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..
9 #include <linux/config.h> /* for CONFIG_SMP */
11 #include <asm/asm-offsets.h>
12 #include <asm/unistd.h>
13 #include <asm/errno.h>
15 #include <asm/thread_info.h>
17 #include <asm/assembly.h>
18 #include <asm/processor.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
34 .import syscall_exit,code
35 .import syscall_exit_rfi,code
36 .export linux_gateway_page
38 /* Linux gateway page is aliased to virtual page 0 in the kernel
39 * address space. Since it is a gateway page it cannot be
40 * dereferenced, so null pointers will still fault. We start
41 * the actual entry point at 0x100. We put break instructions
42 * at the beginning of the page to trap null indirect function
49 /* ADDRESS 0x00 to 0xb0 = 176 bytes / 4 bytes per insn = 44 insns */
54 /* ADDRESS 0xb0 to 0xb4, lws uses 1 insns for entry */
55 /* Light-weight-syscall entry must always be located at 0xb0 */
56 /* WARNING: Keep this number updated with table size changes */
57 #define __NR_lws_entries (2)
60 /* Unconditional branch to lws_start, located on the
64 /* Fill from 0xb4 to 0xe0 */
69 /* This function MUST be located at 0xe0 for glibc's threading
70 mechanism to work. DO NOT MOVE THIS CODE EVER! */
72 gate .+8, %r0 /* increase privilege */
73 depi 3, 31, 2, %r31 /* Ensure we return into user mode. */
74 be 0(%sr7,%r31) /* return to user space */
75 mtctl %r26, %cr27 /* move arg0 to the control register */
77 /* Increase the chance of trapping if random jumps occur to this
78 address, fill from 0xf0 to 0x100 */
83 /* This address must remain fixed at 0x100 for glibc's syscalls to work */
86 gate .+8, %r0 /* become privileged */
87 mtsp %r0,%sr4 /* get kernel space into sr4 */
88 mtsp %r0,%sr5 /* get kernel space into sr5 */
89 mtsp %r0,%sr6 /* get kernel space into sr6 */
90 mfsp %sr7,%r1 /* save user sr7 */
91 mtsp %r1,%sr3 /* and store it in sr3 */
94 /* for now we can *always* set the W bit on entry to the syscall
95 * since we don't support wide userland processes. We could
96 * also save the current SM other than in r0 and restore it on
97 * exit from the syscall, and also use that value to know
98 * whether to do narrow or wide syscalls. -PB
101 extrd,u %r1,PSW_W_BIT,1,%r1
102 /* sp must be aligned on 4, so deposit the W bit setting into
103 * the bottom of sp temporarily */
106 /* The top halves of argument registers must be cleared on syscall
107 * entry from narrow executable.
109 depdi 0, 31, 32, %r26
110 depdi 0, 31, 32, %r25
111 depdi 0, 31, 32, %r24
112 depdi 0, 31, 32, %r23
113 depdi 0, 31, 32, %r22
114 depdi 0, 31, 32, %r21
118 xor %r1,%r30,%r30 /* ye olde xor trick */
122 ldo THREAD_SZ_ALGN+FRAME_SIZE(%r30),%r30 /* set up kernel stack */
124 /* N.B.: It is critical that we don't set sr7 to 0 until r30
125 * contains a valid kernel stack pointer. It is also
126 * critical that we don't start using the kernel stack
127 * until after sr7 has been set to 0.
130 mtsp %r0,%sr7 /* get kernel space into sr7 */
131 STREGM %r1,FRAME_SIZE(%r30) /* save r1 (usp) here for now */
132 mfctl %cr30,%r1 /* get task ptr in %r1 */
133 LDREG TI_TASK(%r1),%r1
135 /* Save some registers for sigcontext and potential task
136 switch (see entry.S for the details of which ones are
137 saved/restored). TASK_PT_PSW is zeroed so we can see whether
138 a process is on a syscall or not. For an interrupt the real
139 PSW value is stored. This is needed for gdb and sys_ptrace. */
140 STREG %r0, TASK_PT_PSW(%r1)
141 STREG %r2, TASK_PT_GR2(%r1) /* preserve rp */
142 STREG %r19, TASK_PT_GR19(%r1)
144 LDREGM -FRAME_SIZE(%r30), %r2 /* get users sp back */
146 extrd,u %r2,63,1,%r19 /* W hidden in bottom bit */
148 xor %r19,%r2,%r2 /* clear bottom bit */
150 std %r19,TASK_PT_PSW(%r1)
153 STREG %r2, TASK_PT_GR30(%r1) /* ... and save it */
155 STREG %r20, TASK_PT_GR20(%r1) /* Syscall number */
156 STREG %r21, TASK_PT_GR21(%r1)
157 STREG %r22, TASK_PT_GR22(%r1)
158 STREG %r23, TASK_PT_GR23(%r1) /* 4th argument */
159 STREG %r24, TASK_PT_GR24(%r1) /* 3rd argument */
160 STREG %r25, TASK_PT_GR25(%r1) /* 2nd argument */
161 STREG %r26, TASK_PT_GR26(%r1) /* 1st argument */
162 STREG %r27, TASK_PT_GR27(%r1) /* user dp */
163 STREG %r28, TASK_PT_GR28(%r1) /* return value 0 */
164 STREG %r28, TASK_PT_ORIG_R28(%r1) /* return value 0 (saved for signals) */
165 STREG %r29, TASK_PT_GR29(%r1) /* return value 1 */
166 STREG %r31, TASK_PT_GR31(%r1) /* preserve syscall return ptr */
168 ldo TASK_PT_FR0(%r1), %r27 /* save fpregs from the kernel */
169 save_fp %r27 /* or potential task switch */
171 mfctl %cr11, %r27 /* i.e. SAR */
172 STREG %r27, TASK_PT_SAR(%r1)
177 ldo -16(%r30),%r29 /* Reference param save area */
178 copy %r19,%r2 /* W bit back to r2 */
180 /* no need to save these on stack in wide mode because the first 8
181 * args are passed in registers */
182 stw %r22, -52(%r30) /* 5th argument */
183 stw %r21, -56(%r30) /* 6th argument */
186 /* Are we being ptraced? */
188 LDREG TI_TASK(%r1),%r1
189 LDREG TASK_PTRACE(%r1), %r1
190 bb,<,n %r1,31,.Ltracesys
192 /* Note! We cannot use the syscall table that is mapped
193 nearby since the gateway page is mapped execute-only. */
196 ldil L%sys_call_table, %r1
198 addil L%(sys_call_table64-sys_call_table), %r1
199 ldo R%sys_call_table(%r1), %r19
201 ldo R%sys_call_table64(%r1), %r19
203 ldil L%sys_call_table, %r1
204 ldo R%sys_call_table(%r1), %r19
206 comiclr,>>= __NR_Linux_syscalls, %r20, %r0
209 LDREGX %r20(%r19), %r19
211 /* If this is a sys_rt_sigreturn call, and the signal was received
212 * when not in_syscall, then we want to return via syscall_exit_rfi,
213 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
214 * trampoline code in signal.c).
216 ldi __NR_rt_sigreturn,%r2
217 comb,= %r2,%r20,.Lrt_sigreturn
219 ldil L%syscall_exit,%r2
221 ldo R%syscall_exit(%r2),%r2
223 comib,<> 0,%r25,.Lin_syscall
224 ldil L%syscall_exit_rfi,%r2
226 ldo R%syscall_exit_rfi(%r2),%r2
228 /* Note! Because we are not running where we were linked, any
229 calls to functions external to this file must be indirect. To
230 be safe, we apply the opposite rule to functions within this
231 file, with local labels given to them to ensure correctness. */
235 ldil L%syscall_exit,%r1
236 be R%syscall_exit(%sr7,%r1)
237 ldo -ENOSYS(%r0),%r28 /* set errno */
240 /* Warning! This trace code is a virtual duplicate of the code above so be
241 * sure to maintain both! */
244 /* Need to save more registers so the debugger can see where we
245 * are. This saves only the lower 8 bits of PSW, so that the C
246 * bit is still clear on syscalls, and the D bit is set if this
247 * full register save path has been executed. We check the D
248 * bit on syscall_return_rfi to determine which registers to
249 * restore. An interrupt results in a full PSW saved with the
250 * C bit set, a non-straced syscall entry results in C and D clear
253 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
254 LDREG TI_TASK(%r1), %r1
256 STREG %r2,TASK_PT_PSW(%r1) /* Lower 8 bits only!! */
258 STREG %r2,TASK_PT_SR0(%r1)
260 STREG %r2,TASK_PT_SR1(%r1)
262 STREG %r2,TASK_PT_SR2(%r1)
264 STREG %r2,TASK_PT_SR3(%r1)
265 STREG %r2,TASK_PT_SR4(%r1)
266 STREG %r2,TASK_PT_SR5(%r1)
267 STREG %r2,TASK_PT_SR6(%r1)
268 STREG %r2,TASK_PT_SR7(%r1)
269 STREG %r2,TASK_PT_IASQ0(%r1)
270 STREG %r2,TASK_PT_IASQ1(%r1)
271 LDREG TASK_PT_GR31(%r1),%r2
272 STREG %r2,TASK_PT_IAOQ0(%r1)
274 STREG %r2,TASK_PT_IAOQ1(%r1)
275 ldo TASK_REGS(%r1),%r2
277 STREG %r3,PT_GR3(%r2)
278 STREG %r4,PT_GR4(%r2)
279 STREG %r5,PT_GR5(%r2)
280 STREG %r6,PT_GR6(%r2)
281 STREG %r7,PT_GR7(%r2)
282 STREG %r8,PT_GR8(%r2)
283 STREG %r9,PT_GR9(%r2)
284 STREG %r10,PT_GR10(%r2)
285 STREG %r11,PT_GR11(%r2)
286 STREG %r12,PT_GR12(%r2)
287 STREG %r13,PT_GR13(%r2)
288 STREG %r14,PT_GR14(%r2)
289 STREG %r15,PT_GR15(%r2)
290 STREG %r16,PT_GR16(%r2)
291 STREG %r17,PT_GR17(%r2)
292 STREG %r18,PT_GR18(%r2)
293 /* Finished saving things for the debugger */
295 ldil L%syscall_trace,%r1
296 ldil L%tracesys_next,%r2
297 be R%syscall_trace(%sr7,%r1)
298 ldo R%tracesys_next(%r2),%r2
301 ldil L%sys_call_table,%r1
302 ldo R%sys_call_table(%r1), %r19
304 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
305 LDREG TI_TASK(%r1), %r1
306 LDREG TASK_PT_GR20(%r1), %r20
307 LDREG TASK_PT_GR26(%r1), %r26 /* Restore the users args */
308 LDREG TASK_PT_GR25(%r1), %r25
309 LDREG TASK_PT_GR24(%r1), %r24
310 LDREG TASK_PT_GR23(%r1), %r23
312 LDREG TASK_PT_GR22(%r1), %r22
313 LDREG TASK_PT_GR21(%r1), %r21
314 ldo -16(%r30),%r29 /* Reference param save area */
317 comiclr,>>= __NR_Linux_syscalls, %r20, %r0
320 LDREGX %r20(%r19), %r19
322 /* If this is a sys_rt_sigreturn call, and the signal was received
323 * when not in_syscall, then we want to return via syscall_exit_rfi,
324 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
325 * trampoline code in signal.c).
327 ldi __NR_rt_sigreturn,%r2
328 comb,= %r2,%r20,.Ltrace_rt_sigreturn
330 ldil L%tracesys_exit,%r2
332 ldo R%tracesys_exit(%r2),%r2
334 /* Do *not* call this function on the gateway page, because it
335 makes a direct call to syscall_trace. */
338 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
339 LDREG TI_TASK(%r1), %r1
341 ldo -16(%r30),%r29 /* Reference param save area */
343 bl syscall_trace, %r2
344 STREG %r28,TASK_PT_GR28(%r1) /* save return value now */
345 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
346 LDREG TI_TASK(%r1), %r1
347 LDREG TASK_PT_GR28(%r1), %r28 /* Restore return val. */
349 ldil L%syscall_exit,%r1
350 be,n R%syscall_exit(%sr7,%r1)
352 .Ltrace_rt_sigreturn:
353 comib,<> 0,%r25,.Ltrace_in_syscall
354 ldil L%tracesys_sigexit,%r2
356 ldo R%tracesys_sigexit(%r2),%r2
359 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
362 ldo -16(%r30),%r29 /* Reference param save area */
364 bl syscall_trace, %r2
367 ldil L%syscall_exit_rfi,%r1
368 be,n R%syscall_exit_rfi(%sr7,%r1)
371 /*********************************************************
372 Light-weight-syscall code
375 r26,r25,r24,r23,r22 - Input registers
376 r28 - Function return register
379 Scracth: Any of the above that aren't being
380 currently used, including r1.
382 Return pointer: r31 (Not usable)
384 Error codes returned by entry path:
386 ENOSYS - r20 was an invalid LWS number.
388 *********************************************************/
390 /* Gate and ensure we return to userspace */
392 depi 3, 31, 2, %r31 /* Ensure we return to userspace */
395 /* FIXME: If we are a 64-bit kernel just
396 * turn this on unconditionally.
399 extrd,u %r1,PSW_W_BIT,1,%r1
400 /* sp must be aligned on 4, so deposit the W bit setting into
401 * the bottom of sp temporarily */
404 /* Clip LWS number to a 32-bit value always */
405 depdi 0, 31, 32, %r20
408 /* Is the lws entry number valid? */
409 comiclr,>>= __NR_lws_entries, %r20, %r0
412 /* WARNING: Trashing sr2 and sr3 */
413 mfsp %sr7,%r1 /* get userspace into sr3 */
415 mtsp %r0,%sr2 /* get kernel space into sr2 */
417 /* Load table start */
418 ldil L%lws_table, %r1
419 ldo R%lws_table(%r1), %r28 /* Scratch use of r28 */
420 LDREGX %r20(%sr2,r28), %r21 /* Scratch use of r21 */
422 /* Jump to lws, lws table pointers already relocated */
426 ldo -ENOSYS(%r0),%r21 /* set errno */
427 /* Fall through: Return to userspace */
431 /* decide whether to reset the wide mode bit
433 * For a syscall, the W bit is stored in the lowest bit
434 * of sp. Extract it and reset W if it is zero */
435 extrd,u,*<> %r30,63,1,%r1
437 /* now reset the lowest bit of sp if it was set */
444 /***************************************************
445 Implementing CAS as an atomic operation:
447 %r26 - Address to examine
448 %r25 - Old value to check (old)
449 %r24 - New value to set (new)
450 %r28 - Return prev through this register.
451 %r21 - Kernel error code
453 If debugging is DISabled:
455 %r21 has the following meanings:
457 EAGAIN - CAS is busy, ldcw failed, try again.
458 EFAULT - Read or write failed.
460 If debugging is enabled:
462 EDEADLOCK - CAS called recursively.
463 EAGAIN && r28 == 1 - CAS is busy. Lock contended.
464 EAGAIN && r28 == 2 - CAS is busy. ldcw failed.
465 EFAULT - Read or write failed.
467 Scratch: r20, r28, r1
469 ****************************************************/
471 /* Do not enable LWS debugging */
472 #define ENABLE_LWS_DEBUG 0
474 /* ELF64 Process entry path */
475 lws_compare_and_swap64:
477 b,n lws_compare_and_swap
479 /* If we are not a 64-bit kernel, then we don't
480 * implement having 64-bit input registers
485 /* ELF32 Process entry path */
486 lws_compare_and_swap32:
488 /* Clip all the input registers */
489 depdi 0, 31, 32, %r26
490 depdi 0, 31, 32, %r25
491 depdi 0, 31, 32, %r24
494 lws_compare_and_swap:
496 /* Load start of lock table */
497 ldil L%lws_lock_start, %r20
498 ldo R%lws_lock_start(%r20), %r28
500 /* Extract four bits from r26 and hash lock (Bits 4-7) */
501 extru %r26, 27, 4, %r20
503 /* Find lock to use, the hash is either one of 0 to
504 15, multiplied by 16 (keep it 16-byte aligned)
505 and add to the lock table offset. */
509 # ifdef ENABLE_LWS_DEBUG
511 DEBUG, check for deadlock!
512 If the thread register values are the same
513 then we were the one that locked it last and
514 this is a recurisve call that will deadlock.
515 We *must* giveup this call and fail.
517 ldw 4(%sr2,%r20), %r28 /* Load thread register */
518 /* WARNING: If cr27 cycles to the same value we have problems */
519 mfctl %cr27, %r21 /* Get current thread register */
520 cmpb,<>,n %r21, %r28, cas_lock /* Called recursive? */
521 b lws_exit /* Return error! */
522 ldo -EDEADLOCK(%r0), %r21
524 cmpb,=,n %r0, %r28, cas_nocontend /* Is nobody using it? */
525 ldo 1(%r0), %r28 /* 1st case */
526 b lws_exit /* Contended... */
527 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
530 /* ENABLE_LWS_DEBUG */
532 LDCW 0(%sr2,%r20), %r28 /* Try to acquire the lock */
533 cmpb,<>,n %r0, %r28, cas_action /* Did we get it? */
535 ldo 2(%r0), %r28 /* 2nd case */
536 b lws_exit /* Contended... */
537 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
549 This all works becuse intr_do_signal
550 and schedule both check the return iasq
551 and see that we are on the kernel page
552 so this process is never scheduled off
553 or is ever sent any signal of any sort,
554 thus it is wholly atomic from usrspaces
558 #if defined CONFIG_SMP && defined ENABLE_LWS_DEBUG
561 stw %r1, 4(%sr2,%r20)
563 /* The load and store could fail */
564 1: ldw 0(%sr3,%r26), %r28
565 sub,<> %r28, %r25, %r0
566 2: stw %r24, 0(%sr3,%r26)
569 stw %r20, 0(%sr2,%r20)
570 # ifdef ENABLE_LWS_DEBUG
571 /* Clear thread register indicator */
572 stw %r0, 4(%sr2,%r20)
575 /* Return to userspace, set no error */
580 /* Error occured on load or store */
583 stw %r20, 0(%sr2,%r20)
584 # ifdef ENABLE_LWS_DEBUG
585 stw %r0, 4(%sr2,%r20)
589 ldo -EFAULT(%r0),%r21 /* set errno */
595 /* Two exception table entries, one for the load,
596 the other for the store. Either return -EFAULT.
597 Each of the entries must be relocated. */
598 .section __ex_table,"aw"
600 /* Pad the address calculation */
601 .word 0,(2b - linux_gateway_page)
602 .word 0,(3b - linux_gateway_page)
604 .word (2b - linux_gateway_page)
605 .word (3b - linux_gateway_page)
609 .section __ex_table,"aw"
611 /* Pad the address calculation */
612 .word 0,(1b - linux_gateway_page)
613 .word 0,(3b - linux_gateway_page)
615 .word (1b - linux_gateway_page)
616 .word (3b - linux_gateway_page)
620 end_compare_and_swap:
622 /* Make sure nothing else is placed on this page */
624 .export end_linux_gateway_page
625 end_linux_gateway_page:
627 /* Relocate symbols assuming linux_gateway_page is mapped
628 to virtual address 0x0 */
630 /* FIXME: The code will always be on the gateay page
631 and thus it will be on the first 4k, the
632 assembler seems to think that the final
633 subtraction result is only a word in
634 length, so we pad the value.
636 #define LWS_ENTRY(_name_) .word 0,(lws_##_name_ - linux_gateway_page)
638 #define LWS_ENTRY(_name_) .word (lws_##_name_ - linux_gateway_page)
644 /* Light-weight-syscall table */
645 /* Start of lws table. */
649 LWS_ENTRY(compare_and_swap32) /* 0 - ELF32 Atomic compare and swap */
650 LWS_ENTRY(compare_and_swap64) /* 1 - ELF64 Atomic compare and swap */
651 /* End of lws table */
654 .export sys_call_table
657 #include "syscall_table.S"
661 .export sys_call_table64
664 #define SYSCALL_TABLE_64BIT
665 #include "syscall_table.S"
670 All light-weight-syscall atomic operations
671 will use this set of locks
675 .export lws_lock_start
681 /* Keep locks aligned at 16-bytes */
689 /* CONFIG_SMP for lws_lock_start */