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 0xb8, lws uses two 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 gate lws_start, %r0 /* increase privilege */
57 depi 3, 31, 2, %r31 /* Ensure we return into user mode. */
59 /* Fill from 0xb8 to 0xe0 */
64 /* This function MUST be located at 0xe0 for glibc's threading
65 mechanism to work. DO NOT MOVE THIS CODE EVER! */
67 gate .+8, %r0 /* increase privilege */
68 depi 3, 31, 2, %r31 /* Ensure we return into user mode. */
69 be 0(%sr7,%r31) /* return to user space */
70 mtctl %r26, %cr27 /* move arg0 to the control register */
72 /* Increase the chance of trapping if random jumps occur to this
73 address, fill from 0xf0 to 0x100 */
78 /* This address must remain fixed at 0x100 for glibc's syscalls to work */
81 gate .+8, %r0 /* become privileged */
82 mtsp %r0,%sr4 /* get kernel space into sr4 */
83 mtsp %r0,%sr5 /* get kernel space into sr5 */
84 mtsp %r0,%sr6 /* get kernel space into sr6 */
87 /* for now we can *always* set the W bit on entry to the syscall
88 * since we don't support wide userland processes. We could
89 * also save the current SM other than in r0 and restore it on
90 * exit from the syscall, and also use that value to know
91 * whether to do narrow or wide syscalls. -PB
94 extrd,u %r1,PSW_W_BIT,1,%r1
95 /* sp must be aligned on 4, so deposit the W bit setting into
96 * the bottom of sp temporarily */
99 /* The top halves of argument registers must be cleared on syscall
100 * entry from narrow executable.
102 depdi 0, 31, 32, %r26
103 depdi 0, 31, 32, %r25
104 depdi 0, 31, 32, %r24
105 depdi 0, 31, 32, %r23
106 depdi 0, 31, 32, %r22
107 depdi 0, 31, 32, %r21
111 /* We use a rsm/ssm pair to prevent sr3 from being clobbered
112 * by external interrupts.
114 mfsp %sr7,%r1 /* save user sr7 */
115 rsm PSW_SM_I, %r0 /* disable interrupts */
116 mtsp %r1,%sr3 /* and store it in sr3 */
119 xor %r1,%r30,%r30 /* ye olde xor trick */
123 ldo THREAD_SZ_ALGN+FRAME_SIZE(%r30),%r30 /* set up kernel stack */
125 /* N.B.: It is critical that we don't set sr7 to 0 until r30
126 * contains a valid kernel stack pointer. It is also
127 * critical that we don't start using the kernel stack
128 * until after sr7 has been set to 0.
131 mtsp %r0,%sr7 /* get kernel space into sr7 */
132 ssm PSW_SM_I, %r0 /* enable interrupts */
133 STREGM %r1,FRAME_SIZE(%r30) /* save r1 (usp) here for now */
134 mfctl %cr30,%r1 /* get task ptr in %r1 */
135 LDREG TI_TASK(%r1),%r1
137 /* Save some registers for sigcontext and potential task
138 switch (see entry.S for the details of which ones are
139 saved/restored). TASK_PT_PSW is zeroed so we can see whether
140 a process is on a syscall or not. For an interrupt the real
141 PSW value is stored. This is needed for gdb and sys_ptrace. */
142 STREG %r0, TASK_PT_PSW(%r1)
143 STREG %r2, TASK_PT_GR2(%r1) /* preserve rp */
144 STREG %r19, TASK_PT_GR19(%r1)
146 LDREGM -FRAME_SIZE(%r30), %r2 /* get users sp back */
148 extrd,u %r2,63,1,%r19 /* W hidden in bottom bit */
150 xor %r19,%r2,%r2 /* clear bottom bit */
152 std %r19,TASK_PT_PSW(%r1)
155 STREG %r2, TASK_PT_GR30(%r1) /* ... and save it */
157 STREG %r20, TASK_PT_GR20(%r1) /* Syscall number */
158 STREG %r21, TASK_PT_GR21(%r1)
159 STREG %r22, TASK_PT_GR22(%r1)
160 STREG %r23, TASK_PT_GR23(%r1) /* 4th argument */
161 STREG %r24, TASK_PT_GR24(%r1) /* 3rd argument */
162 STREG %r25, TASK_PT_GR25(%r1) /* 2nd argument */
163 STREG %r26, TASK_PT_GR26(%r1) /* 1st argument */
164 STREG %r27, TASK_PT_GR27(%r1) /* user dp */
165 STREG %r28, TASK_PT_GR28(%r1) /* return value 0 */
166 STREG %r28, TASK_PT_ORIG_R28(%r1) /* return value 0 (saved for signals) */
167 STREG %r29, TASK_PT_GR29(%r1) /* return value 1 */
168 STREG %r31, TASK_PT_GR31(%r1) /* preserve syscall return ptr */
170 ldo TASK_PT_FR0(%r1), %r27 /* save fpregs from the kernel */
171 save_fp %r27 /* or potential task switch */
173 mfctl %cr11, %r27 /* i.e. SAR */
174 STREG %r27, TASK_PT_SAR(%r1)
179 ldo -16(%r30),%r29 /* Reference param save area */
180 copy %r19,%r2 /* W bit back to r2 */
182 /* no need to save these on stack in wide mode because the first 8
183 * args are passed in registers */
184 stw %r22, -52(%r30) /* 5th argument */
185 stw %r21, -56(%r30) /* 6th argument */
188 /* Are we being ptraced? */
190 LDREG TI_TASK(%r1),%r1
191 ldw TASK_PTRACE(%r1), %r1
192 bb,<,n %r1,31,.Ltracesys
194 /* Note! We cannot use the syscall table that is mapped
195 nearby since the gateway page is mapped execute-only. */
198 ldil L%sys_call_table, %r1
200 addil L%(sys_call_table64-sys_call_table), %r1
201 ldo R%sys_call_table(%r1), %r19
203 ldo R%sys_call_table64(%r1), %r19
205 ldil L%sys_call_table, %r1
206 ldo R%sys_call_table(%r1), %r19
208 comiclr,>> __NR_Linux_syscalls, %r20, %r0
211 LDREGX %r20(%r19), %r19
213 /* If this is a sys_rt_sigreturn call, and the signal was received
214 * when not in_syscall, then we want to return via syscall_exit_rfi,
215 * not syscall_exit. Signal no. in r20, in_syscall in r25 (see
216 * trampoline code in signal.c).
218 ldi __NR_rt_sigreturn,%r2
219 comb,= %r2,%r20,.Lrt_sigreturn
221 ldil L%syscall_exit,%r2
223 ldo R%syscall_exit(%r2),%r2
225 comib,<> 0,%r25,.Lin_syscall
226 ldil L%syscall_exit_rfi,%r2
228 ldo R%syscall_exit_rfi(%r2),%r2
230 /* Note! Because we are not running where we were linked, any
231 calls to functions external to this file must be indirect. To
232 be safe, we apply the opposite rule to functions within this
233 file, with local labels given to them to ensure correctness. */
237 ldil L%syscall_exit,%r1
238 be R%syscall_exit(%sr7,%r1)
239 ldo -ENOSYS(%r0),%r28 /* set errno */
242 /* Warning! This trace code is a virtual duplicate of the code above so be
243 * sure to maintain both! */
246 /* Need to save more registers so the debugger can see where we
247 * are. This saves only the lower 8 bits of PSW, so that the C
248 * bit is still clear on syscalls, and the D bit is set if this
249 * full register save path has been executed. We check the D
250 * bit on syscall_return_rfi to determine which registers to
251 * restore. An interrupt results in a full PSW saved with the
252 * C bit set, a non-straced syscall entry results in C and D clear
255 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
256 LDREG TI_TASK(%r1), %r1
258 STREG %r2,TASK_PT_PSW(%r1) /* Lower 8 bits only!! */
260 STREG %r2,TASK_PT_SR0(%r1)
262 STREG %r2,TASK_PT_SR1(%r1)
264 STREG %r2,TASK_PT_SR2(%r1)
266 STREG %r2,TASK_PT_SR3(%r1)
267 STREG %r2,TASK_PT_SR4(%r1)
268 STREG %r2,TASK_PT_SR5(%r1)
269 STREG %r2,TASK_PT_SR6(%r1)
270 STREG %r2,TASK_PT_SR7(%r1)
271 STREG %r2,TASK_PT_IASQ0(%r1)
272 STREG %r2,TASK_PT_IASQ1(%r1)
273 LDREG TASK_PT_GR31(%r1),%r2
274 STREG %r2,TASK_PT_IAOQ0(%r1)
276 STREG %r2,TASK_PT_IAOQ1(%r1)
277 ldo TASK_REGS(%r1),%r2
279 STREG %r3,PT_GR3(%r2)
280 STREG %r4,PT_GR4(%r2)
281 STREG %r5,PT_GR5(%r2)
282 STREG %r6,PT_GR6(%r2)
283 STREG %r7,PT_GR7(%r2)
284 STREG %r8,PT_GR8(%r2)
285 STREG %r9,PT_GR9(%r2)
286 STREG %r10,PT_GR10(%r2)
287 STREG %r11,PT_GR11(%r2)
288 STREG %r12,PT_GR12(%r2)
289 STREG %r13,PT_GR13(%r2)
290 STREG %r14,PT_GR14(%r2)
291 STREG %r15,PT_GR15(%r2)
292 STREG %r16,PT_GR16(%r2)
293 STREG %r17,PT_GR17(%r2)
294 STREG %r18,PT_GR18(%r2)
295 /* Finished saving things for the debugger */
298 ldil L%do_syscall_trace_enter,%r1
299 ldil L%tracesys_next,%r2
300 be R%do_syscall_trace_enter(%sr7,%r1)
301 ldo R%tracesys_next(%r2),%r2
304 /* do_syscall_trace_enter either returned the syscallno, or -1L,
305 * so we skip restoring the PT_GR20 below, since we pulled it from
306 * task->thread.regs.gr[20] above.
309 ldil L%sys_call_table,%r1
310 ldo R%sys_call_table(%r1), %r19
312 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
313 LDREG TI_TASK(%r1), %r1
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 ldo TASK_REGS(%r1),%r26
351 bl do_syscall_trace_exit,%r2
352 STREG %r28,TASK_PT_GR28(%r1) /* save return value now */
353 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
354 LDREG TI_TASK(%r1), %r1
355 LDREG TASK_PT_GR28(%r1), %r28 /* Restore return val. */
357 ldil L%syscall_exit,%r1
358 be,n R%syscall_exit(%sr7,%r1)
360 .Ltrace_rt_sigreturn:
361 comib,<> 0,%r25,.Ltrace_in_syscall
362 ldil L%tracesys_sigexit,%r2
364 ldo R%tracesys_sigexit(%r2),%r2
367 ldo -THREAD_SZ_ALGN-FRAME_SIZE(%r30),%r1 /* get task ptr */
368 LDREG TI_TASK(%r1), %r1
370 ldo -16(%r30),%r29 /* Reference param save area */
372 bl do_syscall_trace_exit,%r2
373 ldo TASK_REGS(%r1),%r26
375 ldil L%syscall_exit_rfi,%r1
376 be,n R%syscall_exit_rfi(%sr7,%r1)
379 /*********************************************************
380 32/64-bit Light-Weight-Syscall ABI
382 * - Indicates a hint for userspace inline asm
385 Syscall number (caller-saves)
389 Argument registers (caller-saves)
390 - %r26, %r25, %r24, %r23, %r22
393 Return registers (caller-saves)
394 - %r28 (return), %r21 (errno)
397 Caller-saves registers
399 - %r2 (return pointer)
400 - %r31 (ble link register)
403 Callee-saves registers
405 - %r30 (stack pointer)
406 * Not in asm clobber.
408 If userspace is 32-bit:
409 Callee-saves registers
410 - %r19 (32-bit PIC register)
412 Differences from 32-bit calling convention:
413 - Syscall number in %r20
414 - Additional argument register %r22 (arg4)
417 If userspace is 64-bit:
418 Callee-saves registers
419 - %r27 (64-bit PIC register)
421 Differences from 64-bit calling convention:
422 - Syscall number in %r20
423 - Additional argument register %r22 (arg4)
426 Error codes returned by entry path:
428 ENOSYS - r20 was an invalid LWS number.
430 *********************************************************/
434 /* FIXME: If we are a 64-bit kernel just
435 * turn this on unconditionally.
438 extrd,u %r1,PSW_W_BIT,1,%r1
439 /* sp must be aligned on 4, so deposit the W bit setting into
440 * the bottom of sp temporarily */
443 /* Clip LWS number to a 32-bit value always */
444 depdi 0, 31, 32, %r20
447 /* Is the lws entry number valid? */
448 comiclr,>> __NR_lws_entries, %r20, %r0
451 /* WARNING: Trashing sr2 and sr3 */
452 mfsp %sr7,%r1 /* get userspace into sr3 */
454 mtsp %r0,%sr2 /* get kernel space into sr2 */
456 /* Load table start */
457 ldil L%lws_table, %r1
458 ldo R%lws_table(%r1), %r28 /* Scratch use of r28 */
459 LDREGX %r20(%sr2,r28), %r21 /* Scratch use of r21 */
461 /* Jump to lws, lws table pointers already relocated */
465 ldo -ENOSYS(%r0),%r21 /* set errno */
466 /* Fall through: Return to userspace */
470 /* decide whether to reset the wide mode bit
472 * For a syscall, the W bit is stored in the lowest bit
473 * of sp. Extract it and reset W if it is zero */
474 extrd,u,*<> %r30,63,1,%r1
476 /* now reset the lowest bit of sp if it was set */
483 /***************************************************
484 Implementing CAS as an atomic operation:
486 %r26 - Address to examine
487 %r25 - Old value to check (old)
488 %r24 - New value to set (new)
489 %r28 - Return prev through this register.
490 %r21 - Kernel error code
492 If debugging is DISabled:
494 %r21 has the following meanings:
496 EAGAIN - CAS is busy, ldcw failed, try again.
497 EFAULT - Read or write failed.
499 If debugging is enabled:
501 EDEADLOCK - CAS called recursively.
502 EAGAIN && r28 == 1 - CAS is busy. Lock contended.
503 EAGAIN && r28 == 2 - CAS is busy. ldcw failed.
504 EFAULT - Read or write failed.
506 Scratch: r20, r28, r1
508 ****************************************************/
510 /* Do not enable LWS debugging */
511 #define ENABLE_LWS_DEBUG 0
513 /* ELF64 Process entry path */
514 lws_compare_and_swap64:
516 b,n lws_compare_and_swap
518 /* If we are not a 64-bit kernel, then we don't
519 * have 64-bit input registers, and calling
520 * the 64-bit LWS CAS returns ENOSYS.
525 /* ELF32 Process entry path */
526 lws_compare_and_swap32:
528 /* Clip all the input registers */
529 depdi 0, 31, 32, %r26
530 depdi 0, 31, 32, %r25
531 depdi 0, 31, 32, %r24
534 lws_compare_and_swap:
535 /* Load start of lock table */
536 ldil L%lws_lock_start, %r20
537 ldo R%lws_lock_start(%r20), %r28
539 /* Extract four bits from r26 and hash lock (Bits 4-7) */
540 extru %r26, 27, 4, %r20
542 /* Find lock to use, the hash is either one of 0 to
543 15, multiplied by 16 (keep it 16-byte aligned)
544 and add to the lock table offset. */
548 # if ENABLE_LWS_DEBUG
550 DEBUG, check for deadlock!
551 If the thread register values are the same
552 then we were the one that locked it last and
553 this is a recurisve call that will deadlock.
554 We *must* giveup this call and fail.
556 ldw 4(%sr2,%r20), %r28 /* Load thread register */
557 /* WARNING: If cr27 cycles to the same value we have problems */
558 mfctl %cr27, %r21 /* Get current thread register */
559 cmpb,<>,n %r21, %r28, cas_lock /* Called recursive? */
560 b lws_exit /* Return error! */
561 ldo -EDEADLOCK(%r0), %r21
563 cmpb,=,n %r0, %r28, cas_nocontend /* Is nobody using it? */
564 ldo 1(%r0), %r28 /* 1st case */
565 b lws_exit /* Contended... */
566 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
569 /* ENABLE_LWS_DEBUG */
571 LDCW 0(%sr2,%r20), %r28 /* Try to acquire the lock */
572 cmpb,<>,n %r0, %r28, cas_action /* Did we get it? */
574 ldo 2(%r0), %r28 /* 2nd case */
575 b lws_exit /* Contended... */
576 ldo -EAGAIN(%r0), %r21 /* Spin in userspace */
586 This all works becuse intr_do_signal
587 and schedule both check the return iasq
588 and see that we are on the kernel page
589 so this process is never scheduled off
590 or is ever sent any signal of any sort,
591 thus it is wholly atomic from usrspaces
595 #if defined CONFIG_SMP && ENABLE_LWS_DEBUG
598 stw %r1, 4(%sr2,%r20)
600 /* The load and store could fail */
601 1: ldw 0(%sr3,%r26), %r28
602 sub,<> %r28, %r25, %r0
603 2: stw %r24, 0(%sr3,%r26)
605 stw %r20, 0(%sr2,%r20)
607 /* Clear thread register indicator */
608 stw %r0, 4(%sr2,%r20)
610 /* Return to userspace, set no error */
615 /* Error occurred on load or store */
617 stw %r20, 0(%sr2,%r20)
619 stw %r0, 4(%sr2,%r20)
622 ldo -EFAULT(%r0),%r21 /* set errno */
628 /* Two exception table entries, one for the load,
629 the other for the store. Either return -EFAULT.
630 Each of the entries must be relocated. */
631 .section __ex_table,"aw"
632 ASM_ULONG_INSN (1b - linux_gateway_page), (3b - linux_gateway_page)
633 ASM_ULONG_INSN (2b - linux_gateway_page), (3b - linux_gateway_page)
637 /* Make sure nothing else is placed on this page */
639 END(linux_gateway_page)
640 ENTRY(end_linux_gateway_page)
642 /* Relocate symbols assuming linux_gateway_page is mapped
643 to virtual address 0x0 */
645 #define LWS_ENTRY(_name_) ASM_ULONG_INSN (lws_##_name_ - linux_gateway_page)
650 /* Light-weight-syscall table */
651 /* Start of lws table. */
653 LWS_ENTRY(compare_and_swap32) /* 0 - ELF32 Atomic compare and swap */
654 LWS_ENTRY(compare_and_swap64) /* 1 - ELF64 Atomic compare and swap */
656 /* End of lws table */
659 ENTRY(sys_call_table)
660 #include "syscall_table.S"
665 ENTRY(sys_call_table64)
666 #define SYSCALL_TABLE_64BIT
667 #include "syscall_table.S"
668 END(sys_call_table64)
672 All light-weight-syscall atomic operations
673 will use this set of locks
675 NOTE: The lws_lock_start symbol must be
676 at least 16-byte aligned for safe use
681 ENTRY(lws_lock_start)
684 /* Keep locks aligned at 16-bytes */