86 | addq v0, 8, v0 # E : addr += 8 | ||
87 | beq t1, $loop # U : | ||
88 | |||
89 | $found: cttz t1, t2 # U0 : | ||
90 | addq v0, t2, v0 # E : | ||
91 | subq v0, a0, v0 # E : | ||
92 | ret # L0 : | ||
93 | |||
94 | $exception: | ||
95 | nop | ||
96 | nop | ||
97 | nop | ||
98 | ret | ||
99 | |||
100 | .align 4 # currently redundant | ||
101 | $limit: | ||
102 | nop | ||
103 | nop | ||
104 | subq a1, t2, v0 | ||
105 | ret | ||
106 | |||
107 | .end __strlen_user | ||