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>
20 #include <linux/init.h>
22 /* We fill the empty parts of the gateway page with
23 * something that will kill the kernel or a
24 * userspace application.
26 #define KILL_INSN break 0,0
32 .import syscall_exit,code
33 .import syscall_exit_rfi,code
35 /* Linux gateway page is aliased to virtual page 0 in the kernel
36 * address space. Since it is a gateway page it cannot be
37 * dereferenced, so null pointers will still fault. We start
38 * the actual entry point at 0x100. We put break instructions
39 * at the beginning of the page to trap null indirect function
44 ENTRY(linux_gateway_page)
46 /* ADDRESS 0x00 to 0xb0 = 176 bytes / 4 bytes per insn = 44 insns */
51 /* ADDRESS 0xb0 to 0xb4, lws uses 1 insns for entry */
52 /* Light-weight-syscall entry must always be located at 0xb0 */
53 /* WARNING: Keep this number updated with table size changes */
54 #define __NR_lws_entries (2)
57 /* Unconditional branch to lws_start, located on the
61 /* Fill from 0xb4 to 0xe0 */
66 /* This function MUST be located at 0xe0 for glibc's threading
67 mechanism to work. DO NOT MOVE THIS CODE EVER! */
69 gate .+8, %r0 /* increase privilege */
70 depi 3, 31, 2, %r31 /* Ensure we return into user mode. */
71 be 0(%sr7,%r31) /* return to user space */
72 mtctl %r26, %cr27 /* move arg0 to the control register */
74 /* Increase the chance of trapping if random jumps occur to this
75 address, fill from 0xf0 to 0x100 */
80 /* This address must remain fixed at 0x100 for glibc's syscalls to work */
83 gate .+8, %r0 /* become privileged */
84 mtsp %r0,%sr4 /* get kernel space into sr4 */
85 mtsp %r0,%sr5 /* get kernel space into sr5 */
86 mtsp %r0,%sr6 /* get kernel space into sr6 */
87 mfsp %sr7,%r1 /* save user sr7 */
88 mtsp %r1,%sr3 /* and store it in sr3 */
91 /* for now we can *always* set the W bit on entry to the syscall
92 * since we don't support wide userland processes. We could
93 * also save the current SM other than in r0 and restore it on
94 * exit from the syscall, and also use that value to know
95 * whether to do narrow or wide syscalls. -PB
98 extrd,u %r1,PSW_W_BIT,1,%r1
99 /* sp must be aligned on 4, so deposit the W bit setting into
100 * the bottom of sp temporarily */
103 /* The top halves of argument registers must be cleared on syscall
104 * entry from narrow executable.
106 depdi 0, 31, 32, %r26
107 depdi 0, 31, 32, %r25
108 depdi 0, 31, 32, %r24
109 depdi 0, 31, 32, %r23
110 depdi 0, 31, 32, %r22
111 depdi 0, 31, 32, %r21
115 xor %r1,%r30,%r30 /* ye olde xor trick */
119 ldo THREAD_SZ_ALGN+FRAME_SIZE(%r30),%r30 /* set up kernel stack */
121 /* N.B.: It is critical that we don't set sr7 to 0 until r30
122 * contains a valid kernel stack pointer. It is also
123 * critical that we don't start using the kernel stack
124 * until after sr7 has been set to 0.
127 mtsp %r0,%sr7 /* get kernel space into sr7 */
128 STREGM %r1,FRAME_SIZE(%r30) /* save r1 (usp) here for now */
129 mfctl %cr30,%r1 /* get task ptr in %r1 */
130 LDREG TI_TASK(%r1),%r1
132 /* Save some registers for sigcontext and potential task
133 switch (see entry.S for the details of which ones are
134 saved/restored). TASK_PT_PSW is zeroed so we can see whether
135 a process is on a syscall or not. For an interrupt the real
136 PSW value is stored. This is needed for gdb and sys_ptrace. */
137 STREG %r0, TASK_PT_PSW(%r1)
138 STREG %r2, TASK_PT_GR2(%r1) /* preserve rp */
139 STREG %r19, TASK_PT_GR19(%r1)
141 LDREGM -FRAME_SIZE(%r30), %r2 /* get users sp back */
143 extrd,u %r2,63,1,%r19 /* W hidden in bottom bit */
145 xor %r19,%r2,%r2 /* clear bottom bit */
147 std %r19,TASK_PT_PSW(%r1)
150 STREG %r2, TASK_PT_GR30(%r1) /* ... and save it */
152 STREG %r20, TASK_PT_GR20(%r1) /* Syscall number */
153 STREG %r21, TASK_PT_GR21(%r1)
154 STREG %r22, TASK_PT_GR22(%r1)
155 STREG %r23, TASK_PT_GR23(%r1) /* 4th argument */
156 STREG %r24, TASK_PT_GR24(%r1) /* 3rd argument */
157 STREG %r25, TASK_PT_GR25(%r1) /* 2nd argument */
158 STREG %r26, TASK_PT_GR26(%r1) /* 1st argument */
159 STREG %r27, TASK_PT_GR27(%r1) /* user dp */
160 STREG %r28, TASK_PT_GR28(%r1) /* return value 0 */
161 STREG %r28, TASK_PT_ORIG_R28(%r1) /* return value 0 (saved for signals) */
162 STREG %r29, TASK_PT_GR29(%r1) /* return value 1 */
163 STREG %r31, TASK_PT_GR31(%r1) /* preserve syscall return ptr */
165 ldo TASK_PT_FR0(%r1), %r27 /* save fpregs from the kernel */
166 save_fp %r27 /* or potential task switch */
168 mfctl %cr11, %r27 /* i.e. SAR */
169 STREG %r27, TASK_PT_SAR(%r1)
174 ldo -16(%r30),%r29 /* Reference param save area */
175 copy %r19,%r2 /* W bit back to r2 */
177 /* no need to save these on stack in wide mode because the first 8
178 * args are passed in registers */
179 stw %r22, -52(%r30) /* 5th argument */
180 stw %r21, -56(%r30) /* 6th argument */
183 /* Are we being ptraced? */
185 LDREG TI_TASK(%r1),%r1
186 ldw TASK_PTRACE(%r1), %r1
187 bb,<,n %r1,31,.Ltracesys
189 /* Note! We cannot use the syscall table that is mapped
190 nearby since the gateway page is mapped execute-only. */
193 ldil L%sys_call_table, %r1
195 addil L%(sys_call_table64-sys_call_table), %r1
196 ldo R%sys_call_table(%r1), %r19
198 ldo R%sys_call_table64(%r1), %r19
200 ldil L%sys_call_table, %r1
201 ldo R%sys_call_table(%r1), %r19
203 comiclr,>> __NR_Linux_syscalls, %r20, %r0
206 LDREGX %r20(%r19), %r19
208 /* If this is a sys_rt_sigreturn call, and the signal was received
209 * when not in_syscall, then we want to return via syscall_exit_rfi,
210 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
211 * trampoline code in signal.c).
213 ldi __NR_rt_sigreturn,%r2
214 comb,= %r2,%r20,.Lrt_sigreturn
216 ldil L%syscall_exit,%r2
218 ldo R%syscall_exit(%r2),%r2
220 comib,<> 0,%r25,.Lin_syscall
221 ldil L%syscall_exit_rfi,%r2
223 ldo R%syscall_exit_rfi(%r2),%r2
225 /* Note! Because we are not running where we were linked, any
226 calls to functions external to this file must be indirect. To
227 be safe, we apply the opposite rule to functions within this
228 file, with local labels given to them to ensure correctness. */
232 ldil L%syscall_exit,%r1
233 be R%syscall_exit(%sr7,%r1)
234 ldo -ENOSYS(%r0),%r28 /* set errno */
237 /* Warning! This trace code is a virtual duplicate of the code above so be
238 * sure to maintain both! */
241 /* Need to save more registers so the debugger can see where we
242 * are. This saves only the lower 8 bits of PSW, so that the C
243 * bit is still clear on syscalls, and the D bit is set if this
244 * full register save path has been executed. We check the D
245 * bit on syscall_return_rfi to determine which registers to
246 * restore. An interrupt results in a full PSW saved with the
247 * C bit set, a non-straced syscall entry results in C and D clear
250 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
251 LDREG TI_TASK(%r1), %r1
253 STREG %r2,TASK_PT_PSW(%r1) /* Lower 8 bits only!! */
255 STREG %r2,TASK_PT_SR0(%r1)
257 STREG %r2,TASK_PT_SR1(%r1)
259 STREG %r2,TASK_PT_SR2(%r1)
261 STREG %r2,TASK_PT_SR3(%r1)
262 STREG %r2,TASK_PT_SR4(%r1)
263 STREG %r2,TASK_PT_SR5(%r1)
264 STREG %r2,TASK_PT_SR6(%r1)
265 STREG %r2,TASK_PT_SR7(%r1)
266 STREG %r2,TASK_PT_IASQ0(%r1)
267 STREG %r2,TASK_PT_IASQ1(%r1)
268 LDREG TASK_PT_GR31(%r1),%r2
269 STREG %r2,TASK_PT_IAOQ0(%r1)
271 STREG %r2,TASK_PT_IAOQ1(%r1)
272 ldo TASK_REGS(%r1),%r2
274 STREG %r3,PT_GR3(%r2)
275 STREG %r4,PT_GR4(%r2)
276 STREG %r5,PT_GR5(%r2)
277 STREG %r6,PT_GR6(%r2)
278 STREG %r7,PT_GR7(%r2)
279 STREG %r8,PT_GR8(%r2)
280 STREG %r9,PT_GR9(%r2)
281 STREG %r10,PT_GR10(%r2)
282 STREG %r11,PT_GR11(%r2)
283 STREG %r12,PT_GR12(%r2)
284 STREG %r13,PT_GR13(%r2)
285 STREG %r14,PT_GR14(%r2)
286 STREG %r15,PT_GR15(%r2)
287 STREG %r16,PT_GR16(%r2)
288 STREG %r17,PT_GR17(%r2)
289 STREG %r18,PT_GR18(%r2)
290 /* Finished saving things for the debugger */
292 ldil L%syscall_trace,%r1
293 ldil L%tracesys_next,%r2
294 be R%syscall_trace(%sr7,%r1)
295 ldo R%tracesys_next(%r2),%r2
298 ldil L%sys_call_table,%r1
299 ldo R%sys_call_table(%r1), %r19
301 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
302 LDREG TI_TASK(%r1), %r1
303 LDREG TASK_PT_GR20(%r1), %r20
304 LDREG TASK_PT_GR26(%r1), %r26 /* Restore the users args */
305 LDREG TASK_PT_GR25(%r1), %r25
306 LDREG TASK_PT_GR24(%r1), %r24
307 LDREG TASK_PT_GR23(%r1), %r23
309 LDREG TASK_PT_GR22(%r1), %r22
310 LDREG TASK_PT_GR21(%r1), %r21
311 ldo -16(%r30),%r29 /* Reference param save area */
314 comiclr,>>= __NR_Linux_syscalls, %r20, %r0
317 LDREGX %r20(%r19), %r19
319 /* If this is a sys_rt_sigreturn call, and the signal was received
320 * when not in_syscall, then we want to return via syscall_exit_rfi,
321 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
322 * trampoline code in signal.c).
324 ldi __NR_rt_sigreturn,%r2
325 comb,= %r2,%r20,.Ltrace_rt_sigreturn
327 ldil L%tracesys_exit,%r2
329 ldo R%tracesys_exit(%r2),%r2
331 /* Do *not* call this function on the gateway page, because it
332 makes a direct call to syscall_trace. */
335 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
336 LDREG TI_TASK(%r1), %r1
338 ldo -16(%r30),%r29 /* Reference param save area */
340 bl syscall_trace, %r2
341 STREG %r28,TASK_PT_GR28(%r1) /* save return value now */
342 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
343 LDREG TI_TASK(%r1), %r1
344 LDREG TASK_PT_GR28(%r1), %r28 /* Restore return val. */
346 ldil L%syscall_exit,%r1
347 be,n R%syscall_exit(%sr7,%r1)
349 .Ltrace_rt_sigreturn:
350 comib,<> 0,%r25,.Ltrace_in_syscall
351 ldil L%tracesys_sigexit,%r2
353 ldo R%tracesys_sigexit(%r2),%r2
356 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
359 ldo -16(%r30),%r29 /* Reference param save area */
361 bl syscall_trace, %r2
364 ldil L%syscall_exit_rfi,%r1
365 be,n R%syscall_exit_rfi(%sr7,%r1)
368 /*********************************************************
369 Light-weight-syscall code
372 r26,r25,r24,r23,r22 - Input registers
373 r28 - Function return register
376 Scracth: Any of the above that aren't being
377 currently used, including r1.
379 Return pointer: r31 (Not usable)
381 Error codes returned by entry path:
383 ENOSYS - r20 was an invalid LWS number.
385 *********************************************************/
387 /* Gate and ensure we return to userspace */
389 depi 3, 31, 2, %r31 /* Ensure we return to userspace */
392 /* FIXME: If we are a 64-bit kernel just
393 * turn this on unconditionally.
396 extrd,u %r1,PSW_W_BIT,1,%r1
397 /* sp must be aligned on 4, so deposit the W bit setting into
398 * the bottom of sp temporarily */
401 /* Clip LWS number to a 32-bit value always */
402 depdi 0, 31, 32, %r20
405 /* Is the lws entry number valid? */
406 comiclr,>>= __NR_lws_entries, %r20, %r0
409 /* WARNING: Trashing sr2 and sr3 */
410 mfsp %sr7,%r1 /* get userspace into sr3 */
412 mtsp %r0,%sr2 /* get kernel space into sr2 */
414 /* Load table start */
415 ldil L%lws_table, %r1
416 ldo R%lws_table(%r1), %r28 /* Scratch use of r28 */
417 LDREGX %r20(%sr2,r28), %r21 /* Scratch use of r21 */
419 /* Jump to lws, lws table pointers already relocated */
423 ldo -ENOSYS(%r0),%r21 /* set errno */
424 /* Fall through: Return to userspace */
428 /* decide whether to reset the wide mode bit
430 * For a syscall, the W bit is stored in the lowest bit
431 * of sp. Extract it and reset W if it is zero */
432 extrd,u,*<> %r30,63,1,%r1
434 /* now reset the lowest bit of sp if it was set */
441 /***************************************************
442 Implementing CAS as an atomic operation:
444 %r26 - Address to examine
445 %r25 - Old value to check (old)
446 %r24 - New value to set (new)
447 %r28 - Return prev through this register.
448 %r21 - Kernel error code
450 If debugging is DISabled:
452 %r21 has the following meanings:
454 EAGAIN - CAS is busy, ldcw failed, try again.
455 EFAULT - Read or write failed.
457 If debugging is enabled:
459 EDEADLOCK - CAS called recursively.
460 EAGAIN && r28 == 1 - CAS is busy. Lock contended.
461 EAGAIN && r28 == 2 - CAS is busy. ldcw failed.
462 EFAULT - Read or write failed.
464 Scratch: r20, r28, r1
466 ****************************************************/
468 /* Do not enable LWS debugging */
469 #define ENABLE_LWS_DEBUG 0
471 /* ELF64 Process entry path */
472 lws_compare_and_swap64:
474 b,n lws_compare_and_swap
476 /* If we are not a 64-bit kernel, then we don't
477 * implement having 64-bit input registers
482 /* ELF32 Process entry path */
483 lws_compare_and_swap32:
485 /* Clip all the input registers */
486 depdi 0, 31, 32, %r26
487 depdi 0, 31, 32, %r25
488 depdi 0, 31, 32, %r24
491 lws_compare_and_swap:
493 /* Load start of lock table */
494 ldil L%lws_lock_start, %r20
495 ldo R%lws_lock_start(%r20), %r28
497 /* Extract four bits from r26 and hash lock (Bits 4-7) */
498 extru %r26, 27, 4, %r20
500 /* Find lock to use, the hash is either one of 0 to
501 15, multiplied by 16 (keep it 16-byte aligned)
502 and add to the lock table offset. */
506 # if ENABLE_LWS_DEBUG
508 DEBUG, check for deadlock!
509 If the thread register values are the same
510 then we were the one that locked it last and
511 this is a recurisve call that will deadlock.
512 We *must* giveup this call and fail.
514 ldw 4(%sr2,%r20), %r28 /* Load thread register */
515 /* WARNING: If cr27 cycles to the same value we have problems */
516 mfctl %cr27, %r21 /* Get current thread register */
517 cmpb,<>,n %r21, %r28, cas_lock /* Called recursive? */
518 b lws_exit /* Return error! */
519 ldo -EDEADLOCK(%r0), %r21
521 cmpb,=,n %r0, %r28, cas_nocontend /* Is nobody using it? */
522 ldo 1(%r0), %r28 /* 1st case */
523 b lws_exit /* Contended... */
524 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
527 /* ENABLE_LWS_DEBUG */
529 LDCW 0(%sr2,%r20), %r28 /* Try to acquire the lock */
530 cmpb,<>,n %r0, %r28, cas_action /* Did we get it? */
532 ldo 2(%r0), %r28 /* 2nd case */
533 b lws_exit /* Contended... */
534 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
546 This all works becuse intr_do_signal
547 and schedule both check the return iasq
548 and see that we are on the kernel page
549 so this process is never scheduled off
550 or is ever sent any signal of any sort,
551 thus it is wholly atomic from usrspaces
555 #if defined CONFIG_SMP && ENABLE_LWS_DEBUG
558 stw %r1, 4(%sr2,%r20)
560 /* The load and store could fail */
561 1: ldw 0(%sr3,%r26), %r28
562 sub,<> %r28, %r25, %r0
563 2: stw %r24, 0(%sr3,%r26)
566 stw %r20, 0(%sr2,%r20)
567 # if ENABLE_LWS_DEBUG
568 /* Clear thread register indicator */
569 stw %r0, 4(%sr2,%r20)
572 /* Return to userspace, set no error */
577 /* Error occured on load or store */
580 stw %r20, 0(%sr2,%r20)
581 # if ENABLE_LWS_DEBUG
582 stw %r0, 4(%sr2,%r20)
586 ldo -EFAULT(%r0),%r21 /* set errno */
592 /* Two exception table entries, one for the load,
593 the other for the store. Either return -EFAULT.
594 Each of the entries must be relocated. */
595 .section __ex_table,"aw"
596 ASM_ULONG_INSN (1b - linux_gateway_page), (3b - linux_gateway_page)
597 ASM_ULONG_INSN (2b - linux_gateway_page), (3b - linux_gateway_page)
601 /* Make sure nothing else is placed on this page */
603 END(linux_gateway_page)
604 ENTRY(end_linux_gateway_page)
606 /* Relocate symbols assuming linux_gateway_page is mapped
607 to virtual address 0x0 */
609 #define LWS_ENTRY(_name_) ASM_ULONG_INSN (lws_##_name_ - linux_gateway_page)
614 /* Light-weight-syscall table */
615 /* Start of lws table. */
617 LWS_ENTRY(compare_and_swap32) /* 0 - ELF32 Atomic compare and swap */
618 LWS_ENTRY(compare_and_swap64) /* 1 - ELF64 Atomic compare and swap */
620 /* End of lws table */
623 ENTRY(sys_call_table)
624 #include "syscall_table.S"
629 ENTRY(sys_call_table64)
630 #define SYSCALL_TABLE_64BIT
631 #include "syscall_table.S"
632 END(sys_call_table64)
637 All light-weight-syscall atomic operations
638 will use this set of locks
642 ENTRY(lws_lock_start)
646 /* Keep locks aligned at 16-bytes */
655 /* CONFIG_SMP for lws_lock_start */