aboutsummaryrefslogtreecommitdiffstats
path: root/arch
diff options
context:
space:
mode:
authorMark Nutter <mnutter@us.ibm.com>2005-11-15 15:53:51 -0500
committerPaul Mackerras <paulus@samba.org>2006-01-08 22:49:25 -0500
commit05b841174c289ca62a6b42d883b8791d9ac3a4bd (patch)
treef121dff4f40e5caa99a5ab5fd041c0c268dd1f8c /arch
parent7c038749d1e6a2d5fb37ed14aed0fffa34c4e504 (diff)
[PATCH] spufs: add spu-side context switch code
Add the source code that is used to generate spu_save_dump.h and spu_restore_dump.h. Since a full spu tool chain is needed to generate these files, the default remains to use the shipped versions in order to keep the number of tools for building the kernel down. Signed-off-by: Arnd Bergmann <arndb@de.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
Diffstat (limited to 'arch')
-rw-r--r--arch/powerpc/platforms/cell/spufs/Makefile49
-rw-r--r--arch/powerpc/platforms/cell/spufs/spu_restore.c336
-rw-r--r--arch/powerpc/platforms/cell/spufs/spu_restore_crt0.S116
-rw-r--r--arch/powerpc/platforms/cell/spufs/spu_save.c195
-rw-r--r--arch/powerpc/platforms/cell/spufs/spu_save_crt0.S102
-rw-r--r--arch/powerpc/platforms/cell/spufs/spu_utils.h160
6 files changed, 958 insertions, 0 deletions
diff --git a/arch/powerpc/platforms/cell/spufs/Makefile b/arch/powerpc/platforms/cell/spufs/Makefile
index e70e3cc1158f..b38ab747efd7 100644
--- a/arch/powerpc/platforms/cell/spufs/Makefile
+++ b/arch/powerpc/platforms/cell/spufs/Makefile
@@ -2,4 +2,53 @@ obj-$(CONFIG_SPU_FS) += spufs.o
2 2
3spufs-y += inode.o file.o context.o switch.o syscalls.o 3spufs-y += inode.o file.o context.o switch.o syscalls.o
4 4
5# Rules to build switch.o with the help of SPU tool chain
6SPU_CROSS := spu-
7SPU_CC := $(SPU_CROSS)gcc
8SPU_AS := $(SPU_CROSS)gcc
9SPU_LD := $(SPU_CROSS)ld
10SPU_OBJCOPY := $(SPU_CROSS)objcopy
11SPU_CFLAGS := -O2 -Wall -I$(srctree)/include -I$(objtree)/include2
12SPU_AFLAGS := -c -D__ASSEMBLY__ -I$(srctree)/include -I$(objtree)/include2
13SPU_LDFLAGS := -N -Ttext=0x0
14
5$(obj)/switch.o: $(obj)/spu_save_dump.h $(obj)/spu_restore_dump.h 15$(obj)/switch.o: $(obj)/spu_save_dump.h $(obj)/spu_restore_dump.h
16
17# Compile SPU files
18 cmd_spu_cc = $(SPU_CC) $(SPU_CFLAGS) -c -o $@ $<
19quiet_cmd_spu_cc = SPU_CC $@
20$(obj)/spu_%.o: $(src)/spu_%.c
21 $(call if_changed,spu_cc)
22
23# Assemble SPU files
24 cmd_spu_as = $(SPU_AS) $(SPU_AFLAGS) -o $@ $<
25quiet_cmd_spu_as = SPU_AS $@
26$(obj)/spu_%.o: $(src)/spu_%.S
27 $(call if_changed,spu_as)
28
29# Link SPU Executables
30 cmd_spu_ld = $(SPU_LD) $(SPU_LDFLAGS) -o $@ $^
31quiet_cmd_spu_ld = SPU_LD $@
32$(obj)/spu_%: $(obj)/spu_%_crt0.o $(obj)/spu_%.o
33 $(call if_changed,spu_ld)
34
35# Copy into binary format
36 cmd_spu_objcopy = $(SPU_OBJCOPY) -O binary $< $@
37quiet_cmd_spu_objcopy = OBJCOPY $@
38$(obj)/spu_%.bin: $(src)/spu_%
39 $(call if_changed,spu_objcopy)
40
41# create C code from ELF executable
42cmd_hexdump = ( \
43 echo "/*" ; \
44 echo " * $*_dump.h: Copyright (C) 2005 IBM." ; \
45 echo " * Hex-dump auto generated from $*.c." ; \
46 echo " * Do not edit!" ; \
47 echo " */" ; \
48 echo "static unsigned int $*_code[] __page_aligned = {" ; \
49 hexdump -v -e '4/4 "0x%08x, " "\n"' $< ; \
50 echo "};" ; \
51 ) > $@
52quiet_cmd_hexdump = HEXDUMP $@
53$(obj)/%_dump.h: $(obj)/%.bin
54 $(call if_changed,hexdump)
diff --git a/arch/powerpc/platforms/cell/spufs/spu_restore.c b/arch/powerpc/platforms/cell/spufs/spu_restore.c
new file mode 100644
index 000000000000..0bf723dcd677
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_restore.c
@@ -0,0 +1,336 @@
1/*
2 * spu_restore.c
3 *
4 * (C) Copyright IBM Corp. 2005
5 *
6 * SPU-side context restore sequence outlined in
7 * Synergistic Processor Element Book IV
8 *
9 * Author: Mark Nutter <mnutter@us.ibm.com>
10 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2, or (at your option)
14 * any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 *
25 */
26
27
28#ifndef LS_SIZE
29#define LS_SIZE 0x40000 /* 256K (in bytes) */
30#endif
31
32typedef unsigned int u32;
33typedef unsigned long long u64;
34
35#include <spu_intrinsics.h>
36#include <asm/spu_csa.h>
37#include "spu_utils.h"
38
39#define BR_INSTR 0x327fff80 /* br -4 */
40#define NOP_INSTR 0x40200000 /* nop */
41#define HEQ_INSTR 0x7b000000 /* heq $0, $0 */
42#define STOP_INSTR 0x00000000 /* stop 0x0 */
43#define ILLEGAL_INSTR 0x00800000 /* illegal instr */
44#define RESTORE_COMPLETE 0x00003ffc /* stop 0x3ffc */
45
46static inline void fetch_regs_from_mem(addr64 lscsa_ea)
47{
48 unsigned int ls = (unsigned int)&regs_spill[0];
49 unsigned int size = sizeof(regs_spill);
50 unsigned int tag_id = 0;
51 unsigned int cmd = 0x40; /* GET */
52
53 spu_writech(MFC_LSA, ls);
54 spu_writech(MFC_EAH, lscsa_ea.ui[0]);
55 spu_writech(MFC_EAL, lscsa_ea.ui[1]);
56 spu_writech(MFC_Size, size);
57 spu_writech(MFC_TagID, tag_id);
58 spu_writech(MFC_Cmd, cmd);
59}
60
61static inline void restore_upper_240kb(addr64 lscsa_ea)
62{
63 unsigned int ls = 16384;
64 unsigned int list = (unsigned int)&dma_list[0];
65 unsigned int size = sizeof(dma_list);
66 unsigned int tag_id = 0;
67 unsigned int cmd = 0x44; /* GETL */
68
69 /* Restore, Step 4:
70 * Enqueue the GETL command (tag 0) to the MFC SPU command
71 * queue to transfer the upper 240 kb of LS from CSA.
72 */
73 spu_writech(MFC_LSA, ls);
74 spu_writech(MFC_EAH, lscsa_ea.ui[0]);
75 spu_writech(MFC_EAL, list);
76 spu_writech(MFC_Size, size);
77 spu_writech(MFC_TagID, tag_id);
78 spu_writech(MFC_Cmd, cmd);
79}
80
81static inline void restore_decr(void)
82{
83 unsigned int offset;
84 unsigned int decr_running;
85 unsigned int decr;
86
87 /* Restore, Step 6:
88 * If the LSCSA "decrementer running" flag is set
89 * then write the SPU_WrDec channel with the
90 * decrementer value from LSCSA.
91 */
92 offset = LSCSA_QW_OFFSET(decr_status);
93 decr_running = regs_spill[offset].slot[0];
94 if (decr_running) {
95 offset = LSCSA_QW_OFFSET(decr);
96 decr = regs_spill[offset].slot[0];
97 spu_writech(SPU_WrDec, decr);
98 }
99}
100
101static inline void write_ppu_mb(void)
102{
103 unsigned int offset;
104 unsigned int data;
105
106 /* Restore, Step 11:
107 * Write the MFC_WrOut_MB channel with the PPU_MB
108 * data from LSCSA.
109 */
110 offset = LSCSA_QW_OFFSET(ppu_mb);
111 data = regs_spill[offset].slot[0];
112 spu_writech(SPU_WrOutMbox, data);
113}
114
115static inline void write_ppuint_mb(void)
116{
117 unsigned int offset;
118 unsigned int data;
119
120 /* Restore, Step 12:
121 * Write the MFC_WrInt_MB channel with the PPUINT_MB
122 * data from LSCSA.
123 */
124 offset = LSCSA_QW_OFFSET(ppuint_mb);
125 data = regs_spill[offset].slot[0];
126 spu_writech(SPU_WrOutIntrMbox, data);
127}
128
129static inline void restore_fpcr(void)
130{
131 unsigned int offset;
132 vector unsigned int fpcr;
133
134 /* Restore, Step 13:
135 * Restore the floating-point status and control
136 * register from the LSCSA.
137 */
138 offset = LSCSA_QW_OFFSET(fpcr);
139 fpcr = regs_spill[offset].v;
140 spu_mtfpscr(fpcr);
141}
142
143static inline void restore_srr0(void)
144{
145 unsigned int offset;
146 unsigned int srr0;
147
148 /* Restore, Step 14:
149 * Restore the SPU SRR0 data from the LSCSA.
150 */
151 offset = LSCSA_QW_OFFSET(srr0);
152 srr0 = regs_spill[offset].slot[0];
153 spu_writech(SPU_WrSRR0, srr0);
154}
155
156static inline void restore_event_mask(void)
157{
158 unsigned int offset;
159 unsigned int event_mask;
160
161 /* Restore, Step 15:
162 * Restore the SPU_RdEventMsk data from the LSCSA.
163 */
164 offset = LSCSA_QW_OFFSET(event_mask);
165 event_mask = regs_spill[offset].slot[0];
166 spu_writech(SPU_WrEventMask, event_mask);
167}
168
169static inline void restore_tag_mask(void)
170{
171 unsigned int offset;
172 unsigned int tag_mask;
173
174 /* Restore, Step 16:
175 * Restore the SPU_RdTagMsk data from the LSCSA.
176 */
177 offset = LSCSA_QW_OFFSET(tag_mask);
178 tag_mask = regs_spill[offset].slot[0];
179 spu_writech(MFC_WrTagMask, tag_mask);
180}
181
182static inline void restore_complete(void)
183{
184 extern void exit_fini(void);
185 unsigned int *exit_instrs = (unsigned int *)exit_fini;
186 unsigned int offset;
187 unsigned int stopped_status;
188 unsigned int stopped_code;
189
190 /* Restore, Step 18:
191 * Issue a stop-and-signal instruction with
192 * "good context restore" signal value.
193 *
194 * Restore, Step 19:
195 * There may be additional instructions placed
196 * here by the PPE Sequence for SPU Context
197 * Restore in order to restore the correct
198 * "stopped state".
199 *
200 * This step is handled here by analyzing the
201 * LSCSA.stopped_status and then modifying the
202 * exit() function to behave appropriately.
203 */
204
205 offset = LSCSA_QW_OFFSET(stopped_status);
206 stopped_status = regs_spill[offset].slot[0];
207 stopped_code = regs_spill[offset].slot[1];
208
209 switch (stopped_status) {
210 case SPU_STOPPED_STATUS_P_I:
211 /* SPU_Status[P,I]=1. Add illegal instruction
212 * followed by stop-and-signal instruction after
213 * end of restore code.
214 */
215 exit_instrs[0] = RESTORE_COMPLETE;
216 exit_instrs[1] = ILLEGAL_INSTR;
217 exit_instrs[2] = STOP_INSTR | stopped_code;
218 break;
219 case SPU_STOPPED_STATUS_P_H:
220 /* SPU_Status[P,H]=1. Add 'heq $0, $0' followed
221 * by stop-and-signal instruction after end of
222 * restore code.
223 */
224 exit_instrs[0] = RESTORE_COMPLETE;
225 exit_instrs[1] = HEQ_INSTR;
226 exit_instrs[2] = STOP_INSTR | stopped_code;
227 break;
228 case SPU_STOPPED_STATUS_S_P:
229 /* SPU_Status[S,P]=1. Add nop instruction
230 * followed by 'br -4' after end of restore
231 * code.
232 */
233 exit_instrs[0] = RESTORE_COMPLETE;
234 exit_instrs[1] = STOP_INSTR | stopped_code;
235 exit_instrs[2] = NOP_INSTR;
236 exit_instrs[3] = BR_INSTR;
237 break;
238 case SPU_STOPPED_STATUS_S_I:
239 /* SPU_Status[S,I]=1. Add illegal instruction
240 * followed by 'br -4' after end of restore code.
241 */
242 exit_instrs[0] = RESTORE_COMPLETE;
243 exit_instrs[1] = ILLEGAL_INSTR;
244 exit_instrs[2] = NOP_INSTR;
245 exit_instrs[3] = BR_INSTR;
246 break;
247 case SPU_STOPPED_STATUS_I:
248 /* SPU_Status[I]=1. Add illegal instruction followed
249 * by infinite loop after end of restore sequence.
250 */
251 exit_instrs[0] = RESTORE_COMPLETE;
252 exit_instrs[1] = ILLEGAL_INSTR;
253 exit_instrs[2] = NOP_INSTR;
254 exit_instrs[3] = BR_INSTR;
255 break;
256 case SPU_STOPPED_STATUS_S:
257 /* SPU_Status[S]=1. Add two 'nop' instructions. */
258 exit_instrs[0] = RESTORE_COMPLETE;
259 exit_instrs[1] = NOP_INSTR;
260 exit_instrs[2] = NOP_INSTR;
261 exit_instrs[3] = BR_INSTR;
262 break;
263 case SPU_STOPPED_STATUS_H:
264 /* SPU_Status[H]=1. Add 'heq $0, $0' instruction
265 * after end of restore code.
266 */
267 exit_instrs[0] = RESTORE_COMPLETE;
268 exit_instrs[1] = HEQ_INSTR;
269 exit_instrs[2] = NOP_INSTR;
270 exit_instrs[3] = BR_INSTR;
271 break;
272 case SPU_STOPPED_STATUS_P:
273 /* SPU_Status[P]=1. Add stop-and-signal instruction
274 * after end of restore code.
275 */
276 exit_instrs[0] = RESTORE_COMPLETE;
277 exit_instrs[1] = STOP_INSTR | stopped_code;
278 break;
279 case SPU_STOPPED_STATUS_R:
280 /* SPU_Status[I,S,H,P,R]=0. Add infinite loop. */
281 exit_instrs[0] = RESTORE_COMPLETE;
282 exit_instrs[1] = NOP_INSTR;
283 exit_instrs[2] = NOP_INSTR;
284 exit_instrs[3] = BR_INSTR;
285 break;
286 default:
287 /* SPU_Status[R]=1. No additonal instructions. */
288 break;
289 }
290 spu_sync();
291}
292
293/**
294 * main - entry point for SPU-side context restore.
295 *
296 * This code deviates from the documented sequence in the
297 * following aspects:
298 *
299 * 1. The EA for LSCSA is passed from PPE in the
300 * signal notification channels.
301 * 2. The register spill area is pulled by SPU
302 * into LS, rather than pushed by PPE.
303 * 3. All 128 registers are restored by exit().
304 * 4. The exit() function is modified at run
305 * time in order to properly restore the
306 * SPU_Status register.
307 */
308int main()
309{
310 addr64 lscsa_ea;
311
312 lscsa_ea.ui[0] = spu_readch(SPU_RdSigNotify1);
313 lscsa_ea.ui[1] = spu_readch(SPU_RdSigNotify2);
314 fetch_regs_from_mem(lscsa_ea);
315
316 set_event_mask(); /* Step 1. */
317 set_tag_mask(); /* Step 2. */
318 build_dma_list(lscsa_ea); /* Step 3. */
319 restore_upper_240kb(lscsa_ea); /* Step 4. */
320 /* Step 5: done by 'exit'. */
321 restore_decr(); /* Step 6. */
322 enqueue_putllc(lscsa_ea); /* Step 7. */
323 set_tag_update(); /* Step 8. */
324 read_tag_status(); /* Step 9. */
325 read_llar_status(); /* Step 10. */
326 write_ppu_mb(); /* Step 11. */
327 write_ppuint_mb(); /* Step 12. */
328 restore_fpcr(); /* Step 13. */
329 restore_srr0(); /* Step 14. */
330 restore_event_mask(); /* Step 15. */
331 restore_tag_mask(); /* Step 16. */
332 /* Step 17. done by 'exit'. */
333 restore_complete(); /* Step 18. */
334
335 return 0;
336}
diff --git a/arch/powerpc/platforms/cell/spufs/spu_restore_crt0.S b/arch/powerpc/platforms/cell/spufs/spu_restore_crt0.S
new file mode 100644
index 000000000000..2905949debe1
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_restore_crt0.S
@@ -0,0 +1,116 @@
1/*
2 * crt0_r.S: Entry function for SPU-side context restore.
3 *
4 * Copyright (C) 2005 IBM
5 *
6 * Entry and exit function for SPU-side of the context restore
7 * sequence. Sets up an initial stack frame, then branches to
8 * 'main'. On return, restores all 128 registers from the LSCSA
9 * and exits.
10 *
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License as published by
14 * the Free Software Foundation; either version 2, or (at your option)
15 * any later version.
16 *
17 * This program is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 * GNU General Public License for more details.
21 *
22 * You should have received a copy of the GNU General Public License
23 * along with this program; if not, write to the Free Software
24 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
25 */
26
27#include <asm/spu_csa.h>
28
29.data
30.align 7
31.globl regs_spill
32regs_spill:
33.space SIZEOF_SPU_SPILL_REGS, 0x0
34
35.text
36.global _start
37_start:
38 /* Initialize the stack pointer to point to 16368
39 * (16kb-16). The back chain pointer is initialized
40 * to NULL.
41 */
42 il $0, 0
43 il $SP, 16368
44 stqd $0, 0($SP)
45
46 /* Allocate a minimum stack frame for the called main.
47 * This is needed so that main has a place to save the
48 * link register when it calls another function.
49 */
50 stqd $SP, -160($SP)
51 ai $SP, $SP, -160
52
53 /* Call the program's main function. */
54 brsl $0, main
55
56.global exit
57.global _exit
58exit:
59_exit:
60 /* SPU Context Restore, Step 5: Restore the remaining 112 GPRs. */
61 ila $3, regs_spill + 256
62restore_regs:
63 lqr $4, restore_reg_insts
64restore_reg_loop:
65 ai $4, $4, 4
66 .balignl 16, 0x40200000
67restore_reg_insts: /* must be quad-word aligned. */
68 lqd $16, 0($3)
69 lqd $17, 16($3)
70 lqd $18, 32($3)
71 lqd $19, 48($3)
72 andi $5, $4, 0x7F
73 stqr $4, restore_reg_insts
74 ai $3, $3, 64
75 brnz $5, restore_reg_loop
76
77 /* SPU Context Restore Step 17: Restore the first 16 GPRs. */
78 lqa $0, regs_spill + 0
79 lqa $1, regs_spill + 16
80 lqa $2, regs_spill + 32
81 lqa $3, regs_spill + 48
82 lqa $4, regs_spill + 64
83 lqa $5, regs_spill + 80
84 lqa $6, regs_spill + 96
85 lqa $7, regs_spill + 112
86 lqa $8, regs_spill + 128
87 lqa $9, regs_spill + 144
88 lqa $10, regs_spill + 160
89 lqa $11, regs_spill + 176
90 lqa $12, regs_spill + 192
91 lqa $13, regs_spill + 208
92 lqa $14, regs_spill + 224
93 lqa $15, regs_spill + 240
94
95 /* Under normal circumstances, the 'exit' function
96 * terminates with 'stop SPU_RESTORE_COMPLETE',
97 * indicating that the SPU-side restore code has
98 * completed.
99 *
100 * However it is possible that instructions immediately
101 * following the 'stop 0x3ffc' have been modified at run
102 * time so as to recreate the exact SPU_Status settings
103 * from the application, e.g. illegal instruciton, halt,
104 * etc.
105 */
106.global exit_fini
107.global _exit_fini
108exit_fini:
109_exit_fini:
110 stop SPU_RESTORE_COMPLETE
111 stop 0
112 stop 0
113 stop 0
114
115 /* Pad the size of this crt0.o to be multiple of 16 bytes. */
116.balignl 16, 0x0
diff --git a/arch/powerpc/platforms/cell/spufs/spu_save.c b/arch/powerpc/platforms/cell/spufs/spu_save.c
new file mode 100644
index 000000000000..196033b8a579
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_save.c
@@ -0,0 +1,195 @@
1/*
2 * spu_save.c
3 *
4 * (C) Copyright IBM Corp. 2005
5 *
6 * SPU-side context save sequence outlined in
7 * Synergistic Processor Element Book IV
8 *
9 * Author: Mark Nutter <mnutter@us.ibm.com>
10 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2, or (at your option)
14 * any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 *
25 */
26
27
28#ifndef LS_SIZE
29#define LS_SIZE 0x40000 /* 256K (in bytes) */
30#endif
31
32typedef unsigned int u32;
33typedef unsigned long long u64;
34
35#include <spu_intrinsics.h>
36#include <asm/spu_csa.h>
37#include "spu_utils.h"
38
39static inline void save_event_mask(void)
40{
41 unsigned int offset;
42
43 /* Save, Step 2:
44 * Read the SPU_RdEventMsk channel and save to the LSCSA.
45 */
46 offset = LSCSA_QW_OFFSET(event_mask);
47 regs_spill[offset].slot[0] = spu_readch(SPU_RdEventStatMask);
48}
49
50static inline void save_tag_mask(void)
51{
52 unsigned int offset;
53
54 /* Save, Step 3:
55 * Read the SPU_RdTagMsk channel and save to the LSCSA.
56 */
57 offset = LSCSA_QW_OFFSET(tag_mask);
58 regs_spill[offset].slot[0] = spu_readch(MFC_RdTagMask);
59}
60
61static inline void save_upper_240kb(addr64 lscsa_ea)
62{
63 unsigned int ls = 16384;
64 unsigned int list = (unsigned int)&dma_list[0];
65 unsigned int size = sizeof(dma_list);
66 unsigned int tag_id = 0;
67 unsigned int cmd = 0x24; /* PUTL */
68
69 /* Save, Step 7:
70 * Enqueue the PUTL command (tag 0) to the MFC SPU command
71 * queue to transfer the remaining 240 kb of LS to CSA.
72 */
73 spu_writech(MFC_LSA, ls);
74 spu_writech(MFC_EAH, lscsa_ea.ui[0]);
75 spu_writech(MFC_EAL, list);
76 spu_writech(MFC_Size, size);
77 spu_writech(MFC_TagID, tag_id);
78 spu_writech(MFC_Cmd, cmd);
79}
80
81static inline void save_fpcr(void)
82{
83 // vector unsigned int fpcr;
84 unsigned int offset;
85
86 /* Save, Step 9:
87 * Issue the floating-point status and control register
88 * read instruction, and save to the LSCSA.
89 */
90 offset = LSCSA_QW_OFFSET(fpcr);
91 regs_spill[offset].v = spu_mffpscr();
92}
93
94static inline void save_decr(void)
95{
96 unsigned int offset;
97
98 /* Save, Step 10:
99 * Read and save the SPU_RdDec channel data to
100 * the LSCSA.
101 */
102 offset = LSCSA_QW_OFFSET(decr);
103 regs_spill[offset].slot[0] = spu_readch(SPU_RdDec);
104}
105
106static inline void save_srr0(void)
107{
108 unsigned int offset;
109
110 /* Save, Step 11:
111 * Read and save the SPU_WSRR0 channel data to
112 * the LSCSA.
113 */
114 offset = LSCSA_QW_OFFSET(srr0);
115 regs_spill[offset].slot[0] = spu_readch(SPU_RdSRR0);
116}
117
118static inline void spill_regs_to_mem(addr64 lscsa_ea)
119{
120 unsigned int ls = (unsigned int)&regs_spill[0];
121 unsigned int size = sizeof(regs_spill);
122 unsigned int tag_id = 0;
123 unsigned int cmd = 0x20; /* PUT */
124
125 /* Save, Step 13:
126 * Enqueue a PUT command (tag 0) to send the LSCSA
127 * to the CSA.
128 */
129 spu_writech(MFC_LSA, ls);
130 spu_writech(MFC_EAH, lscsa_ea.ui[0]);
131 spu_writech(MFC_EAL, lscsa_ea.ui[1]);
132 spu_writech(MFC_Size, size);
133 spu_writech(MFC_TagID, tag_id);
134 spu_writech(MFC_Cmd, cmd);
135}
136
137static inline void enqueue_sync(addr64 lscsa_ea)
138{
139 unsigned int tag_id = 0;
140 unsigned int cmd = 0xCC;
141
142 /* Save, Step 14:
143 * Enqueue an MFC_SYNC command (tag 0).
144 */
145 spu_writech(MFC_TagID, tag_id);
146 spu_writech(MFC_Cmd, cmd);
147}
148
149static inline void save_complete(void)
150{
151 /* Save, Step 18:
152 * Issue a stop-and-signal instruction indicating
153 * "save complete". Note: This function will not
154 * return!!
155 */
156 spu_stop(SPU_SAVE_COMPLETE);
157}
158
159/**
160 * main - entry point for SPU-side context save.
161 *
162 * This code deviates from the documented sequence as follows:
163 *
164 * 1. The EA for LSCSA is passed from PPE in the
165 * signal notification channels.
166 * 2. All 128 registers are saved by crt0.o.
167 */
168int main()
169{
170 addr64 lscsa_ea;
171
172 lscsa_ea.ui[0] = spu_readch(SPU_RdSigNotify1);
173 lscsa_ea.ui[1] = spu_readch(SPU_RdSigNotify2);
174
175 /* Step 1: done by exit(). */
176 save_event_mask(); /* Step 2. */
177 save_tag_mask(); /* Step 3. */
178 set_event_mask(); /* Step 4. */
179 set_tag_mask(); /* Step 5. */
180 build_dma_list(lscsa_ea); /* Step 6. */
181 save_upper_240kb(lscsa_ea); /* Step 7. */
182 /* Step 8: done by exit(). */
183 save_fpcr(); /* Step 9. */
184 save_decr(); /* Step 10. */
185 save_srr0(); /* Step 11. */
186 enqueue_putllc(lscsa_ea); /* Step 12. */
187 spill_regs_to_mem(lscsa_ea); /* Step 13. */
188 enqueue_sync(lscsa_ea); /* Step 14. */
189 set_tag_update(); /* Step 15. */
190 read_tag_status(); /* Step 16. */
191 read_llar_status(); /* Step 17. */
192 save_complete(); /* Step 18. */
193
194 return 0;
195}
diff --git a/arch/powerpc/platforms/cell/spufs/spu_save_crt0.S b/arch/powerpc/platforms/cell/spufs/spu_save_crt0.S
new file mode 100644
index 000000000000..6659d6a66faa
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_save_crt0.S
@@ -0,0 +1,102 @@
1/*
2 * crt0_s.S: Entry function for SPU-side context save.
3 *
4 * Copyright (C) 2005 IBM
5 *
6 * Entry function for SPU-side of the context save sequence.
7 * Saves all 128 GPRs, sets up an initial stack frame, then
8 * branches to 'main'.
9 *
10 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2, or (at your option)
14 * any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 */
25
26#include <asm/spu_csa.h>
27
28.data
29.align 7
30.globl regs_spill
31regs_spill:
32.space SIZEOF_SPU_SPILL_REGS, 0x0
33
34.text
35.global _start
36_start:
37 /* SPU Context Save Step 1: Save the first 16 GPRs. */
38 stqa $0, regs_spill + 0
39 stqa $1, regs_spill + 16
40 stqa $2, regs_spill + 32
41 stqa $3, regs_spill + 48
42 stqa $4, regs_spill + 64
43 stqa $5, regs_spill + 80
44 stqa $6, regs_spill + 96
45 stqa $7, regs_spill + 112
46 stqa $8, regs_spill + 128
47 stqa $9, regs_spill + 144
48 stqa $10, regs_spill + 160
49 stqa $11, regs_spill + 176
50 stqa $12, regs_spill + 192
51 stqa $13, regs_spill + 208
52 stqa $14, regs_spill + 224
53 stqa $15, regs_spill + 240
54
55 /* SPU Context Save, Step 8: Save the remaining 112 GPRs. */
56 ila $3, regs_spill + 256
57save_regs:
58 lqr $4, save_reg_insts
59save_reg_loop:
60 ai $4, $4, 4
61 .balignl 16, 0x40200000
62save_reg_insts: /* must be quad-word aligned. */
63 stqd $16, 0($3)
64 stqd $17, 16($3)
65 stqd $18, 32($3)
66 stqd $19, 48($3)
67 andi $5, $4, 0x7F
68 stqr $4, save_reg_insts
69 ai $3, $3, 64
70 brnz $5, save_reg_loop
71
72 /* Initialize the stack pointer to point to 16368
73 * (16kb-16). The back chain pointer is initialized
74 * to NULL.
75 */
76 il $0, 0
77 il $SP, 16368
78 stqd $0, 0($SP)
79
80 /* Allocate a minimum stack frame for the called main.
81 * This is needed so that main has a place to save the
82 * link register when it calls another function.
83 */
84 stqd $SP, -160($SP)
85 ai $SP, $SP, -160
86
87 /* Call the program's main function. */
88 brsl $0, main
89
90 /* In this case main should not return; if it does
91 * there has been an error in the sequence. Execute
92 * stop-and-signal with code=0.
93 */
94.global exit
95.global _exit
96exit:
97_exit:
98 stop 0x0
99
100 /* Pad the size of this crt0.o to be multiple of 16 bytes. */
101.balignl 16, 0x0
102
diff --git a/arch/powerpc/platforms/cell/spufs/spu_utils.h b/arch/powerpc/platforms/cell/spufs/spu_utils.h
new file mode 100644
index 000000000000..58359feb6c95
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/spu_utils.h
@@ -0,0 +1,160 @@
1/*
2 * utils.h: Utilities for SPU-side of the context switch operation.
3 *
4 * (C) Copyright IBM 2005
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2, or (at your option)
9 * any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 */
20
21#ifndef _SPU_CONTEXT_UTILS_H_
22#define _SPU_CONTEXT_UTILS_H_
23
24/*
25 * 64-bit safe EA.
26 */
27typedef union {
28 unsigned long long ull;
29 unsigned int ui[2];
30} addr64;
31
32/*
33 * 128-bit register template.
34 */
35typedef union {
36 unsigned int slot[4];
37 vector unsigned int v;
38} spu_reg128v;
39
40/*
41 * DMA list structure.
42 */
43struct dma_list_elem {
44 unsigned int size;
45 unsigned int ea_low;
46};
47
48/*
49 * Declare storage for 8-byte aligned DMA list.
50 */
51struct dma_list_elem dma_list[15] __attribute__ ((aligned(8)));
52
53/*
54 * External definition for storage
55 * declared in crt0.
56 */
57extern spu_reg128v regs_spill[NR_SPU_SPILL_REGS];
58
59/*
60 * Compute LSCSA byte offset for a given field.
61 */
62static struct spu_lscsa *dummy = (struct spu_lscsa *)0;
63#define LSCSA_BYTE_OFFSET(_field) \
64 ((char *)(&(dummy->_field)) - (char *)(&(dummy->gprs[0].slot[0])))
65#define LSCSA_QW_OFFSET(_field) (LSCSA_BYTE_OFFSET(_field) >> 4)
66
67static inline void set_event_mask(void)
68{
69 unsigned int event_mask = 0;
70
71 /* Save, Step 4:
72 * Restore, Step 1:
73 * Set the SPU_RdEventMsk channel to zero to mask
74 * all events.
75 */
76 spu_writech(SPU_WrEventMask, event_mask);
77}
78
79static inline void set_tag_mask(void)
80{
81 unsigned int tag_mask = 1;
82
83 /* Save, Step 5:
84 * Restore, Step 2:
85 * Set the SPU_WrTagMsk channel to '01' to unmask
86 * only tag group 0.
87 */
88 spu_writech(MFC_WrTagMask, tag_mask);
89}
90
91static inline void build_dma_list(addr64 lscsa_ea)
92{
93 unsigned int ea_low;
94 int i;
95
96 /* Save, Step 6:
97 * Restore, Step 3:
98 * Update the effective address for the CSA in the
99 * pre-canned DMA-list in local storage.
100 */
101 ea_low = lscsa_ea.ui[1];
102 ea_low += LSCSA_BYTE_OFFSET(ls[16384]);
103
104 for (i = 0; i < 15; i++, ea_low += 16384) {
105 dma_list[i].size = 16384;
106 dma_list[i].ea_low = ea_low;
107 }
108}
109
110static inline void enqueue_putllc(addr64 lscsa_ea)
111{
112 unsigned int ls = 0;
113 unsigned int size = 128;
114 unsigned int tag_id = 0;
115 unsigned int cmd = 0xB4; /* PUTLLC */
116
117 /* Save, Step 12:
118 * Restore, Step 7:
119 * Send a PUTLLC (tag 0) command to the MFC using
120 * an effective address in the CSA in order to
121 * remove any possible lock-line reservation.
122 */
123 spu_writech(MFC_LSA, ls);
124 spu_writech(MFC_EAH, lscsa_ea.ui[0]);
125 spu_writech(MFC_EAL, lscsa_ea.ui[1]);
126 spu_writech(MFC_Size, size);
127 spu_writech(MFC_TagID, tag_id);
128 spu_writech(MFC_Cmd, cmd);
129}
130
131static inline void set_tag_update(void)
132{
133 unsigned int update_any = 1;
134
135 /* Save, Step 15:
136 * Restore, Step 8:
137 * Write the MFC_TagUpdate channel with '01'.
138 */
139 spu_writech(MFC_WrTagUpdate, update_any);
140}
141
142static inline void read_tag_status(void)
143{
144 /* Save, Step 16:
145 * Restore, Step 9:
146 * Read the MFC_TagStat channel data.
147 */
148 spu_readch(MFC_RdTagStat);
149}
150
151static inline void read_llar_status(void)
152{
153 /* Save, Step 17:
154 * Restore, Step 10:
155 * Read the MFC_AtomicStat channel data.
156 */
157 spu_readch(MFC_RdAtomicStat);
158}
159
160#endif /* _SPU_CONTEXT_UTILS_H_ */
generated by cgit v1.2.2 (git 2.25.0) at 2024-09-30 20:18:58 -0400