2 * Linux/PA-RISC Project (http://www.parisc-linux.org/)
4 * System call entry code Copyright (c) Matthew Wilcox 1999 <willy@bofh.ai>
5 * Licensed under the GNU GPL.
6 * thanks to Philipp Rumpf, Mike Shaver and various others
7 * sorry about the wall, puffin..
10 #include <asm/asm-offsets.h>
11 #include <asm/unistd.h>
12 #include <asm/errno.h>
14 #include <asm/thread_info.h>
15 #include <asm/assembly.h>
16 #include <asm/processor.h>
18 #include <linux/linkage.h>
20 /* We fill the empty parts of the gateway page with
21 * something that will kill the kernel or a
22 * userspace application.
24 #define KILL_INSN break 0,0
30 .import syscall_exit,code
31 .import syscall_exit_rfi,code
33 /* Linux gateway page is aliased to virtual page 0 in the kernel
34 * address space. Since it is a gateway page it cannot be
35 * dereferenced, so null pointers will still fault. We start
36 * the actual entry point at 0x100. We put break instructions
37 * at the beginning of the page to trap null indirect function
42 ENTRY(linux_gateway_page)
44 /* ADDRESS 0x00 to 0xb0 = 176 bytes / 4 bytes per insn = 44 insns */
49 /* ADDRESS 0xb0 to 0xb4, lws uses 1 insns for entry */
50 /* Light-weight-syscall entry must always be located at 0xb0 */
51 /* WARNING: Keep this number updated with table size changes */
52 #define __NR_lws_entries (2)
55 /* Unconditional branch to lws_start, located on the
59 /* Fill from 0xb4 to 0xe0 */
64 /* This function MUST be located at 0xe0 for glibc's threading
65 mechanism to work. DO NOT MOVE THIS CODE EVER! */
67 gate .+8, %r0 /* increase privilege */
68 depi 3, 31, 2, %r31 /* Ensure we return into user mode. */
69 be 0(%sr7,%r31) /* return to user space */
70 mtctl %r26, %cr27 /* move arg0 to the control register */
72 /* Increase the chance of trapping if random jumps occur to this
73 address, fill from 0xf0 to 0x100 */
78 /* This address must remain fixed at 0x100 for glibc's syscalls to work */
81 gate .+8, %r0 /* become privileged */
82 mtsp %r0,%sr4 /* get kernel space into sr4 */
83 mtsp %r0,%sr5 /* get kernel space into sr5 */
84 mtsp %r0,%sr6 /* get kernel space into sr6 */
85 mfsp %sr7,%r1 /* save user sr7 */
86 mtsp %r1,%sr3 /* and store it in sr3 */
89 /* for now we can *always* set the W bit on entry to the syscall
90 * since we don't support wide userland processes. We could
91 * also save the current SM other than in r0 and restore it on
92 * exit from the syscall, and also use that value to know
93 * whether to do narrow or wide syscalls. -PB
96 extrd,u %r1,PSW_W_BIT,1,%r1
97 /* sp must be aligned on 4, so deposit the W bit setting into
98 * the bottom of sp temporarily */
101 /* The top halves of argument registers must be cleared on syscall
102 * entry from narrow executable.
104 depdi 0, 31, 32, %r26
105 depdi 0, 31, 32, %r25
106 depdi 0, 31, 32, %r24
107 depdi 0, 31, 32, %r23
108 depdi 0, 31, 32, %r22
109 depdi 0, 31, 32, %r21
113 xor %r1,%r30,%r30 /* ye olde xor trick */
117 ldo THREAD_SZ_ALGN+FRAME_SIZE(%r30),%r30 /* set up kernel stack */
119 /* N.B.: It is critical that we don't set sr7 to 0 until r30
120 * contains a valid kernel stack pointer. It is also
121 * critical that we don't start using the kernel stack
122 * until after sr7 has been set to 0.
125 mtsp %r0,%sr7 /* get kernel space into sr7 */
126 STREGM %r1,FRAME_SIZE(%r30) /* save r1 (usp) here for now */
127 mfctl %cr30,%r1 /* get task ptr in %r1 */
128 LDREG TI_TASK(%r1),%r1
130 /* Save some registers for sigcontext and potential task
131 switch (see entry.S for the details of which ones are
132 saved/restored). TASK_PT_PSW is zeroed so we can see whether
133 a process is on a syscall or not. For an interrupt the real
134 PSW value is stored. This is needed for gdb and sys_ptrace. */
135 STREG %r0, TASK_PT_PSW(%r1)
136 STREG %r2, TASK_PT_GR2(%r1) /* preserve rp */
137 STREG %r19, TASK_PT_GR19(%r1)
139 LDREGM -FRAME_SIZE(%r30), %r2 /* get users sp back */
141 extrd,u %r2,63,1,%r19 /* W hidden in bottom bit */
143 xor %r19,%r2,%r2 /* clear bottom bit */
145 std %r19,TASK_PT_PSW(%r1)
148 STREG %r2, TASK_PT_GR30(%r1) /* ... and save it */
150 STREG %r20, TASK_PT_GR20(%r1) /* Syscall number */
151 STREG %r21, TASK_PT_GR21(%r1)
152 STREG %r22, TASK_PT_GR22(%r1)
153 STREG %r23, TASK_PT_GR23(%r1) /* 4th argument */
154 STREG %r24, TASK_PT_GR24(%r1) /* 3rd argument */
155 STREG %r25, TASK_PT_GR25(%r1) /* 2nd argument */
156 STREG %r26, TASK_PT_GR26(%r1) /* 1st argument */
157 STREG %r27, TASK_PT_GR27(%r1) /* user dp */
158 STREG %r28, TASK_PT_GR28(%r1) /* return value 0 */
159 STREG %r28, TASK_PT_ORIG_R28(%r1) /* return value 0 (saved for signals) */
160 STREG %r29, TASK_PT_GR29(%r1) /* return value 1 */
161 STREG %r31, TASK_PT_GR31(%r1) /* preserve syscall return ptr */
163 ldo TASK_PT_FR0(%r1), %r27 /* save fpregs from the kernel */
164 save_fp %r27 /* or potential task switch */
166 mfctl %cr11, %r27 /* i.e. SAR */
167 STREG %r27, TASK_PT_SAR(%r1)
172 ldo -16(%r30),%r29 /* Reference param save area */
173 copy %r19,%r2 /* W bit back to r2 */
175 /* no need to save these on stack in wide mode because the first 8
176 * args are passed in registers */
177 stw %r22, -52(%r30) /* 5th argument */
178 stw %r21, -56(%r30) /* 6th argument */
181 /* Are we being ptraced? */
183 LDREG TI_TASK(%r1),%r1
184 ldw TASK_PTRACE(%r1), %r1
185 bb,<,n %r1,31,.Ltracesys
187 /* Note! We cannot use the syscall table that is mapped
188 nearby since the gateway page is mapped execute-only. */
191 ldil L%sys_call_table, %r1
193 addil L%(sys_call_table64-sys_call_table), %r1
194 ldo R%sys_call_table(%r1), %r19
196 ldo R%sys_call_table64(%r1), %r19
198 ldil L%sys_call_table, %r1
199 ldo R%sys_call_table(%r1), %r19
201 comiclr,>> __NR_Linux_syscalls, %r20, %r0
204 LDREGX %r20(%r19), %r19
206 /* If this is a sys_rt_sigreturn call, and the signal was received
207 * when not in_syscall, then we want to return via syscall_exit_rfi,
208 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
209 * trampoline code in signal.c).
211 ldi __NR_rt_sigreturn,%r2
212 comb,= %r2,%r20,.Lrt_sigreturn
214 ldil L%syscall_exit,%r2
216 ldo R%syscall_exit(%r2),%r2
218 comib,<> 0,%r25,.Lin_syscall
219 ldil L%syscall_exit_rfi,%r2
221 ldo R%syscall_exit_rfi(%r2),%r2
223 /* Note! Because we are not running where we were linked, any
224 calls to functions external to this file must be indirect. To
225 be safe, we apply the opposite rule to functions within this
226 file, with local labels given to them to ensure correctness. */
230 ldil L%syscall_exit,%r1
231 be R%syscall_exit(%sr7,%r1)
232 ldo -ENOSYS(%r0),%r28 /* set errno */
235 /* Warning! This trace code is a virtual duplicate of the code above so be
236 * sure to maintain both! */
239 /* Need to save more registers so the debugger can see where we
240 * are. This saves only the lower 8 bits of PSW, so that the C
241 * bit is still clear on syscalls, and the D bit is set if this
242 * full register save path has been executed. We check the D
243 * bit on syscall_return_rfi to determine which registers to
244 * restore. An interrupt results in a full PSW saved with the
245 * C bit set, a non-straced syscall entry results in C and D clear
248 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
249 LDREG TI_TASK(%r1), %r1
251 STREG %r2,TASK_PT_PSW(%r1) /* Lower 8 bits only!! */
253 STREG %r2,TASK_PT_SR0(%r1)
255 STREG %r2,TASK_PT_SR1(%r1)
257 STREG %r2,TASK_PT_SR2(%r1)
259 STREG %r2,TASK_PT_SR3(%r1)
260 STREG %r2,TASK_PT_SR4(%r1)
261 STREG %r2,TASK_PT_SR5(%r1)
262 STREG %r2,TASK_PT_SR6(%r1)
263 STREG %r2,TASK_PT_SR7(%r1)
264 STREG %r2,TASK_PT_IASQ0(%r1)
265 STREG %r2,TASK_PT_IASQ1(%r1)
266 LDREG TASK_PT_GR31(%r1),%r2
267 STREG %r2,TASK_PT_IAOQ0(%r1)
269 STREG %r2,TASK_PT_IAOQ1(%r1)
270 ldo TASK_REGS(%r1),%r2
272 STREG %r3,PT_GR3(%r2)
273 STREG %r4,PT_GR4(%r2)
274 STREG %r5,PT_GR5(%r2)
275 STREG %r6,PT_GR6(%r2)
276 STREG %r7,PT_GR7(%r2)
277 STREG %r8,PT_GR8(%r2)
278 STREG %r9,PT_GR9(%r2)
279 STREG %r10,PT_GR10(%r2)
280 STREG %r11,PT_GR11(%r2)
281 STREG %r12,PT_GR12(%r2)
282 STREG %r13,PT_GR13(%r2)
283 STREG %r14,PT_GR14(%r2)
284 STREG %r15,PT_GR15(%r2)
285 STREG %r16,PT_GR16(%r2)
286 STREG %r17,PT_GR17(%r2)
287 STREG %r18,PT_GR18(%r2)
288 /* Finished saving things for the debugger */
290 ldil L%syscall_trace,%r1
291 ldil L%tracesys_next,%r2
292 be R%syscall_trace(%sr7,%r1)
293 ldo R%tracesys_next(%r2),%r2
296 ldil L%sys_call_table,%r1
297 ldo R%sys_call_table(%r1), %r19
299 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
300 LDREG TI_TASK(%r1), %r1
301 LDREG TASK_PT_GR20(%r1), %r20
302 LDREG TASK_PT_GR26(%r1), %r26 /* Restore the users args */
303 LDREG TASK_PT_GR25(%r1), %r25
304 LDREG TASK_PT_GR24(%r1), %r24
305 LDREG TASK_PT_GR23(%r1), %r23
307 LDREG TASK_PT_GR22(%r1), %r22
308 LDREG TASK_PT_GR21(%r1), %r21
309 ldo -16(%r30),%r29 /* Reference param save area */
312 comiclr,>>= __NR_Linux_syscalls, %r20, %r0
315 LDREGX %r20(%r19), %r19
317 /* If this is a sys_rt_sigreturn call, and the signal was received
318 * when not in_syscall, then we want to return via syscall_exit_rfi,
319 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
320 * trampoline code in signal.c).
322 ldi __NR_rt_sigreturn,%r2
323 comb,= %r2,%r20,.Ltrace_rt_sigreturn
325 ldil L%tracesys_exit,%r2
327 ldo R%tracesys_exit(%r2),%r2
329 /* Do *not* call this function on the gateway page, because it
330 makes a direct call to syscall_trace. */
333 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
334 LDREG TI_TASK(%r1), %r1
336 ldo -16(%r30),%r29 /* Reference param save area */
338 bl syscall_trace, %r2
339 STREG %r28,TASK_PT_GR28(%r1) /* save return value now */
340 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
341 LDREG TI_TASK(%r1), %r1
342 LDREG TASK_PT_GR28(%r1), %r28 /* Restore return val. */
344 ldil L%syscall_exit,%r1
345 be,n R%syscall_exit(%sr7,%r1)
347 .Ltrace_rt_sigreturn:
348 comib,<> 0,%r25,.Ltrace_in_syscall
349 ldil L%tracesys_sigexit,%r2
351 ldo R%tracesys_sigexit(%r2),%r2
354 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
357 ldo -16(%r30),%r29 /* Reference param save area */
359 bl syscall_trace, %r2
362 ldil L%syscall_exit_rfi,%r1
363 be,n R%syscall_exit_rfi(%sr7,%r1)
366 /*********************************************************
367 Light-weight-syscall code
370 r26,r25,r24,r23,r22 - Input registers
371 r28 - Function return register
374 Scracth: Any of the above that aren't being
375 currently used, including r1.
377 Return pointer: r31 (Not usable)
379 Error codes returned by entry path:
381 ENOSYS - r20 was an invalid LWS number.
383 *********************************************************/
385 /* Gate and ensure we return to userspace */
387 depi 3, 31, 2, %r31 /* Ensure we return to userspace */
390 /* FIXME: If we are a 64-bit kernel just
391 * turn this on unconditionally.
394 extrd,u %r1,PSW_W_BIT,1,%r1
395 /* sp must be aligned on 4, so deposit the W bit setting into
396 * the bottom of sp temporarily */
399 /* Clip LWS number to a 32-bit value always */
400 depdi 0, 31, 32, %r20
403 /* Is the lws entry number valid? */
404 comiclr,>>= __NR_lws_entries, %r20, %r0
407 /* WARNING: Trashing sr2 and sr3 */
408 mfsp %sr7,%r1 /* get userspace into sr3 */
410 mtsp %r0,%sr2 /* get kernel space into sr2 */
412 /* Load table start */
413 ldil L%lws_table, %r1
414 ldo R%lws_table(%r1), %r28 /* Scratch use of r28 */
415 LDREGX %r20(%sr2,r28), %r21 /* Scratch use of r21 */
417 /* Jump to lws, lws table pointers already relocated */
421 ldo -ENOSYS(%r0),%r21 /* set errno */
422 /* Fall through: Return to userspace */
426 /* decide whether to reset the wide mode bit
428 * For a syscall, the W bit is stored in the lowest bit
429 * of sp. Extract it and reset W if it is zero */
430 extrd,u,*<> %r30,63,1,%r1
432 /* now reset the lowest bit of sp if it was set */
439 /***************************************************
440 Implementing CAS as an atomic operation:
442 %r26 - Address to examine
443 %r25 - Old value to check (old)
444 %r24 - New value to set (new)
445 %r28 - Return prev through this register.
446 %r21 - Kernel error code
448 If debugging is DISabled:
450 %r21 has the following meanings:
452 EAGAIN - CAS is busy, ldcw failed, try again.
453 EFAULT - Read or write failed.
455 If debugging is enabled:
457 EDEADLOCK - CAS called recursively.
458 EAGAIN && r28 == 1 - CAS is busy. Lock contended.
459 EAGAIN && r28 == 2 - CAS is busy. ldcw failed.
460 EFAULT - Read or write failed.
462 Scratch: r20, r28, r1
464 ****************************************************/
466 /* Do not enable LWS debugging */
467 #define ENABLE_LWS_DEBUG 0
469 /* ELF64 Process entry path */
470 lws_compare_and_swap64:
472 b,n lws_compare_and_swap
474 /* If we are not a 64-bit kernel, then we don't
475 * implement having 64-bit input registers
480 /* ELF32 Process entry path */
481 lws_compare_and_swap32:
483 /* Clip all the input registers */
484 depdi 0, 31, 32, %r26
485 depdi 0, 31, 32, %r25
486 depdi 0, 31, 32, %r24
489 lws_compare_and_swap:
491 /* Load start of lock table */
492 ldil L%lws_lock_start, %r20
493 ldo R%lws_lock_start(%r20), %r28
495 /* Extract four bits from r26 and hash lock (Bits 4-7) */
496 extru %r26, 27, 4, %r20
498 /* Find lock to use, the hash is either one of 0 to
499 15, multiplied by 16 (keep it 16-byte aligned)
500 and add to the lock table offset. */
504 # if ENABLE_LWS_DEBUG
506 DEBUG, check for deadlock!
507 If the thread register values are the same
508 then we were the one that locked it last and
509 this is a recurisve call that will deadlock.
510 We *must* giveup this call and fail.
512 ldw 4(%sr2,%r20), %r28 /* Load thread register */
513 /* WARNING: If cr27 cycles to the same value we have problems */
514 mfctl %cr27, %r21 /* Get current thread register */
515 cmpb,<>,n %r21, %r28, cas_lock /* Called recursive? */
516 b lws_exit /* Return error! */
517 ldo -EDEADLOCK(%r0), %r21
519 cmpb,=,n %r0, %r28, cas_nocontend /* Is nobody using it? */
520 ldo 1(%r0), %r28 /* 1st case */
521 b lws_exit /* Contended... */
522 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
525 /* ENABLE_LWS_DEBUG */
527 LDCW 0(%sr2,%r20), %r28 /* Try to acquire the lock */
528 cmpb,<>,n %r0, %r28, cas_action /* Did we get it? */
530 ldo 2(%r0), %r28 /* 2nd case */
531 b lws_exit /* Contended... */
532 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
544 This all works becuse intr_do_signal
545 and schedule both check the return iasq
546 and see that we are on the kernel page
547 so this process is never scheduled off
548 or is ever sent any signal of any sort,
549 thus it is wholly atomic from usrspaces
553 #if defined CONFIG_SMP && ENABLE_LWS_DEBUG
556 stw %r1, 4(%sr2,%r20)
558 /* The load and store could fail */
559 1: ldw 0(%sr3,%r26), %r28
560 sub,<> %r28, %r25, %r0
561 2: stw %r24, 0(%sr3,%r26)
564 stw %r20, 0(%sr2,%r20)
565 # if ENABLE_LWS_DEBUG
566 /* Clear thread register indicator */
567 stw %r0, 4(%sr2,%r20)
570 /* Return to userspace, set no error */
575 /* Error occured on load or store */
578 stw %r20, 0(%sr2,%r20)
579 # if ENABLE_LWS_DEBUG
580 stw %r0, 4(%sr2,%r20)
584 ldo -EFAULT(%r0),%r21 /* set errno */
590 /* Two exception table entries, one for the load,
591 the other for the store. Either return -EFAULT.
592 Each of the entries must be relocated. */
593 .section __ex_table,"aw"
594 ASM_ULONG_INSN (1b - linux_gateway_page), (3b - linux_gateway_page)
595 ASM_ULONG_INSN (2b - linux_gateway_page), (3b - linux_gateway_page)
599 /* Make sure nothing else is placed on this page */
601 END(linux_gateway_page)
602 ENTRY(end_linux_gateway_page)
604 /* Relocate symbols assuming linux_gateway_page is mapped
605 to virtual address 0x0 */
607 #define LWS_ENTRY(_name_) ASM_ULONG_INSN (lws_##_name_ - linux_gateway_page)
612 /* Light-weight-syscall table */
613 /* Start of lws table. */
615 LWS_ENTRY(compare_and_swap32) /* 0 - ELF32 Atomic compare and swap */
616 LWS_ENTRY(compare_and_swap64) /* 1 - ELF64 Atomic compare and swap */
618 /* End of lws table */
621 ENTRY(sys_call_table)
622 #include "syscall_table.S"
627 ENTRY(sys_call_table64)
628 #define SYSCALL_TABLE_64BIT
629 #include "syscall_table.S"
630 END(sys_call_table64)
635 All light-weight-syscall atomic operations
636 will use this set of locks
640 ENTRY(lws_lock_start)
644 /* Keep locks aligned at 16-bytes */
653 /* CONFIG_SMP for lws_lock_start */