Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Linux/PA-RISC Project (http://www.parisc-linux.org/) | |
3 | * | |
dde39798 HD |
4 | * System call entry code / Linux gateway page |
5 | * Copyright (c) Matthew Wilcox 1999 <willy@bofh.ai> | |
1da177e4 LT |
6 | * Licensed under the GNU GPL. |
7 | * thanks to Philipp Rumpf, Mike Shaver and various others | |
8 | * sorry about the wall, puffin.. | |
9 | */ | |
10 | ||
dde39798 HD |
11 | /* |
12 | How does the Linux gateway page on PA-RISC work? | |
13 | ------------------------------------------------ | |
14 | The Linux gateway page on PA-RISC is "special". | |
15 | It actually has PAGE_GATEWAY bits set (this is linux terminology; in parisc | |
16 | terminology it's Execute, promote to PL0) in the page map. So anything | |
17 | executing on this page executes with kernel level privilege (there's more to it | |
18 | than that: to have this happen, you also have to use a branch with a ,gate | |
19 | completer to activate the privilege promotion). The upshot is that everything | |
20 | that runs on the gateway page runs at kernel privilege but with the current | |
21 | user process address space (although you have access to kernel space via %sr2). | |
22 | For the 0x100 syscall entry, we redo the space registers to point to the kernel | |
23 | address space (preserving the user address space in %sr3), move to wide mode if | |
24 | required, save the user registers and branch into the kernel syscall entry | |
25 | point. For all the other functions, we execute at kernel privilege but don't | |
26 | flip address spaces. The basic upshot of this is that these code snippets are | |
27 | executed atomically (because the kernel can't be pre-empted) and they may | |
28 | perform architecturally forbidden (to PL3) operations (like setting control | |
29 | registers). | |
30 | */ | |
31 | ||
32 | ||
0013a854 | 33 | #include <asm/asm-offsets.h> |
1da177e4 LT |
34 | #include <asm/unistd.h> |
35 | #include <asm/errno.h> | |
1c593571 | 36 | #include <asm/page.h> |
1da177e4 LT |
37 | #include <asm/psw.h> |
38 | #include <asm/thread_info.h> | |
1da177e4 LT |
39 | #include <asm/assembly.h> |
40 | #include <asm/processor.h> | |
6a45716a | 41 | #include <asm/cache.h> |
1da177e4 | 42 | |
8e9e9844 HD |
43 | #include <linux/linkage.h> |
44 | ||
1da177e4 LT |
45 | /* We fill the empty parts of the gateway page with |
46 | * something that will kill the kernel or a | |
47 | * userspace application. | |
48 | */ | |
49 | #define KILL_INSN break 0,0 | |
50 | ||
0b3d643f | 51 | .level LEVEL |
8e9e9844 | 52 | |
dfcf753b | 53 | .text |
1da177e4 LT |
54 | |
55 | .import syscall_exit,code | |
56 | .import syscall_exit_rfi,code | |
1da177e4 LT |
57 | |
58 | /* Linux gateway page is aliased to virtual page 0 in the kernel | |
59 | * address space. Since it is a gateway page it cannot be | |
60 | * dereferenced, so null pointers will still fault. We start | |
61 | * the actual entry point at 0x100. We put break instructions | |
62 | * at the beginning of the page to trap null indirect function | |
63 | * pointers. | |
64 | */ | |
65 | ||
1c593571 | 66 | .align PAGE_SIZE |
8e9e9844 | 67 | ENTRY(linux_gateway_page) |
1da177e4 LT |
68 | |
69 | /* ADDRESS 0x00 to 0xb0 = 176 bytes / 4 bytes per insn = 44 insns */ | |
70 | .rept 44 | |
71 | KILL_INSN | |
72 | .endr | |
73 | ||
f4c0346c | 74 | /* ADDRESS 0xb0 to 0xb8, lws uses two insns for entry */ |
1da177e4 LT |
75 | /* Light-weight-syscall entry must always be located at 0xb0 */ |
76 | /* WARNING: Keep this number updated with table size changes */ | |
89206491 | 77 | #define __NR_lws_entries (3) |
1da177e4 LT |
78 | |
79 | lws_entry: | |
f4c0346c JDA |
80 | gate lws_start, %r0 /* increase privilege */ |
81 | depi 3, 31, 2, %r31 /* Ensure we return into user mode. */ | |
1da177e4 | 82 | |
f4c0346c JDA |
83 | /* Fill from 0xb8 to 0xe0 */ |
84 | .rept 10 | |
1da177e4 LT |
85 | KILL_INSN |
86 | .endr | |
87 | ||
88 | /* This function MUST be located at 0xe0 for glibc's threading | |
89 | mechanism to work. DO NOT MOVE THIS CODE EVER! */ | |
90 | set_thread_pointer: | |
91 | gate .+8, %r0 /* increase privilege */ | |
92 | depi 3, 31, 2, %r31 /* Ensure we return into user mode. */ | |
93 | be 0(%sr7,%r31) /* return to user space */ | |
94 | mtctl %r26, %cr27 /* move arg0 to the control register */ | |
95 | ||
96 | /* Increase the chance of trapping if random jumps occur to this | |
97 | address, fill from 0xf0 to 0x100 */ | |
98 | .rept 4 | |
99 | KILL_INSN | |
100 | .endr | |
101 | ||
102 | /* This address must remain fixed at 0x100 for glibc's syscalls to work */ | |
103 | .align 256 | |
104 | linux_gateway_entry: | |
105 | gate .+8, %r0 /* become privileged */ | |
106 | mtsp %r0,%sr4 /* get kernel space into sr4 */ | |
107 | mtsp %r0,%sr5 /* get kernel space into sr5 */ | |
108 | mtsp %r0,%sr6 /* get kernel space into sr6 */ | |
109 | mfsp %sr7,%r1 /* save user sr7 */ | |
110 | mtsp %r1,%sr3 /* and store it in sr3 */ | |
111 | ||
413059f2 | 112 | #ifdef CONFIG_64BIT |
1da177e4 LT |
113 | /* for now we can *always* set the W bit on entry to the syscall |
114 | * since we don't support wide userland processes. We could | |
115 | * also save the current SM other than in r0 and restore it on | |
116 | * exit from the syscall, and also use that value to know | |
117 | * whether to do narrow or wide syscalls. -PB | |
118 | */ | |
119 | ssm PSW_SM_W, %r1 | |
120 | extrd,u %r1,PSW_W_BIT,1,%r1 | |
121 | /* sp must be aligned on 4, so deposit the W bit setting into | |
122 | * the bottom of sp temporarily */ | |
123 | or,ev %r1,%r30,%r30 | |
124 | b,n 1f | |
125 | /* The top halves of argument registers must be cleared on syscall | |
126 | * entry from narrow executable. | |
127 | */ | |
128 | depdi 0, 31, 32, %r26 | |
129 | depdi 0, 31, 32, %r25 | |
130 | depdi 0, 31, 32, %r24 | |
131 | depdi 0, 31, 32, %r23 | |
132 | depdi 0, 31, 32, %r22 | |
133 | depdi 0, 31, 32, %r21 | |
134 | 1: | |
135 | #endif | |
136 | mfctl %cr30,%r1 | |
137 | xor %r1,%r30,%r30 /* ye olde xor trick */ | |
138 | xor %r1,%r30,%r1 | |
139 | xor %r1,%r30,%r30 | |
140 | ||
141 | ldo THREAD_SZ_ALGN+FRAME_SIZE(%r30),%r30 /* set up kernel stack */ | |
142 | ||
143 | /* N.B.: It is critical that we don't set sr7 to 0 until r30 | |
144 | * contains a valid kernel stack pointer. It is also | |
145 | * critical that we don't start using the kernel stack | |
146 | * until after sr7 has been set to 0. | |
147 | */ | |
148 | ||
149 | mtsp %r0,%sr7 /* get kernel space into sr7 */ | |
150 | STREGM %r1,FRAME_SIZE(%r30) /* save r1 (usp) here for now */ | |
151 | mfctl %cr30,%r1 /* get task ptr in %r1 */ | |
152 | LDREG TI_TASK(%r1),%r1 | |
153 | ||
154 | /* Save some registers for sigcontext and potential task | |
155 | switch (see entry.S for the details of which ones are | |
156 | saved/restored). TASK_PT_PSW is zeroed so we can see whether | |
157 | a process is on a syscall or not. For an interrupt the real | |
158 | PSW value is stored. This is needed for gdb and sys_ptrace. */ | |
159 | STREG %r0, TASK_PT_PSW(%r1) | |
160 | STREG %r2, TASK_PT_GR2(%r1) /* preserve rp */ | |
161 | STREG %r19, TASK_PT_GR19(%r1) | |
162 | ||
163 | LDREGM -FRAME_SIZE(%r30), %r2 /* get users sp back */ | |
413059f2 | 164 | #ifdef CONFIG_64BIT |
1da177e4 LT |
165 | extrd,u %r2,63,1,%r19 /* W hidden in bottom bit */ |
166 | #if 0 | |
167 | xor %r19,%r2,%r2 /* clear bottom bit */ | |
168 | depd,z %r19,1,1,%r19 | |
169 | std %r19,TASK_PT_PSW(%r1) | |
170 | #endif | |
171 | #endif | |
172 | STREG %r2, TASK_PT_GR30(%r1) /* ... and save it */ | |
173 | ||
aa0eecb0 | 174 | STREG %r20, TASK_PT_GR20(%r1) /* Syscall number */ |
1da177e4 LT |
175 | STREG %r21, TASK_PT_GR21(%r1) |
176 | STREG %r22, TASK_PT_GR22(%r1) | |
177 | STREG %r23, TASK_PT_GR23(%r1) /* 4th argument */ | |
178 | STREG %r24, TASK_PT_GR24(%r1) /* 3rd argument */ | |
179 | STREG %r25, TASK_PT_GR25(%r1) /* 2nd argument */ | |
180 | STREG %r26, TASK_PT_GR26(%r1) /* 1st argument */ | |
181 | STREG %r27, TASK_PT_GR27(%r1) /* user dp */ | |
182 | STREG %r28, TASK_PT_GR28(%r1) /* return value 0 */ | |
00df111e | 183 | STREG %r0, TASK_PT_ORIG_R28(%r1) /* don't prohibit restarts */ |
1da177e4 LT |
184 | STREG %r29, TASK_PT_GR29(%r1) /* return value 1 */ |
185 | STREG %r31, TASK_PT_GR31(%r1) /* preserve syscall return ptr */ | |
186 | ||
187 | ldo TASK_PT_FR0(%r1), %r27 /* save fpregs from the kernel */ | |
188 | save_fp %r27 /* or potential task switch */ | |
189 | ||
190 | mfctl %cr11, %r27 /* i.e. SAR */ | |
191 | STREG %r27, TASK_PT_SAR(%r1) | |
192 | ||
193 | loadgp | |
194 | ||
413059f2 | 195 | #ifdef CONFIG_64BIT |
1da177e4 LT |
196 | ldo -16(%r30),%r29 /* Reference param save area */ |
197 | copy %r19,%r2 /* W bit back to r2 */ | |
198 | #else | |
199 | /* no need to save these on stack in wide mode because the first 8 | |
200 | * args are passed in registers */ | |
201 | stw %r22, -52(%r30) /* 5th argument */ | |
202 | stw %r21, -56(%r30) /* 6th argument */ | |
203 | #endif | |
204 | ||
205 | /* Are we being ptraced? */ | |
206 | mfctl %cr30, %r1 | |
64482bd8 AV |
207 | LDREG TI_FLAGS(%r1),%r1 |
208 | ldi _TIF_SYSCALL_TRACE_MASK, %r19 | |
209 | and,COND(=) %r1, %r19, %r0 | |
210 | b,n .Ltracesys | |
1da177e4 LT |
211 | |
212 | /* Note! We cannot use the syscall table that is mapped | |
213 | nearby since the gateway page is mapped execute-only. */ | |
214 | ||
413059f2 | 215 | #ifdef CONFIG_64BIT |
1da177e4 LT |
216 | ldil L%sys_call_table, %r1 |
217 | or,= %r2,%r2,%r2 | |
218 | addil L%(sys_call_table64-sys_call_table), %r1 | |
219 | ldo R%sys_call_table(%r1), %r19 | |
220 | or,= %r2,%r2,%r2 | |
221 | ldo R%sys_call_table64(%r1), %r19 | |
222 | #else | |
223 | ldil L%sys_call_table, %r1 | |
224 | ldo R%sys_call_table(%r1), %r19 | |
225 | #endif | |
3bb457af | 226 | comiclr,>> __NR_Linux_syscalls, %r20, %r0 |
1da177e4 LT |
227 | b,n .Lsyscall_nosys |
228 | ||
229 | LDREGX %r20(%r19), %r19 | |
230 | ||
231 | /* If this is a sys_rt_sigreturn call, and the signal was received | |
232 | * when not in_syscall, then we want to return via syscall_exit_rfi, | |
233 | * not syscall_exit. Signal no. in r20, in_syscall in r25 (see | |
234 | * trampoline code in signal.c). | |
235 | */ | |
236 | ldi __NR_rt_sigreturn,%r2 | |
237 | comb,= %r2,%r20,.Lrt_sigreturn | |
238 | .Lin_syscall: | |
239 | ldil L%syscall_exit,%r2 | |
240 | be 0(%sr7,%r19) | |
241 | ldo R%syscall_exit(%r2),%r2 | |
242 | .Lrt_sigreturn: | |
243 | comib,<> 0,%r25,.Lin_syscall | |
244 | ldil L%syscall_exit_rfi,%r2 | |
245 | be 0(%sr7,%r19) | |
246 | ldo R%syscall_exit_rfi(%r2),%r2 | |
247 | ||
248 | /* Note! Because we are not running where we were linked, any | |
249 | calls to functions external to this file must be indirect. To | |
250 | be safe, we apply the opposite rule to functions within this | |
251 | file, with local labels given to them to ensure correctness. */ | |
252 | ||
253 | .Lsyscall_nosys: | |
254 | syscall_nosys: | |
255 | ldil L%syscall_exit,%r1 | |
256 | be R%syscall_exit(%sr7,%r1) | |
257 | ldo -ENOSYS(%r0),%r28 /* set errno */ | |
258 | ||
259 | ||
260 | /* Warning! This trace code is a virtual duplicate of the code above so be | |
261 | * sure to maintain both! */ | |
262 | .Ltracesys: | |
263 | tracesys: | |
264 | /* Need to save more registers so the debugger can see where we | |
265 | * are. This saves only the lower 8 bits of PSW, so that the C | |
266 | * bit is still clear on syscalls, and the D bit is set if this | |
267 | * full register save path has been executed. We check the D | |
268 | * bit on syscall_return_rfi to determine which registers to | |
269 | * restore. An interrupt results in a full PSW saved with the | |
270 | * C bit set, a non-straced syscall entry results in C and D clear | |
271 | * in the saved PSW. | |
272 | */ | |
273 | ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */ | |
274 | LDREG TI_TASK(%r1), %r1 | |
275 | ssm 0,%r2 | |
276 | STREG %r2,TASK_PT_PSW(%r1) /* Lower 8 bits only!! */ | |
277 | mfsp %sr0,%r2 | |
278 | STREG %r2,TASK_PT_SR0(%r1) | |
279 | mfsp %sr1,%r2 | |
280 | STREG %r2,TASK_PT_SR1(%r1) | |
281 | mfsp %sr2,%r2 | |
282 | STREG %r2,TASK_PT_SR2(%r1) | |
283 | mfsp %sr3,%r2 | |
284 | STREG %r2,TASK_PT_SR3(%r1) | |
285 | STREG %r2,TASK_PT_SR4(%r1) | |
286 | STREG %r2,TASK_PT_SR5(%r1) | |
287 | STREG %r2,TASK_PT_SR6(%r1) | |
288 | STREG %r2,TASK_PT_SR7(%r1) | |
289 | STREG %r2,TASK_PT_IASQ0(%r1) | |
290 | STREG %r2,TASK_PT_IASQ1(%r1) | |
291 | LDREG TASK_PT_GR31(%r1),%r2 | |
292 | STREG %r2,TASK_PT_IAOQ0(%r1) | |
293 | ldo 4(%r2),%r2 | |
294 | STREG %r2,TASK_PT_IAOQ1(%r1) | |
295 | ldo TASK_REGS(%r1),%r2 | |
296 | /* reg_save %r2 */ | |
297 | STREG %r3,PT_GR3(%r2) | |
298 | STREG %r4,PT_GR4(%r2) | |
299 | STREG %r5,PT_GR5(%r2) | |
300 | STREG %r6,PT_GR6(%r2) | |
301 | STREG %r7,PT_GR7(%r2) | |
302 | STREG %r8,PT_GR8(%r2) | |
303 | STREG %r9,PT_GR9(%r2) | |
304 | STREG %r10,PT_GR10(%r2) | |
305 | STREG %r11,PT_GR11(%r2) | |
306 | STREG %r12,PT_GR12(%r2) | |
307 | STREG %r13,PT_GR13(%r2) | |
308 | STREG %r14,PT_GR14(%r2) | |
309 | STREG %r15,PT_GR15(%r2) | |
310 | STREG %r16,PT_GR16(%r2) | |
311 | STREG %r17,PT_GR17(%r2) | |
312 | STREG %r18,PT_GR18(%r2) | |
313 | /* Finished saving things for the debugger */ | |
314 | ||
2798af1a KM |
315 | copy %r2,%r26 |
316 | ldil L%do_syscall_trace_enter,%r1 | |
1da177e4 | 317 | ldil L%tracesys_next,%r2 |
2798af1a | 318 | be R%do_syscall_trace_enter(%sr7,%r1) |
1da177e4 LT |
319 | ldo R%tracesys_next(%r2),%r2 |
320 | ||
2798af1a KM |
321 | tracesys_next: |
322 | /* do_syscall_trace_enter either returned the syscallno, or -1L, | |
323 | * so we skip restoring the PT_GR20 below, since we pulled it from | |
324 | * task->thread.regs.gr[20] above. | |
325 | */ | |
326 | copy %ret0,%r20 | |
1da177e4 LT |
327 | ldil L%sys_call_table,%r1 |
328 | ldo R%sys_call_table(%r1), %r19 | |
329 | ||
330 | ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */ | |
331 | LDREG TI_TASK(%r1), %r1 | |
910cd32e | 332 | LDREG TASK_PT_GR28(%r1), %r28 /* Restore return value */ |
1da177e4 LT |
333 | LDREG TASK_PT_GR26(%r1), %r26 /* Restore the users args */ |
334 | LDREG TASK_PT_GR25(%r1), %r25 | |
335 | LDREG TASK_PT_GR24(%r1), %r24 | |
336 | LDREG TASK_PT_GR23(%r1), %r23 | |
1da177e4 LT |
337 | LDREG TASK_PT_GR22(%r1), %r22 |
338 | LDREG TASK_PT_GR21(%r1), %r21 | |
52ab532e | 339 | #ifdef CONFIG_64BIT |
1da177e4 | 340 | ldo -16(%r30),%r29 /* Reference param save area */ |
52ab532e AV |
341 | #else |
342 | stw %r22, -52(%r30) /* 5th argument */ | |
343 | stw %r21, -56(%r30) /* 6th argument */ | |
1da177e4 LT |
344 | #endif |
345 | ||
910cd32e | 346 | cmpib,COND(=),n -1,%r20,tracesys_exit /* seccomp may have returned -1 */ |
f0b22d1b | 347 | comiclr,>> __NR_Linux_syscalls, %r20, %r0 |
98e8b6c9 | 348 | b,n .Ltracesys_nosys |
1da177e4 LT |
349 | |
350 | LDREGX %r20(%r19), %r19 | |
351 | ||
352 | /* If this is a sys_rt_sigreturn call, and the signal was received | |
353 | * when not in_syscall, then we want to return via syscall_exit_rfi, | |
354 | * not syscall_exit. Signal no. in r20, in_syscall in r25 (see | |
355 | * trampoline code in signal.c). | |
356 | */ | |
357 | ldi __NR_rt_sigreturn,%r2 | |
358 | comb,= %r2,%r20,.Ltrace_rt_sigreturn | |
359 | .Ltrace_in_syscall: | |
360 | ldil L%tracesys_exit,%r2 | |
361 | be 0(%sr7,%r19) | |
362 | ldo R%tracesys_exit(%r2),%r2 | |
363 | ||
98e8b6c9 HD |
364 | .Ltracesys_nosys: |
365 | ldo -ENOSYS(%r0),%r28 /* set errno */ | |
366 | ||
1da177e4 LT |
367 | /* Do *not* call this function on the gateway page, because it |
368 | makes a direct call to syscall_trace. */ | |
369 | ||
370 | tracesys_exit: | |
371 | ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */ | |
372 | LDREG TI_TASK(%r1), %r1 | |
413059f2 | 373 | #ifdef CONFIG_64BIT |
1da177e4 LT |
374 | ldo -16(%r30),%r29 /* Reference param save area */ |
375 | #endif | |
2798af1a | 376 | ldo TASK_REGS(%r1),%r26 |
337685e5 | 377 | BL do_syscall_trace_exit,%r2 |
1da177e4 LT |
378 | STREG %r28,TASK_PT_GR28(%r1) /* save return value now */ |
379 | ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */ | |
380 | LDREG TI_TASK(%r1), %r1 | |
381 | LDREG TASK_PT_GR28(%r1), %r28 /* Restore return val. */ | |
382 | ||
383 | ldil L%syscall_exit,%r1 | |
384 | be,n R%syscall_exit(%sr7,%r1) | |
385 | ||
386 | .Ltrace_rt_sigreturn: | |
387 | comib,<> 0,%r25,.Ltrace_in_syscall | |
388 | ldil L%tracesys_sigexit,%r2 | |
389 | be 0(%sr7,%r19) | |
390 | ldo R%tracesys_sigexit(%r2),%r2 | |
391 | ||
392 | tracesys_sigexit: | |
393 | ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */ | |
5837d42f | 394 | LDREG TI_TASK(%r1), %r1 |
413059f2 | 395 | #ifdef CONFIG_64BIT |
1da177e4 LT |
396 | ldo -16(%r30),%r29 /* Reference param save area */ |
397 | #endif | |
337685e5 | 398 | BL do_syscall_trace_exit,%r2 |
2798af1a | 399 | ldo TASK_REGS(%r1),%r26 |
1da177e4 LT |
400 | |
401 | ldil L%syscall_exit_rfi,%r1 | |
402 | be,n R%syscall_exit_rfi(%sr7,%r1) | |
403 | ||
404 | ||
405 | /********************************************************* | |
c84c3a69 | 406 | 32/64-bit Light-Weight-Syscall ABI |
1da177e4 | 407 | |
c84c3a69 HD |
408 | * - Indicates a hint for userspace inline asm |
409 | implementations. | |
1da177e4 | 410 | |
c84c3a69 HD |
411 | Syscall number (caller-saves) |
412 | - %r20 | |
413 | * In asm clobber. | |
1da177e4 | 414 | |
c84c3a69 HD |
415 | Argument registers (caller-saves) |
416 | - %r26, %r25, %r24, %r23, %r22 | |
417 | * In asm input. | |
418 | ||
419 | Return registers (caller-saves) | |
420 | - %r28 (return), %r21 (errno) | |
421 | * In asm output. | |
422 | ||
423 | Caller-saves registers | |
424 | - %r1, %r27, %r29 | |
425 | - %r2 (return pointer) | |
426 | - %r31 (ble link register) | |
427 | * In asm clobber. | |
428 | ||
429 | Callee-saves registers | |
430 | - %r3-%r18 | |
431 | - %r30 (stack pointer) | |
432 | * Not in asm clobber. | |
433 | ||
434 | If userspace is 32-bit: | |
435 | Callee-saves registers | |
436 | - %r19 (32-bit PIC register) | |
437 | ||
438 | Differences from 32-bit calling convention: | |
439 | - Syscall number in %r20 | |
440 | - Additional argument register %r22 (arg4) | |
441 | - Callee-saves %r19. | |
442 | ||
443 | If userspace is 64-bit: | |
444 | Callee-saves registers | |
445 | - %r27 (64-bit PIC register) | |
446 | ||
447 | Differences from 64-bit calling convention: | |
448 | - Syscall number in %r20 | |
449 | - Additional argument register %r22 (arg4) | |
450 | - Callee-saves %r27. | |
1da177e4 LT |
451 | |
452 | Error codes returned by entry path: | |
453 | ||
454 | ENOSYS - r20 was an invalid LWS number. | |
455 | ||
456 | *********************************************************/ | |
457 | lws_start: | |
1da177e4 | 458 | |
413059f2 | 459 | #ifdef CONFIG_64BIT |
1da177e4 LT |
460 | /* FIXME: If we are a 64-bit kernel just |
461 | * turn this on unconditionally. | |
462 | */ | |
463 | ssm PSW_SM_W, %r1 | |
464 | extrd,u %r1,PSW_W_BIT,1,%r1 | |
465 | /* sp must be aligned on 4, so deposit the W bit setting into | |
466 | * the bottom of sp temporarily */ | |
467 | or,ev %r1,%r30,%r30 | |
468 | ||
469 | /* Clip LWS number to a 32-bit value always */ | |
470 | depdi 0, 31, 32, %r20 | |
471 | #endif | |
472 | ||
473 | /* Is the lws entry number valid? */ | |
f4c0346c | 474 | comiclr,>> __NR_lws_entries, %r20, %r0 |
1da177e4 LT |
475 | b,n lws_exit_nosys |
476 | ||
477 | /* WARNING: Trashing sr2 and sr3 */ | |
478 | mfsp %sr7,%r1 /* get userspace into sr3 */ | |
479 | mtsp %r1,%sr3 | |
480 | mtsp %r0,%sr2 /* get kernel space into sr2 */ | |
481 | ||
482 | /* Load table start */ | |
483 | ldil L%lws_table, %r1 | |
484 | ldo R%lws_table(%r1), %r28 /* Scratch use of r28 */ | |
485 | LDREGX %r20(%sr2,r28), %r21 /* Scratch use of r21 */ | |
486 | ||
487 | /* Jump to lws, lws table pointers already relocated */ | |
488 | be,n 0(%sr2,%r21) | |
489 | ||
490 | lws_exit_nosys: | |
491 | ldo -ENOSYS(%r0),%r21 /* set errno */ | |
492 | /* Fall through: Return to userspace */ | |
493 | ||
494 | lws_exit: | |
413059f2 | 495 | #ifdef CONFIG_64BIT |
1da177e4 LT |
496 | /* decide whether to reset the wide mode bit |
497 | * | |
498 | * For a syscall, the W bit is stored in the lowest bit | |
499 | * of sp. Extract it and reset W if it is zero */ | |
500 | extrd,u,*<> %r30,63,1,%r1 | |
501 | rsm PSW_SM_W, %r0 | |
502 | /* now reset the lowest bit of sp if it was set */ | |
503 | xor %r30,%r1,%r30 | |
504 | #endif | |
f4c0346c | 505 | be,n 0(%sr7, %r31) |
1da177e4 LT |
506 | |
507 | ||
508 | ||
509 | /*************************************************** | |
89206491 | 510 | Implementing 32bit CAS as an atomic operation: |
1da177e4 LT |
511 | |
512 | %r26 - Address to examine | |
513 | %r25 - Old value to check (old) | |
514 | %r24 - New value to set (new) | |
515 | %r28 - Return prev through this register. | |
516 | %r21 - Kernel error code | |
517 | ||
518 | If debugging is DISabled: | |
519 | ||
520 | %r21 has the following meanings: | |
521 | ||
522 | EAGAIN - CAS is busy, ldcw failed, try again. | |
523 | EFAULT - Read or write failed. | |
524 | ||
525 | If debugging is enabled: | |
526 | ||
527 | EDEADLOCK - CAS called recursively. | |
528 | EAGAIN && r28 == 1 - CAS is busy. Lock contended. | |
529 | EAGAIN && r28 == 2 - CAS is busy. ldcw failed. | |
530 | EFAULT - Read or write failed. | |
531 | ||
532 | Scratch: r20, r28, r1 | |
533 | ||
534 | ****************************************************/ | |
535 | ||
536 | /* Do not enable LWS debugging */ | |
537 | #define ENABLE_LWS_DEBUG 0 | |
538 | ||
539 | /* ELF64 Process entry path */ | |
540 | lws_compare_and_swap64: | |
413059f2 | 541 | #ifdef CONFIG_64BIT |
1da177e4 LT |
542 | b,n lws_compare_and_swap |
543 | #else | |
544 | /* If we are not a 64-bit kernel, then we don't | |
c84c3a69 HD |
545 | * have 64-bit input registers, and calling |
546 | * the 64-bit LWS CAS returns ENOSYS. | |
1da177e4 LT |
547 | */ |
548 | b,n lws_exit_nosys | |
549 | #endif | |
550 | ||
551 | /* ELF32 Process entry path */ | |
552 | lws_compare_and_swap32: | |
413059f2 | 553 | #ifdef CONFIG_64BIT |
1da177e4 LT |
554 | /* Clip all the input registers */ |
555 | depdi 0, 31, 32, %r26 | |
556 | depdi 0, 31, 32, %r25 | |
557 | depdi 0, 31, 32, %r24 | |
558 | #endif | |
559 | ||
560 | lws_compare_and_swap: | |
1da177e4 LT |
561 | /* Load start of lock table */ |
562 | ldil L%lws_lock_start, %r20 | |
563 | ldo R%lws_lock_start(%r20), %r28 | |
564 | ||
565 | /* Extract four bits from r26 and hash lock (Bits 4-7) */ | |
566 | extru %r26, 27, 4, %r20 | |
567 | ||
568 | /* Find lock to use, the hash is either one of 0 to | |
569 | 15, multiplied by 16 (keep it 16-byte aligned) | |
570 | and add to the lock table offset. */ | |
571 | shlw %r20, 4, %r20 | |
572 | add %r20, %r28, %r20 | |
573 | ||
b5e8b733 | 574 | # if ENABLE_LWS_DEBUG |
1da177e4 LT |
575 | /* |
576 | DEBUG, check for deadlock! | |
577 | If the thread register values are the same | |
578 | then we were the one that locked it last and | |
579 | this is a recurisve call that will deadlock. | |
580 | We *must* giveup this call and fail. | |
581 | */ | |
582 | ldw 4(%sr2,%r20), %r28 /* Load thread register */ | |
aa0eecb0 | 583 | /* WARNING: If cr27 cycles to the same value we have problems */ |
1da177e4 LT |
584 | mfctl %cr27, %r21 /* Get current thread register */ |
585 | cmpb,<>,n %r21, %r28, cas_lock /* Called recursive? */ | |
586 | b lws_exit /* Return error! */ | |
587 | ldo -EDEADLOCK(%r0), %r21 | |
588 | cas_lock: | |
589 | cmpb,=,n %r0, %r28, cas_nocontend /* Is nobody using it? */ | |
590 | ldo 1(%r0), %r28 /* 1st case */ | |
591 | b lws_exit /* Contended... */ | |
592 | ldo -EAGAIN(%r0), %r21 /* Spin in userspace */ | |
593 | cas_nocontend: | |
594 | # endif | |
595 | /* ENABLE_LWS_DEBUG */ | |
596 | ||
c776cd89 JDA |
597 | rsm PSW_SM_I, %r0 /* Disable interrupts */ |
598 | /* COW breaks can cause contention on UP systems */ | |
64f49532 | 599 | LDCW 0(%sr2,%r20), %r28 /* Try to acquire the lock */ |
1da177e4 LT |
600 | cmpb,<>,n %r0, %r28, cas_action /* Did we get it? */ |
601 | cas_wouldblock: | |
602 | ldo 2(%r0), %r28 /* 2nd case */ | |
c776cd89 | 603 | ssm PSW_SM_I, %r0 |
1da177e4 LT |
604 | b lws_exit /* Contended... */ |
605 | ldo -EAGAIN(%r0), %r21 /* Spin in userspace */ | |
1da177e4 LT |
606 | |
607 | /* | |
608 | prev = *addr; | |
609 | if ( prev == old ) | |
610 | *addr = new; | |
611 | return prev; | |
612 | */ | |
613 | ||
614 | /* NOTES: | |
615 | This all works becuse intr_do_signal | |
616 | and schedule both check the return iasq | |
617 | and see that we are on the kernel page | |
618 | so this process is never scheduled off | |
619 | or is ever sent any signal of any sort, | |
620 | thus it is wholly atomic from usrspaces | |
621 | perspective | |
622 | */ | |
623 | cas_action: | |
b5e8b733 | 624 | #if defined CONFIG_SMP && ENABLE_LWS_DEBUG |
1da177e4 LT |
625 | /* DEBUG */ |
626 | mfctl %cr27, %r1 | |
627 | stw %r1, 4(%sr2,%r20) | |
628 | #endif | |
629 | /* The load and store could fail */ | |
c776cd89 | 630 | 1: ldw,ma 0(%sr3,%r26), %r28 |
1da177e4 | 631 | sub,<> %r28, %r25, %r0 |
c776cd89 | 632 | 2: stw,ma %r24, 0(%sr3,%r26) |
1da177e4 | 633 | /* Free lock */ |
c776cd89 | 634 | stw,ma %r20, 0(%sr2,%r20) |
f4c0346c | 635 | #if ENABLE_LWS_DEBUG |
1da177e4 LT |
636 | /* Clear thread register indicator */ |
637 | stw %r0, 4(%sr2,%r20) | |
1da177e4 | 638 | #endif |
c776cd89 JDA |
639 | /* Enable interrupts */ |
640 | ssm PSW_SM_I, %r0 | |
1da177e4 LT |
641 | /* Return to userspace, set no error */ |
642 | b lws_exit | |
643 | copy %r0, %r21 | |
644 | ||
645 | 3: | |
25985edc | 646 | /* Error occurred on load or store */ |
1da177e4 LT |
647 | /* Free lock */ |
648 | stw %r20, 0(%sr2,%r20) | |
f4c0346c | 649 | #if ENABLE_LWS_DEBUG |
1da177e4 | 650 | stw %r0, 4(%sr2,%r20) |
1da177e4 | 651 | #endif |
c776cd89 | 652 | ssm PSW_SM_I, %r0 |
1da177e4 LT |
653 | b lws_exit |
654 | ldo -EFAULT(%r0),%r21 /* set errno */ | |
655 | nop | |
656 | nop | |
657 | nop | |
658 | nop | |
659 | ||
660 | /* Two exception table entries, one for the load, | |
661 | the other for the store. Either return -EFAULT. | |
662 | Each of the entries must be relocated. */ | |
61dbbaeb HD |
663 | ASM_EXCEPTIONTABLE_ENTRY(1b-linux_gateway_page, 3b-linux_gateway_page) |
664 | ASM_EXCEPTIONTABLE_ENTRY(2b-linux_gateway_page, 3b-linux_gateway_page) | |
1da177e4 | 665 | |
1da177e4 | 666 | |
89206491 GM |
667 | /*************************************************** |
668 | New CAS implementation which uses pointers and variable size | |
669 | information. The value pointed by old and new MUST NOT change | |
670 | while performing CAS. The lock only protect the value at %r26. | |
671 | ||
672 | %r26 - Address to examine | |
673 | %r25 - Pointer to the value to check (old) | |
674 | %r24 - Pointer to the value to set (new) | |
675 | %r23 - Size of the variable (0/1/2/3 for 8/16/32/64 bit) | |
676 | %r28 - Return non-zero on failure | |
677 | %r21 - Kernel error code | |
678 | ||
679 | %r21 has the following meanings: | |
680 | ||
681 | EAGAIN - CAS is busy, ldcw failed, try again. | |
682 | EFAULT - Read or write failed. | |
683 | ||
684 | Scratch: r20, r22, r28, r29, r1, fr4 (32bit for 64bit CAS only) | |
685 | ||
686 | ****************************************************/ | |
687 | ||
688 | /* ELF32 Process entry path */ | |
689 | lws_compare_and_swap_2: | |
690 | #ifdef CONFIG_64BIT | |
691 | /* Clip the input registers */ | |
692 | depdi 0, 31, 32, %r26 | |
693 | depdi 0, 31, 32, %r25 | |
694 | depdi 0, 31, 32, %r24 | |
695 | depdi 0, 31, 32, %r23 | |
696 | #endif | |
697 | ||
698 | /* Check the validity of the size pointer */ | |
699 | subi,>>= 4, %r23, %r0 | |
700 | b,n lws_exit_nosys | |
701 | ||
702 | /* Jump to the functions which will load the old and new values into | |
703 | registers depending on the their size */ | |
704 | shlw %r23, 2, %r29 | |
705 | blr %r29, %r0 | |
706 | nop | |
707 | ||
708 | /* 8bit load */ | |
709 | 4: ldb 0(%sr3,%r25), %r25 | |
710 | b cas2_lock_start | |
711 | 5: ldb 0(%sr3,%r24), %r24 | |
712 | nop | |
713 | nop | |
714 | nop | |
715 | nop | |
716 | nop | |
717 | ||
718 | /* 16bit load */ | |
719 | 6: ldh 0(%sr3,%r25), %r25 | |
720 | b cas2_lock_start | |
721 | 7: ldh 0(%sr3,%r24), %r24 | |
722 | nop | |
723 | nop | |
724 | nop | |
725 | nop | |
726 | nop | |
727 | ||
728 | /* 32bit load */ | |
729 | 8: ldw 0(%sr3,%r25), %r25 | |
730 | b cas2_lock_start | |
731 | 9: ldw 0(%sr3,%r24), %r24 | |
732 | nop | |
733 | nop | |
734 | nop | |
735 | nop | |
736 | nop | |
737 | ||
738 | /* 64bit load */ | |
739 | #ifdef CONFIG_64BIT | |
740 | 10: ldd 0(%sr3,%r25), %r25 | |
741 | 11: ldd 0(%sr3,%r24), %r24 | |
742 | #else | |
743 | /* Load new value into r22/r23 - high/low */ | |
744 | 10: ldw 0(%sr3,%r25), %r22 | |
745 | 11: ldw 4(%sr3,%r25), %r23 | |
746 | /* Load new value into fr4 for atomic store later */ | |
747 | 12: flddx 0(%sr3,%r24), %fr4 | |
748 | #endif | |
749 | ||
750 | cas2_lock_start: | |
751 | /* Load start of lock table */ | |
752 | ldil L%lws_lock_start, %r20 | |
753 | ldo R%lws_lock_start(%r20), %r28 | |
754 | ||
755 | /* Extract four bits from r26 and hash lock (Bits 4-7) */ | |
756 | extru %r26, 27, 4, %r20 | |
757 | ||
758 | /* Find lock to use, the hash is either one of 0 to | |
759 | 15, multiplied by 16 (keep it 16-byte aligned) | |
760 | and add to the lock table offset. */ | |
761 | shlw %r20, 4, %r20 | |
762 | add %r20, %r28, %r20 | |
763 | ||
764 | rsm PSW_SM_I, %r0 /* Disable interrupts */ | |
765 | /* COW breaks can cause contention on UP systems */ | |
766 | LDCW 0(%sr2,%r20), %r28 /* Try to acquire the lock */ | |
767 | cmpb,<>,n %r0, %r28, cas2_action /* Did we get it? */ | |
768 | cas2_wouldblock: | |
769 | ldo 2(%r0), %r28 /* 2nd case */ | |
770 | ssm PSW_SM_I, %r0 | |
771 | b lws_exit /* Contended... */ | |
772 | ldo -EAGAIN(%r0), %r21 /* Spin in userspace */ | |
773 | ||
774 | /* | |
775 | prev = *addr; | |
776 | if ( prev == old ) | |
777 | *addr = new; | |
778 | return prev; | |
779 | */ | |
780 | ||
781 | /* NOTES: | |
782 | This all works becuse intr_do_signal | |
783 | and schedule both check the return iasq | |
784 | and see that we are on the kernel page | |
785 | so this process is never scheduled off | |
786 | or is ever sent any signal of any sort, | |
787 | thus it is wholly atomic from usrspaces | |
788 | perspective | |
789 | */ | |
790 | cas2_action: | |
791 | /* Jump to the correct function */ | |
792 | blr %r29, %r0 | |
793 | /* Set %r28 as non-zero for now */ | |
794 | ldo 1(%r0),%r28 | |
795 | ||
796 | /* 8bit CAS */ | |
797 | 13: ldb,ma 0(%sr3,%r26), %r29 | |
798 | sub,= %r29, %r25, %r0 | |
799 | b,n cas2_end | |
800 | 14: stb,ma %r24, 0(%sr3,%r26) | |
801 | b cas2_end | |
802 | copy %r0, %r28 | |
803 | nop | |
804 | nop | |
805 | ||
806 | /* 16bit CAS */ | |
807 | 15: ldh,ma 0(%sr3,%r26), %r29 | |
808 | sub,= %r29, %r25, %r0 | |
809 | b,n cas2_end | |
810 | 16: sth,ma %r24, 0(%sr3,%r26) | |
811 | b cas2_end | |
812 | copy %r0, %r28 | |
813 | nop | |
814 | nop | |
815 | ||
816 | /* 32bit CAS */ | |
817 | 17: ldw,ma 0(%sr3,%r26), %r29 | |
818 | sub,= %r29, %r25, %r0 | |
819 | b,n cas2_end | |
820 | 18: stw,ma %r24, 0(%sr3,%r26) | |
821 | b cas2_end | |
822 | copy %r0, %r28 | |
823 | nop | |
824 | nop | |
825 | ||
826 | /* 64bit CAS */ | |
827 | #ifdef CONFIG_64BIT | |
828 | 19: ldd,ma 0(%sr3,%r26), %r29 | |
1b59ddfc | 829 | sub,*= %r29, %r25, %r0 |
89206491 GM |
830 | b,n cas2_end |
831 | 20: std,ma %r24, 0(%sr3,%r26) | |
832 | copy %r0, %r28 | |
833 | #else | |
834 | /* Compare first word */ | |
835 | 19: ldw,ma 0(%sr3,%r26), %r29 | |
836 | sub,= %r29, %r22, %r0 | |
837 | b,n cas2_end | |
838 | /* Compare second word */ | |
839 | 20: ldw,ma 4(%sr3,%r26), %r29 | |
840 | sub,= %r29, %r23, %r0 | |
841 | b,n cas2_end | |
842 | /* Perform the store */ | |
843 | 21: fstdx %fr4, 0(%sr3,%r26) | |
844 | copy %r0, %r28 | |
845 | #endif | |
846 | ||
847 | cas2_end: | |
848 | /* Free lock */ | |
849 | stw,ma %r20, 0(%sr2,%r20) | |
850 | /* Enable interrupts */ | |
851 | ssm PSW_SM_I, %r0 | |
852 | /* Return to userspace, set no error */ | |
853 | b lws_exit | |
854 | copy %r0, %r21 | |
855 | ||
856 | 22: | |
857 | /* Error occurred on load or store */ | |
858 | /* Free lock */ | |
859 | stw %r20, 0(%sr2,%r20) | |
860 | ssm PSW_SM_I, %r0 | |
861 | ldo 1(%r0),%r28 | |
862 | b lws_exit | |
863 | ldo -EFAULT(%r0),%r21 /* set errno */ | |
864 | nop | |
865 | nop | |
866 | nop | |
867 | ||
868 | /* Exception table entries, for the load and store, return EFAULT. | |
869 | Each of the entries must be relocated. */ | |
870 | ASM_EXCEPTIONTABLE_ENTRY(4b-linux_gateway_page, 22b-linux_gateway_page) | |
871 | ASM_EXCEPTIONTABLE_ENTRY(5b-linux_gateway_page, 22b-linux_gateway_page) | |
872 | ASM_EXCEPTIONTABLE_ENTRY(6b-linux_gateway_page, 22b-linux_gateway_page) | |
873 | ASM_EXCEPTIONTABLE_ENTRY(7b-linux_gateway_page, 22b-linux_gateway_page) | |
874 | ASM_EXCEPTIONTABLE_ENTRY(8b-linux_gateway_page, 22b-linux_gateway_page) | |
875 | ASM_EXCEPTIONTABLE_ENTRY(9b-linux_gateway_page, 22b-linux_gateway_page) | |
876 | ASM_EXCEPTIONTABLE_ENTRY(10b-linux_gateway_page, 22b-linux_gateway_page) | |
877 | ASM_EXCEPTIONTABLE_ENTRY(11b-linux_gateway_page, 22b-linux_gateway_page) | |
878 | ASM_EXCEPTIONTABLE_ENTRY(13b-linux_gateway_page, 22b-linux_gateway_page) | |
879 | ASM_EXCEPTIONTABLE_ENTRY(14b-linux_gateway_page, 22b-linux_gateway_page) | |
880 | ASM_EXCEPTIONTABLE_ENTRY(15b-linux_gateway_page, 22b-linux_gateway_page) | |
881 | ASM_EXCEPTIONTABLE_ENTRY(16b-linux_gateway_page, 22b-linux_gateway_page) | |
882 | ASM_EXCEPTIONTABLE_ENTRY(17b-linux_gateway_page, 22b-linux_gateway_page) | |
883 | ASM_EXCEPTIONTABLE_ENTRY(18b-linux_gateway_page, 22b-linux_gateway_page) | |
884 | ASM_EXCEPTIONTABLE_ENTRY(19b-linux_gateway_page, 22b-linux_gateway_page) | |
885 | ASM_EXCEPTIONTABLE_ENTRY(20b-linux_gateway_page, 22b-linux_gateway_page) | |
886 | #ifndef CONFIG_64BIT | |
887 | ASM_EXCEPTIONTABLE_ENTRY(12b-linux_gateway_page, 22b-linux_gateway_page) | |
888 | ASM_EXCEPTIONTABLE_ENTRY(21b-linux_gateway_page, 22b-linux_gateway_page) | |
889 | #endif | |
890 | ||
1da177e4 | 891 | /* Make sure nothing else is placed on this page */ |
1c593571 | 892 | .align PAGE_SIZE |
8e9e9844 HD |
893 | END(linux_gateway_page) |
894 | ENTRY(end_linux_gateway_page) | |
1da177e4 LT |
895 | |
896 | /* Relocate symbols assuming linux_gateway_page is mapped | |
897 | to virtual address 0x0 */ | |
8e9e9844 | 898 | |
0b3d643f | 899 | #define LWS_ENTRY(_name_) ASM_ULONG_INSN (lws_##_name_ - linux_gateway_page) |
1da177e4 | 900 | |
1bcdd854 HD |
901 | .section .rodata,"a" |
902 | ||
6a45716a | 903 | .align 8 |
1da177e4 LT |
904 | /* Light-weight-syscall table */ |
905 | /* Start of lws table. */ | |
8e9e9844 | 906 | ENTRY(lws_table) |
89206491 GM |
907 | LWS_ENTRY(compare_and_swap32) /* 0 - ELF32 Atomic 32bit CAS */ |
908 | LWS_ENTRY(compare_and_swap64) /* 1 - ELF64 Atomic 32bit CAS */ | |
909 | LWS_ENTRY(compare_and_swap_2) /* 2 - ELF32 Atomic 64bit CAS */ | |
8e9e9844 | 910 | END(lws_table) |
1da177e4 LT |
911 | /* End of lws table */ |
912 | ||
6a45716a | 913 | .align 8 |
8e9e9844 | 914 | ENTRY(sys_call_table) |
1da177e4 | 915 | #include "syscall_table.S" |
8e9e9844 | 916 | END(sys_call_table) |
1da177e4 | 917 | |
413059f2 | 918 | #ifdef CONFIG_64BIT |
6a45716a | 919 | .align 8 |
8e9e9844 | 920 | ENTRY(sys_call_table64) |
1da177e4 LT |
921 | #define SYSCALL_TABLE_64BIT |
922 | #include "syscall_table.S" | |
8e9e9844 | 923 | END(sys_call_table64) |
1da177e4 LT |
924 | #endif |
925 | ||
1da177e4 LT |
926 | /* |
927 | All light-weight-syscall atomic operations | |
928 | will use this set of locks | |
c84c3a69 HD |
929 | |
930 | NOTE: The lws_lock_start symbol must be | |
931 | at least 16-byte aligned for safe use | |
932 | with ldcw. | |
1da177e4 | 933 | */ |
dfcf753b | 934 | .section .data |
6a45716a | 935 | .align L1_CACHE_BYTES |
8e9e9844 | 936 | ENTRY(lws_lock_start) |
1da177e4 | 937 | /* lws locks */ |
1da177e4 LT |
938 | .rept 16 |
939 | /* Keep locks aligned at 16-bytes */ | |
940 | .word 1 | |
941 | .word 0 | |
942 | .word 0 | |
943 | .word 0 | |
944 | .endr | |
8e9e9844 | 945 | END(lws_lock_start) |
1da177e4 | 946 | .previous |
1da177e4 LT |
947 | |
948 | .end | |
949 | ||
950 |