diff options
author | David S. Miller <davem@sunset.davemloft.net> | 2005-10-07 16:30:49 -0400 |
---|---|---|
committer | David S. Miller <davem@sunset.davemloft.net> | 2005-10-07 16:30:49 -0400 |
commit | ba6399334dd8a75bd295de26496196c720abae0a (patch) | |
tree | 0535e4bcfe8da7d0dde0bcf1ba8cb914c5f12e46 /arch/sparc64/kernel | |
parent | d85c3553df5e24cb3117385f0a17e1cc0436d148 (diff) |
[SPARC64]: Fix userland FPU state corruption.
We need to use stricter memory barriers around the block
load and store instructions we use to save and restore the
FPU register file.
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'arch/sparc64/kernel')
-rw-r--r-- | arch/sparc64/kernel/entry.S | 39 | ||||
-rw-r--r-- | arch/sparc64/kernel/rtrap.S | 7 |
2 files changed, 25 insertions, 21 deletions
diff --git a/arch/sparc64/kernel/entry.S b/arch/sparc64/kernel/entry.S index f685035dbdb8..11a848402fb1 100644 --- a/arch/sparc64/kernel/entry.S +++ b/arch/sparc64/kernel/entry.S | |||
@@ -33,7 +33,7 @@ | |||
33 | /* This is trivial with the new code... */ | 33 | /* This is trivial with the new code... */ |
34 | .globl do_fpdis | 34 | .globl do_fpdis |
35 | do_fpdis: | 35 | do_fpdis: |
36 | sethi %hi(TSTATE_PEF), %g4 ! IEU0 | 36 | sethi %hi(TSTATE_PEF), %g4 |
37 | rdpr %tstate, %g5 | 37 | rdpr %tstate, %g5 |
38 | andcc %g5, %g4, %g0 | 38 | andcc %g5, %g4, %g0 |
39 | be,pt %xcc, 1f | 39 | be,pt %xcc, 1f |
@@ -50,18 +50,18 @@ do_fpdis: | |||
50 | add %g0, %g0, %g0 | 50 | add %g0, %g0, %g0 |
51 | ba,a,pt %xcc, rtrap_clr_l6 | 51 | ba,a,pt %xcc, rtrap_clr_l6 |
52 | 52 | ||
53 | 1: ldub [%g6 + TI_FPSAVED], %g5 ! Load Group | 53 | 1: ldub [%g6 + TI_FPSAVED], %g5 |
54 | wr %g0, FPRS_FEF, %fprs ! LSU Group+4bubbles | 54 | wr %g0, FPRS_FEF, %fprs |
55 | andcc %g5, FPRS_FEF, %g0 ! IEU1 Group | 55 | andcc %g5, FPRS_FEF, %g0 |
56 | be,a,pt %icc, 1f ! CTI | 56 | be,a,pt %icc, 1f |
57 | clr %g7 ! IEU0 | 57 | clr %g7 |
58 | ldx [%g6 + TI_GSR], %g7 ! Load Group | 58 | ldx [%g6 + TI_GSR], %g7 |
59 | 1: andcc %g5, FPRS_DL, %g0 ! IEU1 | 59 | 1: andcc %g5, FPRS_DL, %g0 |
60 | bne,pn %icc, 2f ! CTI | 60 | bne,pn %icc, 2f |
61 | fzero %f0 ! FPA | 61 | fzero %f0 |
62 | andcc %g5, FPRS_DU, %g0 ! IEU1 Group | 62 | andcc %g5, FPRS_DU, %g0 |
63 | bne,pn %icc, 1f ! CTI | 63 | bne,pn %icc, 1f |
64 | fzero %f2 ! FPA | 64 | fzero %f2 |
65 | faddd %f0, %f2, %f4 | 65 | faddd %f0, %f2, %f4 |
66 | fmuld %f0, %f2, %f6 | 66 | fmuld %f0, %f2, %f6 |
67 | faddd %f0, %f2, %f8 | 67 | faddd %f0, %f2, %f8 |
@@ -104,8 +104,10 @@ do_fpdis: | |||
104 | add %g6, TI_FPREGS + 0xc0, %g2 | 104 | add %g6, TI_FPREGS + 0xc0, %g2 |
105 | faddd %f0, %f2, %f8 | 105 | faddd %f0, %f2, %f8 |
106 | fmuld %f0, %f2, %f10 | 106 | fmuld %f0, %f2, %f10 |
107 | ldda [%g1] ASI_BLK_S, %f32 ! grrr, where is ASI_BLK_NUCLEUS 8-( | 107 | membar #Sync |
108 | ldda [%g1] ASI_BLK_S, %f32 | ||
108 | ldda [%g2] ASI_BLK_S, %f48 | 109 | ldda [%g2] ASI_BLK_S, %f48 |
110 | membar #Sync | ||
109 | faddd %f0, %f2, %f12 | 111 | faddd %f0, %f2, %f12 |
110 | fmuld %f0, %f2, %f14 | 112 | fmuld %f0, %f2, %f14 |
111 | faddd %f0, %f2, %f16 | 113 | faddd %f0, %f2, %f16 |
@@ -116,7 +118,6 @@ do_fpdis: | |||
116 | fmuld %f0, %f2, %f26 | 118 | fmuld %f0, %f2, %f26 |
117 | faddd %f0, %f2, %f28 | 119 | faddd %f0, %f2, %f28 |
118 | fmuld %f0, %f2, %f30 | 120 | fmuld %f0, %f2, %f30 |
119 | membar #Sync | ||
120 | b,pt %xcc, fpdis_exit | 121 | b,pt %xcc, fpdis_exit |
121 | nop | 122 | nop |
122 | 2: andcc %g5, FPRS_DU, %g0 | 123 | 2: andcc %g5, FPRS_DU, %g0 |
@@ -133,8 +134,10 @@ do_fpdis: | |||
133 | add %g6, TI_FPREGS + 0x40, %g2 | 134 | add %g6, TI_FPREGS + 0x40, %g2 |
134 | faddd %f32, %f34, %f36 | 135 | faddd %f32, %f34, %f36 |
135 | fmuld %f32, %f34, %f38 | 136 | fmuld %f32, %f34, %f38 |
136 | ldda [%g1] ASI_BLK_S, %f0 ! grrr, where is ASI_BLK_NUCLEUS 8-( | 137 | membar #Sync |
138 | ldda [%g1] ASI_BLK_S, %f0 | ||
137 | ldda [%g2] ASI_BLK_S, %f16 | 139 | ldda [%g2] ASI_BLK_S, %f16 |
140 | membar #Sync | ||
138 | faddd %f32, %f34, %f40 | 141 | faddd %f32, %f34, %f40 |
139 | fmuld %f32, %f34, %f42 | 142 | fmuld %f32, %f34, %f42 |
140 | faddd %f32, %f34, %f44 | 143 | faddd %f32, %f34, %f44 |
@@ -147,7 +150,6 @@ do_fpdis: | |||
147 | fmuld %f32, %f34, %f58 | 150 | fmuld %f32, %f34, %f58 |
148 | faddd %f32, %f34, %f60 | 151 | faddd %f32, %f34, %f60 |
149 | fmuld %f32, %f34, %f62 | 152 | fmuld %f32, %f34, %f62 |
150 | membar #Sync | ||
151 | ba,pt %xcc, fpdis_exit | 153 | ba,pt %xcc, fpdis_exit |
152 | nop | 154 | nop |
153 | 3: mov SECONDARY_CONTEXT, %g3 | 155 | 3: mov SECONDARY_CONTEXT, %g3 |
@@ -158,7 +160,8 @@ do_fpdis: | |||
158 | stxa %g2, [%g3] ASI_DMMU | 160 | stxa %g2, [%g3] ASI_DMMU |
159 | membar #Sync | 161 | membar #Sync |
160 | mov 0x40, %g2 | 162 | mov 0x40, %g2 |
161 | ldda [%g1] ASI_BLK_S, %f0 ! grrr, where is ASI_BLK_NUCLEUS 8-( | 163 | membar #Sync |
164 | ldda [%g1] ASI_BLK_S, %f0 | ||
162 | ldda [%g1 + %g2] ASI_BLK_S, %f16 | 165 | ldda [%g1 + %g2] ASI_BLK_S, %f16 |
163 | add %g1, 0x80, %g1 | 166 | add %g1, 0x80, %g1 |
164 | ldda [%g1] ASI_BLK_S, %f32 | 167 | ldda [%g1] ASI_BLK_S, %f32 |
diff --git a/arch/sparc64/kernel/rtrap.S b/arch/sparc64/kernel/rtrap.S index ecfb42a69a44..090dcca00d2a 100644 --- a/arch/sparc64/kernel/rtrap.S +++ b/arch/sparc64/kernel/rtrap.S | |||
@@ -312,32 +312,33 @@ kern_fpucheck: ldub [%g6 + TI_FPDEPTH], %l5 | |||
312 | wr %g1, FPRS_FEF, %fprs | 312 | wr %g1, FPRS_FEF, %fprs |
313 | ldx [%o1 + %o5], %g1 | 313 | ldx [%o1 + %o5], %g1 |
314 | add %g6, TI_XFSR, %o1 | 314 | add %g6, TI_XFSR, %o1 |
315 | membar #StoreLoad | #LoadLoad | ||
316 | sll %o0, 8, %o2 | 315 | sll %o0, 8, %o2 |
317 | add %g6, TI_FPREGS, %o3 | 316 | add %g6, TI_FPREGS, %o3 |
318 | brz,pn %l6, 1f | 317 | brz,pn %l6, 1f |
319 | add %g6, TI_FPREGS+0x40, %o4 | 318 | add %g6, TI_FPREGS+0x40, %o4 |
320 | 319 | ||
320 | membar #Sync | ||
321 | ldda [%o3 + %o2] ASI_BLK_P, %f0 | 321 | ldda [%o3 + %o2] ASI_BLK_P, %f0 |
322 | ldda [%o4 + %o2] ASI_BLK_P, %f16 | 322 | ldda [%o4 + %o2] ASI_BLK_P, %f16 |
323 | membar #Sync | ||
323 | 1: andcc %l2, FPRS_DU, %g0 | 324 | 1: andcc %l2, FPRS_DU, %g0 |
324 | be,pn %icc, 1f | 325 | be,pn %icc, 1f |
325 | wr %g1, 0, %gsr | 326 | wr %g1, 0, %gsr |
326 | add %o2, 0x80, %o2 | 327 | add %o2, 0x80, %o2 |
328 | membar #Sync | ||
327 | ldda [%o3 + %o2] ASI_BLK_P, %f32 | 329 | ldda [%o3 + %o2] ASI_BLK_P, %f32 |
328 | ldda [%o4 + %o2] ASI_BLK_P, %f48 | 330 | ldda [%o4 + %o2] ASI_BLK_P, %f48 |
329 | |||
330 | 1: membar #Sync | 331 | 1: membar #Sync |
331 | ldx [%o1 + %o5], %fsr | 332 | ldx [%o1 + %o5], %fsr |
332 | 2: stb %l5, [%g6 + TI_FPDEPTH] | 333 | 2: stb %l5, [%g6 + TI_FPDEPTH] |
333 | ba,pt %xcc, rt_continue | 334 | ba,pt %xcc, rt_continue |
334 | nop | 335 | nop |
335 | 5: wr %g0, FPRS_FEF, %fprs | 336 | 5: wr %g0, FPRS_FEF, %fprs |
336 | membar #StoreLoad | #LoadLoad | ||
337 | sll %o0, 8, %o2 | 337 | sll %o0, 8, %o2 |
338 | 338 | ||
339 | add %g6, TI_FPREGS+0x80, %o3 | 339 | add %g6, TI_FPREGS+0x80, %o3 |
340 | add %g6, TI_FPREGS+0xc0, %o4 | 340 | add %g6, TI_FPREGS+0xc0, %o4 |
341 | membar #Sync | ||
341 | ldda [%o3 + %o2] ASI_BLK_P, %f32 | 342 | ldda [%o3 + %o2] ASI_BLK_P, %f32 |
342 | ldda [%o4 + %o2] ASI_BLK_P, %f48 | 343 | ldda [%o4 + %o2] ASI_BLK_P, %f48 |
343 | membar #Sync | 344 | membar #Sync |