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>
15 #include <asm/thread_info.h>
16 #include <asm/assembly.h>
17 #include <asm/processor.h>
19 #include <linux/linkage.h>
21 /* We fill the empty parts of the gateway page with
22 * something that will kill the kernel or a
23 * userspace application.
25 #define KILL_INSN break 0,0
31 .import syscall_exit,code
32 .import syscall_exit_rfi,code
34 /* Linux gateway page is aliased to virtual page 0 in the kernel
35 * address space. Since it is a gateway page it cannot be
36 * dereferenced, so null pointers will still fault. We start
37 * the actual entry point at 0x100. We put break instructions
38 * at the beginning of the page to trap null indirect function
43 ENTRY(linux_gateway_page)
45 /* ADDRESS 0x00 to 0xb0 = 176 bytes / 4 bytes per insn = 44 insns */
50 /* ADDRESS 0xb0 to 0xb4, lws uses 1 insns for entry */
51 /* Light-weight-syscall entry must always be located at 0xb0 */
52 /* WARNING: Keep this number updated with table size changes */
53 #define __NR_lws_entries (2)
56 /* Unconditional branch to lws_start, located on the
60 /* Fill from 0xb4 to 0xe0 */
65 /* This function MUST be located at 0xe0 for glibc's threading
66 mechanism to work. DO NOT MOVE THIS CODE EVER! */
68 gate .+8, %r0 /* increase privilege */
69 depi 3, 31, 2, %r31 /* Ensure we return into user mode. */
70 be 0(%sr7,%r31) /* return to user space */
71 mtctl %r26, %cr27 /* move arg0 to the control register */
73 /* Increase the chance of trapping if random jumps occur to this
74 address, fill from 0xf0 to 0x100 */
79 /* This address must remain fixed at 0x100 for glibc's syscalls to work */
82 gate .+8, %r0 /* become privileged */
83 mtsp %r0,%sr4 /* get kernel space into sr4 */
84 mtsp %r0,%sr5 /* get kernel space into sr5 */
85 mtsp %r0,%sr6 /* get kernel space into sr6 */
86 mfsp %sr7,%r1 /* save user sr7 */
87 mtsp %r1,%sr3 /* and store it in sr3 */
90 /* for now we can *always* set the W bit on entry to the syscall
91 * since we don't support wide userland processes. We could
92 * also save the current SM other than in r0 and restore it on
93 * exit from the syscall, and also use that value to know
94 * whether to do narrow or wide syscalls. -PB
97 extrd,u %r1,PSW_W_BIT,1,%r1
98 /* sp must be aligned on 4, so deposit the W bit setting into
99 * the bottom of sp temporarily */
102 /* The top halves of argument registers must be cleared on syscall
103 * entry from narrow executable.
105 depdi 0, 31, 32, %r26
106 depdi 0, 31, 32, %r25
107 depdi 0, 31, 32, %r24
108 depdi 0, 31, 32, %r23
109 depdi 0, 31, 32, %r22
110 depdi 0, 31, 32, %r21
114 xor %r1,%r30,%r30 /* ye olde xor trick */
118 ldo THREAD_SZ_ALGN+FRAME_SIZE(%r30),%r30 /* set up kernel stack */
120 /* N.B.: It is critical that we don't set sr7 to 0 until r30
121 * contains a valid kernel stack pointer. It is also
122 * critical that we don't start using the kernel stack
123 * until after sr7 has been set to 0.
126 mtsp %r0,%sr7 /* get kernel space into sr7 */
127 STREGM %r1,FRAME_SIZE(%r30) /* save r1 (usp) here for now */
128 mfctl %cr30,%r1 /* get task ptr in %r1 */
129 LDREG TI_TASK(%r1),%r1
131 /* Save some registers for sigcontext and potential task
132 switch (see entry.S for the details of which ones are
133 saved/restored). TASK_PT_PSW is zeroed so we can see whether
134 a process is on a syscall or not. For an interrupt the real
135 PSW value is stored. This is needed for gdb and sys_ptrace. */
136 STREG %r0, TASK_PT_PSW(%r1)
137 STREG %r2, TASK_PT_GR2(%r1) /* preserve rp */
138 STREG %r19, TASK_PT_GR19(%r1)
140 LDREGM -FRAME_SIZE(%r30), %r2 /* get users sp back */
142 extrd,u %r2,63,1,%r19 /* W hidden in bottom bit */
144 xor %r19,%r2,%r2 /* clear bottom bit */
146 std %r19,TASK_PT_PSW(%r1)
149 STREG %r2, TASK_PT_GR30(%r1) /* ... and save it */
151 STREG %r20, TASK_PT_GR20(%r1) /* Syscall number */
152 STREG %r21, TASK_PT_GR21(%r1)
153 STREG %r22, TASK_PT_GR22(%r1)
154 STREG %r23, TASK_PT_GR23(%r1) /* 4th argument */
155 STREG %r24, TASK_PT_GR24(%r1) /* 3rd argument */
156 STREG %r25, TASK_PT_GR25(%r1) /* 2nd argument */
157 STREG %r26, TASK_PT_GR26(%r1) /* 1st argument */
158 STREG %r27, TASK_PT_GR27(%r1) /* user dp */
159 STREG %r28, TASK_PT_GR28(%r1) /* return value 0 */
160 STREG %r28, TASK_PT_ORIG_R28(%r1) /* return value 0 (saved for signals) */
161 STREG %r29, TASK_PT_GR29(%r1) /* return value 1 */
162 STREG %r31, TASK_PT_GR31(%r1) /* preserve syscall return ptr */
164 ldo TASK_PT_FR0(%r1), %r27 /* save fpregs from the kernel */
165 save_fp %r27 /* or potential task switch */
167 mfctl %cr11, %r27 /* i.e. SAR */
168 STREG %r27, TASK_PT_SAR(%r1)
173 ldo -16(%r30),%r29 /* Reference param save area */
174 copy %r19,%r2 /* W bit back to r2 */
176 /* no need to save these on stack in wide mode because the first 8
177 * args are passed in registers */
178 stw %r22, -52(%r30) /* 5th argument */
179 stw %r21, -56(%r30) /* 6th argument */
182 /* Are we being ptraced? */
184 LDREG TI_TASK(%r1),%r1
185 ldw TASK_PTRACE(%r1), %r1
186 bb,<,n %r1,31,.Ltracesys
188 /* Note! We cannot use the syscall table that is mapped
189 nearby since the gateway page is mapped execute-only. */
192 ldil L%sys_call_table, %r1
194 addil L%(sys_call_table64-sys_call_table), %r1
195 ldo R%sys_call_table(%r1), %r19
197 ldo R%sys_call_table64(%r1), %r19
199 ldil L%sys_call_table, %r1
200 ldo R%sys_call_table(%r1), %r19
202 comiclr,>> __NR_Linux_syscalls, %r20, %r0
205 LDREGX %r20(%r19), %r19
207 /* If this is a sys_rt_sigreturn call, and the signal was received
208 * when not in_syscall, then we want to return via syscall_exit_rfi,
209 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
210 * trampoline code in signal.c).
212 ldi __NR_rt_sigreturn,%r2
213 comb,= %r2,%r20,.Lrt_sigreturn
215 ldil L%syscall_exit,%r2
217 ldo R%syscall_exit(%r2),%r2
219 comib,<> 0,%r25,.Lin_syscall
220 ldil L%syscall_exit_rfi,%r2
222 ldo R%syscall_exit_rfi(%r2),%r2
224 /* Note! Because we are not running where we were linked, any
225 calls to functions external to this file must be indirect. To
226 be safe, we apply the opposite rule to functions within this
227 file, with local labels given to them to ensure correctness. */
231 ldil L%syscall_exit,%r1
232 be R%syscall_exit(%sr7,%r1)
233 ldo -ENOSYS(%r0),%r28 /* set errno */
236 /* Warning! This trace code is a virtual duplicate of the code above so be
237 * sure to maintain both! */
240 /* Need to save more registers so the debugger can see where we
241 * are. This saves only the lower 8 bits of PSW, so that the C
242 * bit is still clear on syscalls, and the D bit is set if this
243 * full register save path has been executed. We check the D
244 * bit on syscall_return_rfi to determine which registers to
245 * restore. An interrupt results in a full PSW saved with the
246 * C bit set, a non-straced syscall entry results in C and D clear
249 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
250 LDREG TI_TASK(%r1), %r1
252 STREG %r2,TASK_PT_PSW(%r1) /* Lower 8 bits only!! */
254 STREG %r2,TASK_PT_SR0(%r1)
256 STREG %r2,TASK_PT_SR1(%r1)
258 STREG %r2,TASK_PT_SR2(%r1)
260 STREG %r2,TASK_PT_SR3(%r1)
261 STREG %r2,TASK_PT_SR4(%r1)
262 STREG %r2,TASK_PT_SR5(%r1)
263 STREG %r2,TASK_PT_SR6(%r1)
264 STREG %r2,TASK_PT_SR7(%r1)
265 STREG %r2,TASK_PT_IASQ0(%r1)
266 STREG %r2,TASK_PT_IASQ1(%r1)
267 LDREG TASK_PT_GR31(%r1),%r2
268 STREG %r2,TASK_PT_IAOQ0(%r1)
270 STREG %r2,TASK_PT_IAOQ1(%r1)
271 ldo TASK_REGS(%r1),%r2
273 STREG %r3,PT_GR3(%r2)
274 STREG %r4,PT_GR4(%r2)
275 STREG %r5,PT_GR5(%r2)
276 STREG %r6,PT_GR6(%r2)
277 STREG %r7,PT_GR7(%r2)
278 STREG %r8,PT_GR8(%r2)
279 STREG %r9,PT_GR9(%r2)
280 STREG %r10,PT_GR10(%r2)
281 STREG %r11,PT_GR11(%r2)
282 STREG %r12,PT_GR12(%r2)
283 STREG %r13,PT_GR13(%r2)
284 STREG %r14,PT_GR14(%r2)
285 STREG %r15,PT_GR15(%r2)
286 STREG %r16,PT_GR16(%r2)
287 STREG %r17,PT_GR17(%r2)
288 STREG %r18,PT_GR18(%r2)
289 /* Finished saving things for the debugger */
291 ldil L%syscall_trace,%r1
292 ldil L%tracesys_next,%r2
293 be R%syscall_trace(%sr7,%r1)
294 ldo R%tracesys_next(%r2),%r2
297 ldil L%sys_call_table,%r1
298 ldo R%sys_call_table(%r1), %r19
300 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
301 LDREG TI_TASK(%r1), %r1
302 LDREG TASK_PT_GR20(%r1), %r20
303 LDREG TASK_PT_GR26(%r1), %r26 /* Restore the users args */
304 LDREG TASK_PT_GR25(%r1), %r25
305 LDREG TASK_PT_GR24(%r1), %r24
306 LDREG TASK_PT_GR23(%r1), %r23
308 LDREG TASK_PT_GR22(%r1), %r22
309 LDREG TASK_PT_GR21(%r1), %r21
310 ldo -16(%r30),%r29 /* Reference param save area */
313 comiclr,>>= __NR_Linux_syscalls, %r20, %r0
316 LDREGX %r20(%r19), %r19
318 /* If this is a sys_rt_sigreturn call, and the signal was received
319 * when not in_syscall, then we want to return via syscall_exit_rfi,
320 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
321 * trampoline code in signal.c).
323 ldi __NR_rt_sigreturn,%r2
324 comb,= %r2,%r20,.Ltrace_rt_sigreturn
326 ldil L%tracesys_exit,%r2
328 ldo R%tracesys_exit(%r2),%r2
330 /* Do *not* call this function on the gateway page, because it
331 makes a direct call to syscall_trace. */
334 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
335 LDREG TI_TASK(%r1), %r1
337 ldo -16(%r30),%r29 /* Reference param save area */
339 bl syscall_trace, %r2
340 STREG %r28,TASK_PT_GR28(%r1) /* save return value now */
341 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
342 LDREG TI_TASK(%r1), %r1
343 LDREG TASK_PT_GR28(%r1), %r28 /* Restore return val. */
345 ldil L%syscall_exit,%r1
346 be,n R%syscall_exit(%sr7,%r1)
348 .Ltrace_rt_sigreturn:
349 comib,<> 0,%r25,.Ltrace_in_syscall
350 ldil L%tracesys_sigexit,%r2
352 ldo R%tracesys_sigexit(%r2),%r2
355 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
358 ldo -16(%r30),%r29 /* Reference param save area */
360 bl syscall_trace, %r2
363 ldil L%syscall_exit_rfi,%r1
364 be,n R%syscall_exit_rfi(%sr7,%r1)
367 /*********************************************************
368 Light-weight-syscall code
371 r26,r25,r24,r23,r22 - Input registers
372 r28 - Function return register
375 Scracth: Any of the above that aren't being
376 currently used, including r1.
378 Return pointer: r31 (Not usable)
380 Error codes returned by entry path:
382 ENOSYS - r20 was an invalid LWS number.
384 *********************************************************/
386 /* Gate and ensure we return to userspace */
388 depi 3, 31, 2, %r31 /* Ensure we return to userspace */
391 /* FIXME: If we are a 64-bit kernel just
392 * turn this on unconditionally.
395 extrd,u %r1,PSW_W_BIT,1,%r1
396 /* sp must be aligned on 4, so deposit the W bit setting into
397 * the bottom of sp temporarily */
400 /* Clip LWS number to a 32-bit value always */
401 depdi 0, 31, 32, %r20
404 /* Is the lws entry number valid? */
405 comiclr,>>= __NR_lws_entries, %r20, %r0
408 /* WARNING: Trashing sr2 and sr3 */
409 mfsp %sr7,%r1 /* get userspace into sr3 */
411 mtsp %r0,%sr2 /* get kernel space into sr2 */
413 /* Load table start */
414 ldil L%lws_table, %r1
415 ldo R%lws_table(%r1), %r28 /* Scratch use of r28 */
416 LDREGX %r20(%sr2,r28), %r21 /* Scratch use of r21 */
418 /* Jump to lws, lws table pointers already relocated */
422 ldo -ENOSYS(%r0),%r21 /* set errno */
423 /* Fall through: Return to userspace */
427 /* decide whether to reset the wide mode bit
429 * For a syscall, the W bit is stored in the lowest bit
430 * of sp. Extract it and reset W if it is zero */
431 extrd,u,*<> %r30,63,1,%r1
433 /* now reset the lowest bit of sp if it was set */
440 /***************************************************
441 Implementing CAS as an atomic operation:
443 %r26 - Address to examine
444 %r25 - Old value to check (old)
445 %r24 - New value to set (new)
446 %r28 - Return prev through this register.
447 %r21 - Kernel error code
449 If debugging is DISabled:
451 %r21 has the following meanings:
453 EAGAIN - CAS is busy, ldcw failed, try again.
454 EFAULT - Read or write failed.
456 If debugging is enabled:
458 EDEADLOCK - CAS called recursively.
459 EAGAIN && r28 == 1 - CAS is busy. Lock contended.
460 EAGAIN && r28 == 2 - CAS is busy. ldcw failed.
461 EFAULT - Read or write failed.
463 Scratch: r20, r28, r1
465 ****************************************************/
467 /* Do not enable LWS debugging */
468 #define ENABLE_LWS_DEBUG 0
470 /* ELF64 Process entry path */
471 lws_compare_and_swap64:
473 b,n lws_compare_and_swap
475 /* If we are not a 64-bit kernel, then we don't
476 * implement having 64-bit input registers
481 /* ELF32 Process entry path */
482 lws_compare_and_swap32:
484 /* Clip all the input registers */
485 depdi 0, 31, 32, %r26
486 depdi 0, 31, 32, %r25
487 depdi 0, 31, 32, %r24
490 lws_compare_and_swap:
492 /* Load start of lock table */
493 ldil L%lws_lock_start, %r20
494 ldo R%lws_lock_start(%r20), %r28
496 /* Extract four bits from r26 and hash lock (Bits 4-7) */
497 extru %r26, 27, 4, %r20
499 /* Find lock to use, the hash is either one of 0 to
500 15, multiplied by 16 (keep it 16-byte aligned)
501 and add to the lock table offset. */
505 # if ENABLE_LWS_DEBUG
507 DEBUG, check for deadlock!
508 If the thread register values are the same
509 then we were the one that locked it last and
510 this is a recurisve call that will deadlock.
511 We *must* giveup this call and fail.
513 ldw 4(%sr2,%r20), %r28 /* Load thread register */
514 /* WARNING: If cr27 cycles to the same value we have problems */
515 mfctl %cr27, %r21 /* Get current thread register */
516 cmpb,<>,n %r21, %r28, cas_lock /* Called recursive? */
517 b lws_exit /* Return error! */
518 ldo -EDEADLOCK(%r0), %r21
520 cmpb,=,n %r0, %r28, cas_nocontend /* Is nobody using it? */
521 ldo 1(%r0), %r28 /* 1st case */
522 b lws_exit /* Contended... */
523 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
526 /* ENABLE_LWS_DEBUG */
528 LDCW 0(%sr2,%r20), %r28 /* Try to acquire the lock */
529 cmpb,<>,n %r0, %r28, cas_action /* Did we get it? */
531 ldo 2(%r0), %r28 /* 2nd case */
532 b lws_exit /* Contended... */
533 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
545 This all works becuse intr_do_signal
546 and schedule both check the return iasq
547 and see that we are on the kernel page
548 so this process is never scheduled off
549 or is ever sent any signal of any sort,
550 thus it is wholly atomic from usrspaces
554 #if defined CONFIG_SMP && ENABLE_LWS_DEBUG
557 stw %r1, 4(%sr2,%r20)
559 /* The load and store could fail */
560 1: ldw 0(%sr3,%r26), %r28
561 sub,<> %r28, %r25, %r0
562 2: stw %r24, 0(%sr3,%r26)
565 stw %r20, 0(%sr2,%r20)
566 # if ENABLE_LWS_DEBUG
567 /* Clear thread register indicator */
568 stw %r0, 4(%sr2,%r20)
571 /* Return to userspace, set no error */
576 /* Error occured on load or store */
579 stw %r20, 0(%sr2,%r20)
580 # if ENABLE_LWS_DEBUG
581 stw %r0, 4(%sr2,%r20)
585 ldo -EFAULT(%r0),%r21 /* set errno */
591 /* Two exception table entries, one for the load,
592 the other for the store. Either return -EFAULT.
593 Each of the entries must be relocated. */
594 .section __ex_table,"aw"
595 ASM_ULONG_INSN (1b - linux_gateway_page), (3b - linux_gateway_page)
596 ASM_ULONG_INSN (2b - linux_gateway_page), (3b - linux_gateway_page)
600 /* Make sure nothing else is placed on this page */
602 END(linux_gateway_page)
603 ENTRY(end_linux_gateway_page)
605 /* Relocate symbols assuming linux_gateway_page is mapped
606 to virtual address 0x0 */
608 #define LWS_ENTRY(_name_) ASM_ULONG_INSN (lws_##_name_ - linux_gateway_page)
613 /* Light-weight-syscall table */
614 /* Start of lws table. */
616 LWS_ENTRY(compare_and_swap32) /* 0 - ELF32 Atomic compare and swap */
617 LWS_ENTRY(compare_and_swap64) /* 1 - ELF64 Atomic compare and swap */
619 /* End of lws table */
622 ENTRY(sys_call_table)
623 #include "syscall_table.S"
628 ENTRY(sys_call_table64)
629 #define SYSCALL_TABLE_64BIT
630 #include "syscall_table.S"
631 END(sys_call_table64)
636 All light-weight-syscall atomic operations
637 will use this set of locks
641 ENTRY(lws_lock_start)
645 /* Keep locks aligned at 16-bytes */
654 /* CONFIG_SMP for lws_lock_start */