kernel - Fix excessive ipiq recursion (3)
[dragonfly.git] / sys / platform / pc64 / x86_64 / ipl.s
bloba6d609333bf493e94e28748c840e0e6af4e3c7fc
1 /*
2 * Copyright (c) 2008 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
34 * ---
36 * Copyright (c) 1989, 1990 William F. Jolitz.
37 * Copyright (c) 1990 The Regents of the University of California.
38 * All rights reserved.
40 * This code is derived from software contributed to Berkeley by
41 * William Jolitz.
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
45 * are met:
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
51 * 3. Neither the name of the University nor the names of its contributors
52 * may be used to endorse or promote products derived from this software
53 * without specific prior written permission.
55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65 * SUCH DAMAGE.
67 * @(#)ipl.s
69 * $FreeBSD: src/sys/i386/isa/ipl.s,v 1.32.2.3 2002/05/16 16:03:56 bde Exp $
72 #include <machine/asmacros.h>
73 #include <machine/segments.h>
74 #include <machine/ipl.h>
75 #include <machine/lock.h>
76 #include <machine/psl.h>
77 #include <machine/trap.h>
79 #include "assym.s"
82 * AT/386
83 * Vector interrupt control section
85 * ipending - Pending interrupts (set when a masked interrupt occurs)
86 * spending - Pending software interrupts
88 .data
89 ALIGN_DATA
91 .globl fastunpend_count
92 fastunpend_count: .long 0
94 .text
95 SUPERALIGN_TEXT
98 * GENERAL NOTES
100 * - interrupts are always called with a critical section held
102 * - we release our critical section when scheduling interrupt
103 * or softinterrupt threads in order so they can preempt
104 * (unless we are called manually from a critical section, in
105 * which case there will still be a critical section and
106 * they won't preempt anyway).
108 * - TD_NEST_COUNT prevents splz from nesting too deeply within
109 * itself. It is *not* actually an interrupt nesting count.
110 * PCPU(intr_nesting_level) is an interrupt nesting count.
112 * - We have to be careful in regards to local interrupts
113 * occuring simultaniously with our doreti and splz
114 * processing.
116 * - Interrupts must be enabled when calling higher level
117 * functions in order to avoid deadlocking against things
118 * like smp_invltlb.
122 * DORETI
124 * Handle return from interrupts, traps and syscalls. This function
125 * checks the cpl for unmasked pending interrupts (hardware or soft)
126 * and schedules them if appropriate, then irets.
128 * If we are in a critical section we cannot run any pending ints.
130 * The stack contains a trapframe at the start of doreti.
132 SUPERALIGN_TEXT
133 .globl doreti
134 .type doreti,@function
135 doreti:
136 FAKE_MCOUNT(bintr) /* init "from" bintr -> doreti */
137 movq $0,%rax /* irq mask unavailable due to BGL */
138 movq PCPU(curthread),%rbx
139 cli /* interlock with critical section */
140 cmpl $0,PCPU(reqflags) /* short cut if nothing to do */
141 je 5f
142 testl $-1,TD_CRITCOUNT(%rbx) /* can't unpend if in critical sec */
143 jne 5f
144 incl TD_CRITCOUNT(%rbx) /* force all ints to pending */
145 doreti_next:
146 cli /* re-assert cli on loop */
147 movq %rax,%rcx /* irq mask unavailable due to BGL */
148 notq %rcx
149 testl $RQF_IPIQ,PCPU(reqflags)
150 jnz doreti_ipiq
151 testl $RQF_TIMER,PCPU(reqflags)
152 jnz doreti_timer
154 * check for an unmasked int (3 groups)
156 movq $0,%rdx
157 testq PCPU_E8(ipending,%rdx),%rcx
158 jnz doreti_fast
160 movq $1,%rdx
161 testq PCPU_E8(ipending,%rdx),%rcx
162 jnz doreti_fast
164 movq $2,%rdx
165 testq PCPU_E8(ipending,%rdx),%rcx
166 jnz doreti_fast
168 movl PCPU(spending),%ecx /* check for a pending software int */
169 cmpl $0,%ecx
170 jnz doreti_soft
172 testl $RQF_AST_MASK,PCPU(reqflags) /* any pending ASTs? */
173 jz 2f
175 /* ASTs are only applicable when returning to userland */
176 testb $SEL_RPL_MASK,TF_CS(%rsp)
177 jnz doreti_ast
180 * Nothing left to do, finish up. Interrupts are still disabled.
181 * %eax contains the mask of IRQ's that are not available due to
182 * BGL requirements. We can only clear RQF_INTPEND if *ALL* pending
183 * interrupts have been processed.
185 decl TD_CRITCOUNT(%rbx) /* interlocked with cli */
186 testl %eax,%eax
187 jnz 5f
188 andl $~RQF_INTPEND,PCPU(reqflags)
190 MEXITCOUNT
193 * (interrupts are disabled here)
195 * Restore register and iret. iret can fault on %rip (which is
196 * really stupid). If this occurs we re-fault and vector to
197 * doreti_iret_fault().
199 * ...
200 * can be set from user mode, this can result in a kernel mode
201 * exception. The trap code will revector to the *_fault code
202 * which then sets up a T_PROTFLT signal. If the signal is
203 * sent to userland, sendsig() will automatically clean up all
204 * the segment registers to avoid a loop.
206 .globl doreti_iret
207 .globl doreti_syscall_ret
208 doreti_syscall_ret:
209 POP_FRAME /* registers and %gs (+cli) */
210 /* WARNING: special global doreti_iret is also used by exception.S */
211 doreti_iret:
212 iretq
215 * doreti_iret_fault. Alternative return code for the case where
216 * we get a fault in the doreti_exit code above. trap()
217 * (sys/platform/pc64/x86_64/trap.c) catches this specific * case,
218 * sends the process a signal and continues in the corresponding
219 * place in the code below.
221 * Interrupts are likely disabled due to the above interlock
222 * between cli/iretq. We must enable them before calling any
223 * high level function.
225 ALIGN_TEXT
226 .globl doreti_iret_fault
227 doreti_iret_fault:
228 PUSH_FRAME_NOSWAP
230 movq $T_PROTFLT,TF_TRAPNO(%rsp)
231 movq $0,TF_ERR(%rsp) /* XXX should be the error code */
232 movq $0,TF_ADDR(%rsp)
233 FAKE_MCOUNT(TF_RIP(%rsp))
234 jmp calltrap
237 * Interrupt pending. NOTE: stack context holds frame structure
238 * for interrupt procedure, do not do random pushes or pops!
240 ALIGN_TEXT
241 doreti_fast:
242 andq PCPU_E8(ipending,%rdx),%rcx
244 bsfq %rcx, %rcx /* locate the next dispatchable int */
245 btrq %rcx, PCPU_E8(ipending,%rdx)
246 /* is it really still pending? */
247 jnc doreti_next
249 shlq $6, %rdx
250 orq %rdx, %rcx /* form intr number */
252 pushq %rax /* save IRQ mask unavailable for BGL */
253 /* NOTE: is also CPL in frame */
254 call dofastunpend /* unpend intr %rcx */
255 popq %rax
256 jmp doreti_next
259 * SOFT interrupt pending
261 * Temporarily back-out our critical section to allow an interrupt
262 * preempt us when we schedule it. Bump intr_nesting_level to
263 * prevent the switch code from recursing via splz too deeply.
265 ALIGN_TEXT
266 doreti_soft:
268 bsfl %ecx,%ecx /* locate the next pending softint */
269 btrl %ecx,PCPU(spending) /* make sure its still pending */
270 jnc doreti_next
271 addl $FIRST_SOFTINT,%ecx /* actual intr number */
272 pushq %rax
273 movl %ecx,%edi /* argument to C call */
274 incl TD_NEST_COUNT(%rbx) /* prevent doreti/splz nesting */
275 decl TD_CRITCOUNT(%rbx) /* so we can preempt */
276 call sched_ithd_soft /* YYY must pull in imasks */
277 incl TD_CRITCOUNT(%rbx)
278 decl TD_NEST_COUNT(%rbx)
279 popq %rax
280 jmp doreti_next
283 * AST pending. We clear RQF_AST_SIGNAL automatically, the others
284 * are cleared by the trap as they are processed.
286 * Temporarily back-out our critical section because trap() can be
287 * a long-winded call, and we want to be more syscall-like.
289 * YYY theoretically we can call lwkt_switch directly if all we need
290 * to do is a reschedule.
292 doreti_ast:
293 andl $~RQF_AST_SIGNAL,PCPU(reqflags)
295 movl %eax,%r12d /* save cpl (can't use stack) */
296 movl $T_ASTFLT,TF_TRAPNO(%rsp)
297 movq %rsp,%rdi /* pass frame by ref (%edi = C arg) */
298 decl TD_CRITCOUNT(%rbx)
299 call trap
300 incl TD_CRITCOUNT(%rbx)
301 movl %r12d,%eax /* restore cpl for loop */
302 jmp doreti_next
305 * IPIQ message pending. We clear RQF_IPIQ automatically.
307 doreti_ipiq:
308 movl %eax,%r12d /* save cpl (can't use stack) */
309 incl PCPU(intr_nesting_level)
310 andl $~RQF_IPIQ,PCPU(reqflags)
311 subq %rax,%rax
313 xchgl %eax,PCPU(npoll) /* (atomic op) allow another Xipi */
314 subq $8,%rsp /* trapframe->intrframe */
315 movq %rsp,%rdi /* pass frame by ref (C arg) */
316 call lwkt_process_ipiq_frame
317 addq $8,%rsp /* intrframe->trapframe */
318 decl PCPU(intr_nesting_level)
319 movl %r12d,%eax /* restore cpl for loop */
320 jmp doreti_next
322 doreti_timer:
323 movl %eax,%r12d /* save cpl (can't use stack) */
324 incl PCPU(intr_nesting_level)
325 andl $~RQF_TIMER,PCPU(reqflags)
327 subq $8,%rsp /* trapframe->intrframe */
328 movq %rsp,%rdi /* pass frame by ref (C arg) */
329 call pcpu_timer_process_frame
330 addq $8,%rsp /* intrframe->trapframe */
331 decl PCPU(intr_nesting_level)
332 movl %r12d,%eax /* restore cpl for loop */
333 jmp doreti_next
336 * SPLZ() a C callable procedure to dispatch any unmasked pending
337 * interrupts regardless of critical section nesting. ASTs
338 * are not dispatched.
340 * Use %eax to track those IRQs that could not be processed
341 * due to BGL requirements.
343 SUPERALIGN_TEXT
345 ENTRY(splz)
346 pushfq
347 pushq %rbx
348 movq PCPU(curthread),%rbx
349 incl TD_CRITCOUNT(%rbx)
350 movq $0,%rax
352 splz_next:
354 movq %rax,%rcx /* rcx = ~CPL */
355 notq %rcx
356 testl $RQF_IPIQ,PCPU(reqflags)
357 jnz splz_ipiq
358 testl $RQF_TIMER,PCPU(reqflags)
359 jnz splz_timer
361 * check for an unmasked int (3 groups)
363 movq $0,%rdx
364 testq PCPU_E8(ipending,%rdx),%rcx
365 jnz splz_fast
367 movq $1,%rdx
368 testq PCPU_E8(ipending,%rdx),%rcx
369 jnz splz_fast
371 movq $2,%rdx
372 testq PCPU_E8(ipending,%rdx),%rcx
373 jnz splz_fast
375 movl PCPU(spending),%ecx
376 cmpl $0,%ecx
377 jnz splz_soft
379 decl TD_CRITCOUNT(%rbx)
382 * Nothing left to do, finish up. Interrupts are still disabled.
383 * If our mask of IRQs we couldn't process due to BGL requirements
384 * is 0 then there are no pending interrupt sources left and we
385 * can clear RQF_INTPEND.
387 testl %eax,%eax
388 jnz 5f
389 andl $~RQF_INTPEND,PCPU(reqflags)
391 popq %rbx
392 popfq
396 * Interrupt pending
398 ALIGN_TEXT
399 splz_fast:
400 andq PCPU_E8(ipending,%rdx),%rcx
402 bsfq %rcx, %rcx /* locate the next dispatchable int */
403 btrq %rcx, PCPU_E8(ipending,%rdx)
404 /* is it really still pending? */
405 jnc splz_next
407 shlq $6, %rdx
408 orq %rdx, %rcx /* form intr number */
410 pushq %rax
411 call dofastunpend /* unpend intr %rcx */
412 popq %rax
413 jmp splz_next
416 * SOFT interrupt pending
418 * Temporarily back-out our critical section to allow the interrupt
419 * preempt us.
421 ALIGN_TEXT
422 splz_soft:
424 bsfl %ecx,%ecx /* locate the next pending softint */
425 btrl %ecx,PCPU(spending) /* make sure its still pending */
426 jnc splz_next
427 addl $FIRST_SOFTINT,%ecx /* actual intr number */
429 pushq %rax
430 movl %ecx,%edi /* C argument */
431 incl TD_NEST_COUNT(%rbx) /* prevent doreti/splz nesting */
432 decl TD_CRITCOUNT(%rbx)
433 call sched_ithd_soft /* YYY must pull in imasks */
434 incl TD_CRITCOUNT(%rbx)
435 decl TD_NEST_COUNT(%rbx) /* prevent doreti/splz nesting */
436 popq %rax
437 jmp splz_next
439 splz_ipiq:
440 andl $~RQF_IPIQ,PCPU(reqflags)
442 pushq %rax
443 subq %rax,%rax
444 xchgl %eax,PCPU(npoll) /* (atomic op) allow another Xipi */
445 call lwkt_process_ipiq
446 popq %rax
447 jmp splz_next
449 splz_timer:
450 andl $~RQF_TIMER,PCPU(reqflags)
452 pushq %rax
453 call pcpu_timer_process
454 popq %rax
455 jmp splz_next
458 * dofastunpend(%rcx:intr)
460 * A interrupt previously made pending can now be run,
461 * execute it by pushing a dummy interrupt frame and
462 * calling ithread_fast_handler to execute or schedule it.
464 * ithread_fast_handler() returns 0 if it wants us to unmask
465 * further interrupts.
467 #define PUSH_DUMMY \
468 pushfq ; /* phys int frame / flags */ \
469 xorq %rax,%rax ; /* something not SEL_UPL */ \
470 pushq %rax ; /* phys int frame / cs */ \
471 pushq 3*8(%rsp) ; /* original caller eip */ \
472 subq $TF_RIP,%rsp ; /* trap frame */ \
473 movq $0,TF_XFLAGS(%rsp) ; /* extras */ \
474 movq $0,TF_TRAPNO(%rsp) ; /* extras */ \
475 movq $0,TF_ADDR(%rsp) ; /* extras */ \
476 movq $0,TF_FLAGS(%rsp) ; /* extras */ \
477 movq $0,TF_ERR(%rsp) ; /* extras */ \
479 #define POP_DUMMY \
480 addq $TF_RIP+(3*8),%rsp ; \
482 dofastunpend:
483 pushq %rbp /* frame for backtrace */
484 movq %rsp,%rbp
485 PUSH_DUMMY
486 pushq %rcx /* last part of intrframe = intr */
487 incl fastunpend_count
488 movq %rsp,%rdi /* pass frame by reference C arg */
489 call ithread_fast_handler /* returns 0 to unmask */
490 popq %rdi /* intrframe->trapframe */
491 /* + also rdi C arg to next call */
492 cmpl $0,%eax
493 jnz 1f
494 movq MachIntrABI + MACHINTR_INTREN, %rax
495 callq *%rax /* MachIntrABI.intren(intr) */
497 POP_DUMMY
498 popq %rbp