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>
16 #include <asm/assembly.h>
17 #include <asm/processor.h>
19 /* We fill the empty parts of the gateway page with
20 * something that will kill the kernel or a
21 * userspace application.
23 #define KILL_INSN break 0,0
33 .import syscall_exit,code
34 .import syscall_exit_rfi,code
35 .export linux_gateway_page
37 /* Linux gateway page is aliased to virtual page 0 in the kernel
38 * address space. Since it is a gateway page it cannot be
39 * dereferenced, so null pointers will still fault. We start
40 * the actual entry point at 0x100. We put break instructions
41 * at the beginning of the page to trap null indirect function
48 /* ADDRESS 0x00 to 0xb0 = 176 bytes / 4 bytes per insn = 44 insns */
53 /* ADDRESS 0xb0 to 0xb4, lws uses 1 insns for entry */
54 /* Light-weight-syscall entry must always be located at 0xb0 */
55 /* WARNING: Keep this number updated with table size changes */
56 #define __NR_lws_entries (2)
59 /* Unconditional branch to lws_start, located on the
63 /* Fill from 0xb4 to 0xe0 */
68 /* This function MUST be located at 0xe0 for glibc's threading
69 mechanism to work. DO NOT MOVE THIS CODE EVER! */
71 gate .+8, %r0 /* increase privilege */
72 depi 3, 31, 2, %r31 /* Ensure we return into user mode. */
73 be 0(%sr7,%r31) /* return to user space */
74 mtctl %r26, %cr27 /* move arg0 to the control register */
76 /* Increase the chance of trapping if random jumps occur to this
77 address, fill from 0xf0 to 0x100 */
82 /* This address must remain fixed at 0x100 for glibc's syscalls to work */
85 gate .+8, %r0 /* become privileged */
86 mtsp %r0,%sr4 /* get kernel space into sr4 */
87 mtsp %r0,%sr5 /* get kernel space into sr5 */
88 mtsp %r0,%sr6 /* get kernel space into sr6 */
89 mfsp %sr7,%r1 /* save user sr7 */
90 mtsp %r1,%sr3 /* and store it in sr3 */
93 /* for now we can *always* set the W bit on entry to the syscall
94 * since we don't support wide userland processes. We could
95 * also save the current SM other than in r0 and restore it on
96 * exit from the syscall, and also use that value to know
97 * whether to do narrow or wide syscalls. -PB
100 extrd,u %r1,PSW_W_BIT,1,%r1
101 /* sp must be aligned on 4, so deposit the W bit setting into
102 * the bottom of sp temporarily */
105 /* The top halves of argument registers must be cleared on syscall
106 * entry from narrow executable.
108 depdi 0, 31, 32, %r26
109 depdi 0, 31, 32, %r25
110 depdi 0, 31, 32, %r24
111 depdi 0, 31, 32, %r23
112 depdi 0, 31, 32, %r22
113 depdi 0, 31, 32, %r21
117 xor %r1,%r30,%r30 /* ye olde xor trick */
121 ldo THREAD_SZ_ALGN+FRAME_SIZE(%r30),%r30 /* set up kernel stack */
123 /* N.B.: It is critical that we don't set sr7 to 0 until r30
124 * contains a valid kernel stack pointer. It is also
125 * critical that we don't start using the kernel stack
126 * until after sr7 has been set to 0.
129 mtsp %r0,%sr7 /* get kernel space into sr7 */
130 STREGM %r1,FRAME_SIZE(%r30) /* save r1 (usp) here for now */
131 mfctl %cr30,%r1 /* get task ptr in %r1 */
132 LDREG TI_TASK(%r1),%r1
134 /* Save some registers for sigcontext and potential task
135 switch (see entry.S for the details of which ones are
136 saved/restored). TASK_PT_PSW is zeroed so we can see whether
137 a process is on a syscall or not. For an interrupt the real
138 PSW value is stored. This is needed for gdb and sys_ptrace. */
139 STREG %r0, TASK_PT_PSW(%r1)
140 STREG %r2, TASK_PT_GR2(%r1) /* preserve rp */
141 STREG %r19, TASK_PT_GR19(%r1)
143 LDREGM -FRAME_SIZE(%r30), %r2 /* get users sp back */
145 extrd,u %r2,63,1,%r19 /* W hidden in bottom bit */
147 xor %r19,%r2,%r2 /* clear bottom bit */
149 std %r19,TASK_PT_PSW(%r1)
152 STREG %r2, TASK_PT_GR30(%r1) /* ... and save it */
154 STREG %r20, TASK_PT_GR20(%r1) /* Syscall number */
155 STREG %r21, TASK_PT_GR21(%r1)
156 STREG %r22, TASK_PT_GR22(%r1)
157 STREG %r23, TASK_PT_GR23(%r1) /* 4th argument */
158 STREG %r24, TASK_PT_GR24(%r1) /* 3rd argument */
159 STREG %r25, TASK_PT_GR25(%r1) /* 2nd argument */
160 STREG %r26, TASK_PT_GR26(%r1) /* 1st argument */
161 STREG %r27, TASK_PT_GR27(%r1) /* user dp */
162 STREG %r28, TASK_PT_GR28(%r1) /* return value 0 */
163 STREG %r28, TASK_PT_ORIG_R28(%r1) /* return value 0 (saved for signals) */
164 STREG %r29, TASK_PT_GR29(%r1) /* return value 1 */
165 STREG %r31, TASK_PT_GR31(%r1) /* preserve syscall return ptr */
167 ldo TASK_PT_FR0(%r1), %r27 /* save fpregs from the kernel */
168 save_fp %r27 /* or potential task switch */
170 mfctl %cr11, %r27 /* i.e. SAR */
171 STREG %r27, TASK_PT_SAR(%r1)
176 ldo -16(%r30),%r29 /* Reference param save area */
177 copy %r19,%r2 /* W bit back to r2 */
179 /* no need to save these on stack in wide mode because the first 8
180 * args are passed in registers */
181 stw %r22, -52(%r30) /* 5th argument */
182 stw %r21, -56(%r30) /* 6th argument */
185 /* Are we being ptraced? */
187 LDREG TI_TASK(%r1),%r1
188 LDREG TASK_PTRACE(%r1), %r1
189 bb,<,n %r1,31,.Ltracesys
191 /* Note! We cannot use the syscall table that is mapped
192 nearby since the gateway page is mapped execute-only. */
195 ldil L%sys_call_table, %r1
197 addil L%(sys_call_table64-sys_call_table), %r1
198 ldo R%sys_call_table(%r1), %r19
200 ldo R%sys_call_table64(%r1), %r19
202 ldil L%sys_call_table, %r1
203 ldo R%sys_call_table(%r1), %r19
205 comiclr,>>= __NR_Linux_syscalls, %r20, %r0
208 LDREGX %r20(%r19), %r19
210 /* If this is a sys_rt_sigreturn call, and the signal was received
211 * when not in_syscall, then we want to return via syscall_exit_rfi,
212 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
213 * trampoline code in signal.c).
215 ldi __NR_rt_sigreturn,%r2
216 comb,= %r2,%r20,.Lrt_sigreturn
218 ldil L%syscall_exit,%r2
220 ldo R%syscall_exit(%r2),%r2
222 comib,<> 0,%r25,.Lin_syscall
223 ldil L%syscall_exit_rfi,%r2
225 ldo R%syscall_exit_rfi(%r2),%r2
227 /* Note! Because we are not running where we were linked, any
228 calls to functions external to this file must be indirect. To
229 be safe, we apply the opposite rule to functions within this
230 file, with local labels given to them to ensure correctness. */
234 ldil L%syscall_exit,%r1
235 be R%syscall_exit(%sr7,%r1)
236 ldo -ENOSYS(%r0),%r28 /* set errno */
239 /* Warning! This trace code is a virtual duplicate of the code above so be
240 * sure to maintain both! */
243 /* Need to save more registers so the debugger can see where we
244 * are. This saves only the lower 8 bits of PSW, so that the C
245 * bit is still clear on syscalls, and the D bit is set if this
246 * full register save path has been executed. We check the D
247 * bit on syscall_return_rfi to determine which registers to
248 * restore. An interrupt results in a full PSW saved with the
249 * C bit set, a non-straced syscall entry results in C and D clear
252 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
253 LDREG TI_TASK(%r1), %r1
255 STREG %r2,TASK_PT_PSW(%r1) /* Lower 8 bits only!! */
257 STREG %r2,TASK_PT_SR0(%r1)
259 STREG %r2,TASK_PT_SR1(%r1)
261 STREG %r2,TASK_PT_SR2(%r1)
263 STREG %r2,TASK_PT_SR3(%r1)
264 STREG %r2,TASK_PT_SR4(%r1)
265 STREG %r2,TASK_PT_SR5(%r1)
266 STREG %r2,TASK_PT_SR6(%r1)
267 STREG %r2,TASK_PT_SR7(%r1)
268 STREG %r2,TASK_PT_IASQ0(%r1)
269 STREG %r2,TASK_PT_IASQ1(%r1)
270 LDREG TASK_PT_GR31(%r1),%r2
271 STREG %r2,TASK_PT_IAOQ0(%r1)
273 STREG %r2,TASK_PT_IAOQ1(%r1)
274 ldo TASK_REGS(%r1),%r2
276 STREG %r3,PT_GR3(%r2)
277 STREG %r4,PT_GR4(%r2)
278 STREG %r5,PT_GR5(%r2)
279 STREG %r6,PT_GR6(%r2)
280 STREG %r7,PT_GR7(%r2)
281 STREG %r8,PT_GR8(%r2)
282 STREG %r9,PT_GR9(%r2)
283 STREG %r10,PT_GR10(%r2)
284 STREG %r11,PT_GR11(%r2)
285 STREG %r12,PT_GR12(%r2)
286 STREG %r13,PT_GR13(%r2)
287 STREG %r14,PT_GR14(%r2)
288 STREG %r15,PT_GR15(%r2)
289 STREG %r16,PT_GR16(%r2)
290 STREG %r17,PT_GR17(%r2)
291 STREG %r18,PT_GR18(%r2)
292 /* Finished saving things for the debugger */
294 ldil L%syscall_trace,%r1
295 ldil L%tracesys_next,%r2
296 be R%syscall_trace(%sr7,%r1)
297 ldo R%tracesys_next(%r2),%r2
300 ldil L%sys_call_table,%r1
301 ldo R%sys_call_table(%r1), %r19
303 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
304 LDREG TI_TASK(%r1), %r1
305 LDREG TASK_PT_GR20(%r1), %r20
306 LDREG TASK_PT_GR26(%r1), %r26 /* Restore the users args */
307 LDREG TASK_PT_GR25(%r1), %r25
308 LDREG TASK_PT_GR24(%r1), %r24
309 LDREG TASK_PT_GR23(%r1), %r23
311 LDREG TASK_PT_GR22(%r1), %r22
312 LDREG TASK_PT_GR21(%r1), %r21
313 ldo -16(%r30),%r29 /* Reference param save area */
316 comiclr,>>= __NR_Linux_syscalls, %r20, %r0
319 LDREGX %r20(%r19), %r19
321 /* If this is a sys_rt_sigreturn call, and the signal was received
322 * when not in_syscall, then we want to return via syscall_exit_rfi,
323 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
324 * trampoline code in signal.c).
326 ldi __NR_rt_sigreturn,%r2
327 comb,= %r2,%r20,.Ltrace_rt_sigreturn
329 ldil L%tracesys_exit,%r2
331 ldo R%tracesys_exit(%r2),%r2
333 /* Do *not* call this function on the gateway page, because it
334 makes a direct call to syscall_trace. */
337 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
338 LDREG TI_TASK(%r1), %r1
340 ldo -16(%r30),%r29 /* Reference param save area */
342 bl syscall_trace, %r2
343 STREG %r28,TASK_PT_GR28(%r1) /* save return value now */
344 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
345 LDREG TI_TASK(%r1), %r1
346 LDREG TASK_PT_GR28(%r1), %r28 /* Restore return val. */
348 ldil L%syscall_exit,%r1
349 be,n R%syscall_exit(%sr7,%r1)
351 .Ltrace_rt_sigreturn:
352 comib,<> 0,%r25,.Ltrace_in_syscall
353 ldil L%tracesys_sigexit,%r2
355 ldo R%tracesys_sigexit(%r2),%r2
358 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
361 ldo -16(%r30),%r29 /* Reference param save area */
363 bl syscall_trace, %r2
366 ldil L%syscall_exit_rfi,%r1
367 be,n R%syscall_exit_rfi(%sr7,%r1)
370 /*********************************************************
371 Light-weight-syscall code
374 r26,r25,r24,r23,r22 - Input registers
375 r28 - Function return register
378 Scracth: Any of the above that aren't being
379 currently used, including r1.
381 Return pointer: r31 (Not usable)
383 Error codes returned by entry path:
385 ENOSYS - r20 was an invalid LWS number.
387 *********************************************************/
389 /* Gate and ensure we return to userspace */
391 depi 3, 31, 2, %r31 /* Ensure we return to userspace */
394 /* FIXME: If we are a 64-bit kernel just
395 * turn this on unconditionally.
398 extrd,u %r1,PSW_W_BIT,1,%r1
399 /* sp must be aligned on 4, so deposit the W bit setting into
400 * the bottom of sp temporarily */
403 /* Clip LWS number to a 32-bit value always */
404 depdi 0, 31, 32, %r20
407 /* Is the lws entry number valid? */
408 comiclr,>>= __NR_lws_entries, %r20, %r0
411 /* WARNING: Trashing sr2 and sr3 */
412 mfsp %sr7,%r1 /* get userspace into sr3 */
414 mtsp %r0,%sr2 /* get kernel space into sr2 */
416 /* Load table start */
417 ldil L%lws_table, %r1
418 ldo R%lws_table(%r1), %r28 /* Scratch use of r28 */
419 LDREGX %r20(%sr2,r28), %r21 /* Scratch use of r21 */
421 /* Jump to lws, lws table pointers already relocated */
425 ldo -ENOSYS(%r0),%r21 /* set errno */
426 /* Fall through: Return to userspace */
430 /* decide whether to reset the wide mode bit
432 * For a syscall, the W bit is stored in the lowest bit
433 * of sp. Extract it and reset W if it is zero */
434 extrd,u,*<> %r30,63,1,%r1
436 /* now reset the lowest bit of sp if it was set */
443 /***************************************************
444 Implementing CAS as an atomic operation:
446 %r26 - Address to examine
447 %r25 - Old value to check (old)
448 %r24 - New value to set (new)
449 %r28 - Return prev through this register.
450 %r21 - Kernel error code
452 If debugging is DISabled:
454 %r21 has the following meanings:
456 EAGAIN - CAS is busy, ldcw failed, try again.
457 EFAULT - Read or write failed.
459 If debugging is enabled:
461 EDEADLOCK - CAS called recursively.
462 EAGAIN && r28 == 1 - CAS is busy. Lock contended.
463 EAGAIN && r28 == 2 - CAS is busy. ldcw failed.
464 EFAULT - Read or write failed.
466 Scratch: r20, r28, r1
468 ****************************************************/
470 /* Do not enable LWS debugging */
471 #define ENABLE_LWS_DEBUG 0
473 /* ELF64 Process entry path */
474 lws_compare_and_swap64:
476 b,n lws_compare_and_swap
478 /* If we are not a 64-bit kernel, then we don't
479 * implement having 64-bit input registers
484 /* ELF32 Process entry path */
485 lws_compare_and_swap32:
487 /* Clip all the input registers */
488 depdi 0, 31, 32, %r26
489 depdi 0, 31, 32, %r25
490 depdi 0, 31, 32, %r24
493 lws_compare_and_swap:
495 /* Load start of lock table */
496 ldil L%lws_lock_start, %r20
497 ldo R%lws_lock_start(%r20), %r28
499 /* Extract four bits from r26 and hash lock (Bits 4-7) */
500 extru %r26, 27, 4, %r20
502 /* Find lock to use, the hash is either one of 0 to
503 15, multiplied by 16 (keep it 16-byte aligned)
504 and add to the lock table offset. */
508 # ifdef ENABLE_LWS_DEBUG
510 DEBUG, check for deadlock!
511 If the thread register values are the same
512 then we were the one that locked it last and
513 this is a recurisve call that will deadlock.
514 We *must* giveup this call and fail.
516 ldw 4(%sr2,%r20), %r28 /* Load thread register */
517 /* WARNING: If cr27 cycles to the same value we have problems */
518 mfctl %cr27, %r21 /* Get current thread register */
519 cmpb,<>,n %r21, %r28, cas_lock /* Called recursive? */
520 b lws_exit /* Return error! */
521 ldo -EDEADLOCK(%r0), %r21
523 cmpb,=,n %r0, %r28, cas_nocontend /* Is nobody using it? */
524 ldo 1(%r0), %r28 /* 1st case */
525 b lws_exit /* Contended... */
526 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
529 /* ENABLE_LWS_DEBUG */
531 LDCW 0(%sr2,%r20), %r28 /* Try to acquire the lock */
532 cmpb,<>,n %r0, %r28, cas_action /* Did we get it? */
534 ldo 2(%r0), %r28 /* 2nd case */
535 b lws_exit /* Contended... */
536 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
548 This all works becuse intr_do_signal
549 and schedule both check the return iasq
550 and see that we are on the kernel page
551 so this process is never scheduled off
552 or is ever sent any signal of any sort,
553 thus it is wholly atomic from usrspaces
557 #if defined CONFIG_SMP && defined ENABLE_LWS_DEBUG
560 stw %r1, 4(%sr2,%r20)
562 /* The load and store could fail */
563 1: ldw 0(%sr3,%r26), %r28
564 sub,<> %r28, %r25, %r0
565 2: stw %r24, 0(%sr3,%r26)
568 stw %r20, 0(%sr2,%r20)
569 # ifdef ENABLE_LWS_DEBUG
570 /* Clear thread register indicator */
571 stw %r0, 4(%sr2,%r20)
574 /* Return to userspace, set no error */
579 /* Error occured on load or store */
582 stw %r20, 0(%sr2,%r20)
583 # ifdef ENABLE_LWS_DEBUG
584 stw %r0, 4(%sr2,%r20)
588 ldo -EFAULT(%r0),%r21 /* set errno */
594 /* Two exception table entries, one for the load,
595 the other for the store. Either return -EFAULT.
596 Each of the entries must be relocated. */
597 .section __ex_table,"aw"
599 /* Pad the address calculation */
600 .word 0,(2b - linux_gateway_page)
601 .word 0,(3b - linux_gateway_page)
603 .word (2b - linux_gateway_page)
604 .word (3b - linux_gateway_page)
608 .section __ex_table,"aw"
610 /* Pad the address calculation */
611 .word 0,(1b - linux_gateway_page)
612 .word 0,(3b - linux_gateway_page)
614 .word (1b - linux_gateway_page)
615 .word (3b - linux_gateway_page)
619 end_compare_and_swap:
621 /* Make sure nothing else is placed on this page */
623 .export end_linux_gateway_page
624 end_linux_gateway_page:
626 /* Relocate symbols assuming linux_gateway_page is mapped
627 to virtual address 0x0 */
629 /* FIXME: The code will always be on the gateay page
630 and thus it will be on the first 4k, the
631 assembler seems to think that the final
632 subtraction result is only a word in
633 length, so we pad the value.
635 #define LWS_ENTRY(_name_) .word 0,(lws_##_name_ - linux_gateway_page)
637 #define LWS_ENTRY(_name_) .word (lws_##_name_ - linux_gateway_page)
643 /* Light-weight-syscall table */
644 /* Start of lws table. */
648 LWS_ENTRY(compare_and_swap32) /* 0 - ELF32 Atomic compare and swap */
649 LWS_ENTRY(compare_and_swap64) /* 1 - ELF64 Atomic compare and swap */
650 /* End of lws table */
653 .export sys_call_table
656 #include "syscall_table.S"
660 .export sys_call_table64
663 #define SYSCALL_TABLE_64BIT
664 #include "syscall_table.S"
669 All light-weight-syscall atomic operations
670 will use this set of locks
674 .export lws_lock_start
680 /* Keep locks aligned at 16-bytes */
688 /* CONFIG_SMP for lws_lock_start */