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
32 /* on 64bit pad to 64bit values */
34 #define ULONG_WORD(x) .word 0, x
36 #define ULONG_WORD(x) .word x
42 .import syscall_exit,code
43 .import syscall_exit_rfi,code
45 /* Linux gateway page is aliased to virtual page 0 in the kernel
46 * address space. Since it is a gateway page it cannot be
47 * dereferenced, so null pointers will still fault. We start
48 * the actual entry point at 0x100. We put break instructions
49 * at the beginning of the page to trap null indirect function
54 ENTRY(linux_gateway_page)
56 /* ADDRESS 0x00 to 0xb0 = 176 bytes / 4 bytes per insn = 44 insns */
61 /* ADDRESS 0xb0 to 0xb4, lws uses 1 insns for entry */
62 /* Light-weight-syscall entry must always be located at 0xb0 */
63 /* WARNING: Keep this number updated with table size changes */
64 #define __NR_lws_entries (2)
67 /* Unconditional branch to lws_start, located on the
71 /* Fill from 0xb4 to 0xe0 */
76 /* This function MUST be located at 0xe0 for glibc's threading
77 mechanism to work. DO NOT MOVE THIS CODE EVER! */
79 gate .+8, %r0 /* increase privilege */
80 depi 3, 31, 2, %r31 /* Ensure we return into user mode. */
81 be 0(%sr7,%r31) /* return to user space */
82 mtctl %r26, %cr27 /* move arg0 to the control register */
84 /* Increase the chance of trapping if random jumps occur to this
85 address, fill from 0xf0 to 0x100 */
90 /* This address must remain fixed at 0x100 for glibc's syscalls to work */
93 gate .+8, %r0 /* become privileged */
94 mtsp %r0,%sr4 /* get kernel space into sr4 */
95 mtsp %r0,%sr5 /* get kernel space into sr5 */
96 mtsp %r0,%sr6 /* get kernel space into sr6 */
97 mfsp %sr7,%r1 /* save user sr7 */
98 mtsp %r1,%sr3 /* and store it in sr3 */
101 /* for now we can *always* set the W bit on entry to the syscall
102 * since we don't support wide userland processes. We could
103 * also save the current SM other than in r0 and restore it on
104 * exit from the syscall, and also use that value to know
105 * whether to do narrow or wide syscalls. -PB
108 extrd,u %r1,PSW_W_BIT,1,%r1
109 /* sp must be aligned on 4, so deposit the W bit setting into
110 * the bottom of sp temporarily */
113 /* The top halves of argument registers must be cleared on syscall
114 * entry from narrow executable.
116 depdi 0, 31, 32, %r26
117 depdi 0, 31, 32, %r25
118 depdi 0, 31, 32, %r24
119 depdi 0, 31, 32, %r23
120 depdi 0, 31, 32, %r22
121 depdi 0, 31, 32, %r21
125 xor %r1,%r30,%r30 /* ye olde xor trick */
129 ldo THREAD_SZ_ALGN+FRAME_SIZE(%r30),%r30 /* set up kernel stack */
131 /* N.B.: It is critical that we don't set sr7 to 0 until r30
132 * contains a valid kernel stack pointer. It is also
133 * critical that we don't start using the kernel stack
134 * until after sr7 has been set to 0.
137 mtsp %r0,%sr7 /* get kernel space into sr7 */
138 STREGM %r1,FRAME_SIZE(%r30) /* save r1 (usp) here for now */
139 mfctl %cr30,%r1 /* get task ptr in %r1 */
140 LDREG TI_TASK(%r1),%r1
142 /* Save some registers for sigcontext and potential task
143 switch (see entry.S for the details of which ones are
144 saved/restored). TASK_PT_PSW is zeroed so we can see whether
145 a process is on a syscall or not. For an interrupt the real
146 PSW value is stored. This is needed for gdb and sys_ptrace. */
147 STREG %r0, TASK_PT_PSW(%r1)
148 STREG %r2, TASK_PT_GR2(%r1) /* preserve rp */
149 STREG %r19, TASK_PT_GR19(%r1)
151 LDREGM -FRAME_SIZE(%r30), %r2 /* get users sp back */
153 extrd,u %r2,63,1,%r19 /* W hidden in bottom bit */
155 xor %r19,%r2,%r2 /* clear bottom bit */
157 std %r19,TASK_PT_PSW(%r1)
160 STREG %r2, TASK_PT_GR30(%r1) /* ... and save it */
162 STREG %r20, TASK_PT_GR20(%r1) /* Syscall number */
163 STREG %r21, TASK_PT_GR21(%r1)
164 STREG %r22, TASK_PT_GR22(%r1)
165 STREG %r23, TASK_PT_GR23(%r1) /* 4th argument */
166 STREG %r24, TASK_PT_GR24(%r1) /* 3rd argument */
167 STREG %r25, TASK_PT_GR25(%r1) /* 2nd argument */
168 STREG %r26, TASK_PT_GR26(%r1) /* 1st argument */
169 STREG %r27, TASK_PT_GR27(%r1) /* user dp */
170 STREG %r28, TASK_PT_GR28(%r1) /* return value 0 */
171 STREG %r28, TASK_PT_ORIG_R28(%r1) /* return value 0 (saved for signals) */
172 STREG %r29, TASK_PT_GR29(%r1) /* return value 1 */
173 STREG %r31, TASK_PT_GR31(%r1) /* preserve syscall return ptr */
175 ldo TASK_PT_FR0(%r1), %r27 /* save fpregs from the kernel */
176 save_fp %r27 /* or potential task switch */
178 mfctl %cr11, %r27 /* i.e. SAR */
179 STREG %r27, TASK_PT_SAR(%r1)
184 ldo -16(%r30),%r29 /* Reference param save area */
185 copy %r19,%r2 /* W bit back to r2 */
187 /* no need to save these on stack in wide mode because the first 8
188 * args are passed in registers */
189 stw %r22, -52(%r30) /* 5th argument */
190 stw %r21, -56(%r30) /* 6th argument */
193 /* Are we being ptraced? */
195 LDREG TI_TASK(%r1),%r1
196 LDREG TASK_PTRACE(%r1), %r1
197 bb,<,n %r1,31,.Ltracesys
199 /* Note! We cannot use the syscall table that is mapped
200 nearby since the gateway page is mapped execute-only. */
203 ldil L%sys_call_table, %r1
205 addil L%(sys_call_table64-sys_call_table), %r1
206 ldo R%sys_call_table(%r1), %r19
208 ldo R%sys_call_table64(%r1), %r19
210 ldil L%sys_call_table, %r1
211 ldo R%sys_call_table(%r1), %r19
213 comiclr,>>= __NR_Linux_syscalls, %r20, %r0
216 LDREGX %r20(%r19), %r19
218 /* If this is a sys_rt_sigreturn call, and the signal was received
219 * when not in_syscall, then we want to return via syscall_exit_rfi,
220 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
221 * trampoline code in signal.c).
223 ldi __NR_rt_sigreturn,%r2
224 comb,= %r2,%r20,.Lrt_sigreturn
226 ldil L%syscall_exit,%r2
228 ldo R%syscall_exit(%r2),%r2
230 comib,<> 0,%r25,.Lin_syscall
231 ldil L%syscall_exit_rfi,%r2
233 ldo R%syscall_exit_rfi(%r2),%r2
235 /* Note! Because we are not running where we were linked, any
236 calls to functions external to this file must be indirect. To
237 be safe, we apply the opposite rule to functions within this
238 file, with local labels given to them to ensure correctness. */
242 ldil L%syscall_exit,%r1
243 be R%syscall_exit(%sr7,%r1)
244 ldo -ENOSYS(%r0),%r28 /* set errno */
247 /* Warning! This trace code is a virtual duplicate of the code above so be
248 * sure to maintain both! */
251 /* Need to save more registers so the debugger can see where we
252 * are. This saves only the lower 8 bits of PSW, so that the C
253 * bit is still clear on syscalls, and the D bit is set if this
254 * full register save path has been executed. We check the D
255 * bit on syscall_return_rfi to determine which registers to
256 * restore. An interrupt results in a full PSW saved with the
257 * C bit set, a non-straced syscall entry results in C and D clear
260 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
261 LDREG TI_TASK(%r1), %r1
263 STREG %r2,TASK_PT_PSW(%r1) /* Lower 8 bits only!! */
265 STREG %r2,TASK_PT_SR0(%r1)
267 STREG %r2,TASK_PT_SR1(%r1)
269 STREG %r2,TASK_PT_SR2(%r1)
271 STREG %r2,TASK_PT_SR3(%r1)
272 STREG %r2,TASK_PT_SR4(%r1)
273 STREG %r2,TASK_PT_SR5(%r1)
274 STREG %r2,TASK_PT_SR6(%r1)
275 STREG %r2,TASK_PT_SR7(%r1)
276 STREG %r2,TASK_PT_IASQ0(%r1)
277 STREG %r2,TASK_PT_IASQ1(%r1)
278 LDREG TASK_PT_GR31(%r1),%r2
279 STREG %r2,TASK_PT_IAOQ0(%r1)
281 STREG %r2,TASK_PT_IAOQ1(%r1)
282 ldo TASK_REGS(%r1),%r2
284 STREG %r3,PT_GR3(%r2)
285 STREG %r4,PT_GR4(%r2)
286 STREG %r5,PT_GR5(%r2)
287 STREG %r6,PT_GR6(%r2)
288 STREG %r7,PT_GR7(%r2)
289 STREG %r8,PT_GR8(%r2)
290 STREG %r9,PT_GR9(%r2)
291 STREG %r10,PT_GR10(%r2)
292 STREG %r11,PT_GR11(%r2)
293 STREG %r12,PT_GR12(%r2)
294 STREG %r13,PT_GR13(%r2)
295 STREG %r14,PT_GR14(%r2)
296 STREG %r15,PT_GR15(%r2)
297 STREG %r16,PT_GR16(%r2)
298 STREG %r17,PT_GR17(%r2)
299 STREG %r18,PT_GR18(%r2)
300 /* Finished saving things for the debugger */
302 ldil L%syscall_trace,%r1
303 ldil L%tracesys_next,%r2
304 be R%syscall_trace(%sr7,%r1)
305 ldo R%tracesys_next(%r2),%r2
308 ldil L%sys_call_table,%r1
309 ldo R%sys_call_table(%r1), %r19
311 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
312 LDREG TI_TASK(%r1), %r1
313 LDREG TASK_PT_GR20(%r1), %r20
314 LDREG TASK_PT_GR26(%r1), %r26 /* Restore the users args */
315 LDREG TASK_PT_GR25(%r1), %r25
316 LDREG TASK_PT_GR24(%r1), %r24
317 LDREG TASK_PT_GR23(%r1), %r23
319 LDREG TASK_PT_GR22(%r1), %r22
320 LDREG TASK_PT_GR21(%r1), %r21
321 ldo -16(%r30),%r29 /* Reference param save area */
324 comiclr,>>= __NR_Linux_syscalls, %r20, %r0
327 LDREGX %r20(%r19), %r19
329 /* If this is a sys_rt_sigreturn call, and the signal was received
330 * when not in_syscall, then we want to return via syscall_exit_rfi,
331 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
332 * trampoline code in signal.c).
334 ldi __NR_rt_sigreturn,%r2
335 comb,= %r2,%r20,.Ltrace_rt_sigreturn
337 ldil L%tracesys_exit,%r2
339 ldo R%tracesys_exit(%r2),%r2
341 /* Do *not* call this function on the gateway page, because it
342 makes a direct call to syscall_trace. */
345 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
346 LDREG TI_TASK(%r1), %r1
348 ldo -16(%r30),%r29 /* Reference param save area */
350 bl syscall_trace, %r2
351 STREG %r28,TASK_PT_GR28(%r1) /* save return value now */
352 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
353 LDREG TI_TASK(%r1), %r1
354 LDREG TASK_PT_GR28(%r1), %r28 /* Restore return val. */
356 ldil L%syscall_exit,%r1
357 be,n R%syscall_exit(%sr7,%r1)
359 .Ltrace_rt_sigreturn:
360 comib,<> 0,%r25,.Ltrace_in_syscall
361 ldil L%tracesys_sigexit,%r2
363 ldo R%tracesys_sigexit(%r2),%r2
366 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
369 ldo -16(%r30),%r29 /* Reference param save area */
371 bl syscall_trace, %r2
374 ldil L%syscall_exit_rfi,%r1
375 be,n R%syscall_exit_rfi(%sr7,%r1)
378 /*********************************************************
379 Light-weight-syscall code
382 r26,r25,r24,r23,r22 - Input registers
383 r28 - Function return register
386 Scracth: Any of the above that aren't being
387 currently used, including r1.
389 Return pointer: r31 (Not usable)
391 Error codes returned by entry path:
393 ENOSYS - r20 was an invalid LWS number.
395 *********************************************************/
397 /* Gate and ensure we return to userspace */
399 depi 3, 31, 2, %r31 /* Ensure we return to userspace */
402 /* FIXME: If we are a 64-bit kernel just
403 * turn this on unconditionally.
406 extrd,u %r1,PSW_W_BIT,1,%r1
407 /* sp must be aligned on 4, so deposit the W bit setting into
408 * the bottom of sp temporarily */
411 /* Clip LWS number to a 32-bit value always */
412 depdi 0, 31, 32, %r20
415 /* Is the lws entry number valid? */
416 comiclr,>>= __NR_lws_entries, %r20, %r0
419 /* WARNING: Trashing sr2 and sr3 */
420 mfsp %sr7,%r1 /* get userspace into sr3 */
422 mtsp %r0,%sr2 /* get kernel space into sr2 */
424 /* Load table start */
425 ldil L%lws_table, %r1
426 ldo R%lws_table(%r1), %r28 /* Scratch use of r28 */
427 LDREGX %r20(%sr2,r28), %r21 /* Scratch use of r21 */
429 /* Jump to lws, lws table pointers already relocated */
433 ldo -ENOSYS(%r0),%r21 /* set errno */
434 /* Fall through: Return to userspace */
438 /* decide whether to reset the wide mode bit
440 * For a syscall, the W bit is stored in the lowest bit
441 * of sp. Extract it and reset W if it is zero */
442 extrd,u,*<> %r30,63,1,%r1
444 /* now reset the lowest bit of sp if it was set */
451 /***************************************************
452 Implementing CAS as an atomic operation:
454 %r26 - Address to examine
455 %r25 - Old value to check (old)
456 %r24 - New value to set (new)
457 %r28 - Return prev through this register.
458 %r21 - Kernel error code
460 If debugging is DISabled:
462 %r21 has the following meanings:
464 EAGAIN - CAS is busy, ldcw failed, try again.
465 EFAULT - Read or write failed.
467 If debugging is enabled:
469 EDEADLOCK - CAS called recursively.
470 EAGAIN && r28 == 1 - CAS is busy. Lock contended.
471 EAGAIN && r28 == 2 - CAS is busy. ldcw failed.
472 EFAULT - Read or write failed.
474 Scratch: r20, r28, r1
476 ****************************************************/
478 /* Do not enable LWS debugging */
479 #define ENABLE_LWS_DEBUG 0
481 /* ELF64 Process entry path */
482 lws_compare_and_swap64:
484 b,n lws_compare_and_swap
486 /* If we are not a 64-bit kernel, then we don't
487 * implement having 64-bit input registers
492 /* ELF32 Process entry path */
493 lws_compare_and_swap32:
495 /* Clip all the input registers */
496 depdi 0, 31, 32, %r26
497 depdi 0, 31, 32, %r25
498 depdi 0, 31, 32, %r24
501 lws_compare_and_swap:
503 /* Load start of lock table */
504 ldil L%lws_lock_start, %r20
505 ldo R%lws_lock_start(%r20), %r28
507 /* Extract four bits from r26 and hash lock (Bits 4-7) */
508 extru %r26, 27, 4, %r20
510 /* Find lock to use, the hash is either one of 0 to
511 15, multiplied by 16 (keep it 16-byte aligned)
512 and add to the lock table offset. */
516 # ifdef ENABLE_LWS_DEBUG
518 DEBUG, check for deadlock!
519 If the thread register values are the same
520 then we were the one that locked it last and
521 this is a recurisve call that will deadlock.
522 We *must* giveup this call and fail.
524 ldw 4(%sr2,%r20), %r28 /* Load thread register */
525 /* WARNING: If cr27 cycles to the same value we have problems */
526 mfctl %cr27, %r21 /* Get current thread register */
527 cmpb,<>,n %r21, %r28, cas_lock /* Called recursive? */
528 b lws_exit /* Return error! */
529 ldo -EDEADLOCK(%r0), %r21
531 cmpb,=,n %r0, %r28, cas_nocontend /* Is nobody using it? */
532 ldo 1(%r0), %r28 /* 1st case */
533 b lws_exit /* Contended... */
534 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
537 /* ENABLE_LWS_DEBUG */
539 LDCW 0(%sr2,%r20), %r28 /* Try to acquire the lock */
540 cmpb,<>,n %r0, %r28, cas_action /* Did we get it? */
542 ldo 2(%r0), %r28 /* 2nd case */
543 b lws_exit /* Contended... */
544 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
556 This all works becuse intr_do_signal
557 and schedule both check the return iasq
558 and see that we are on the kernel page
559 so this process is never scheduled off
560 or is ever sent any signal of any sort,
561 thus it is wholly atomic from usrspaces
565 #if defined CONFIG_SMP && defined ENABLE_LWS_DEBUG
568 stw %r1, 4(%sr2,%r20)
570 /* The load and store could fail */
571 1: ldw 0(%sr3,%r26), %r28
572 sub,<> %r28, %r25, %r0
573 2: stw %r24, 0(%sr3,%r26)
576 stw %r20, 0(%sr2,%r20)
577 # ifdef ENABLE_LWS_DEBUG
578 /* Clear thread register indicator */
579 stw %r0, 4(%sr2,%r20)
582 /* Return to userspace, set no error */
587 /* Error occured on load or store */
590 stw %r20, 0(%sr2,%r20)
591 # ifdef ENABLE_LWS_DEBUG
592 stw %r0, 4(%sr2,%r20)
596 ldo -EFAULT(%r0),%r21 /* set errno */
602 /* Two exception table entries, one for the load,
603 the other for the store. Either return -EFAULT.
604 Each of the entries must be relocated. */
605 .section __ex_table,"aw"
606 ULONG_WORD(2b - linux_gateway_page)
607 ULONG_WORD(3b - linux_gateway_page)
610 .section __ex_table,"aw"
611 ULONG_WORD(1b - linux_gateway_page)
612 ULONG_WORD(3b - linux_gateway_page)
615 end_compare_and_swap:
617 /* Make sure nothing else is placed on this page */
619 END(linux_gateway_page)
620 ENTRY(end_linux_gateway_page)
622 /* Relocate symbols assuming linux_gateway_page is mapped
623 to virtual address 0x0 */
625 #define LWS_ENTRY(_name_) ULONG_WORD(lws_##_name_ - linux_gateway_page)
630 /* Light-weight-syscall table */
631 /* Start of lws table. */
633 LWS_ENTRY(compare_and_swap32) /* 0 - ELF32 Atomic compare and swap */
634 LWS_ENTRY(compare_and_swap64) /* 1 - ELF64 Atomic compare and swap */
636 /* End of lws table */
639 ENTRY(sys_call_table)
640 #include "syscall_table.S"
645 ENTRY(sys_call_table64)
646 #define SYSCALL_TABLE_64BIT
647 #include "syscall_table.S"
648 END(sys_call_table64)
653 All light-weight-syscall atomic operations
654 will use this set of locks
658 ENTRY(lws_lock_start)
662 /* Keep locks aligned at 16-bytes */
671 /* CONFIG_SMP for lws_lock_start */