SimpleOS

LXR

Navigation



Site hébergé par : enix

The LXR Cross Referencer for SOS

source navigation ]
diff markup ]
identifier search ]
general search ]
 
 
Article:1 ] [ 2 ] [ 3 ] [ 4 ] [ 5 ] [ 6 ] [ 6.5 ] [ 7 ] [ 7.5 ] [ 8 ] [ 9 ] [ 9.5 ]

001 /* Copyright (C) 2005  David Decotigny
002    Copyright (C) 2000-2004, The KOS team
003 
004    This program is free software; you can redistribute it and/or
005    modify it under the terms of the GNU General Public License
006    as published by the Free Software Foundation; either version 2
007    of the License, or (at your option) any later version.
008    
009    This program is distributed in the hope that it will be useful,
010    but WITHOUT ANY WARRANTY; without even the implied warranty of
011    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
012    GNU General Public License for more details.
013    
014    You should have received a copy of the GNU General Public License
015    along with this program; if not, write to the Free Software
016    Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
017    USA. 
018 */
019 
020 
021 #include <sos/assert.h>
022 #include <sos/klibc.h>
023 #include <drivers/bochs.h>
024 #include <drivers/x86_videomem.h>
025 #include <hwcore/segment.h>
026 #include <hwcore/gdt.h>
027 #include <sos/uaccess.h>
028 
029 #include "cpu_context.h"
030 
031 
032 /**
033  * Here is the definition of a CPU context for IA32 processors. This
034  * is a SOS convention, not a specification given by the IA32
035  * spec. However there is a strong constraint related to the x86
036  * interrupt handling specification: the top of the stack MUST be
037  * compatible with the 'iret' instruction, ie there must be the
038  * err_code (might be 0), eip, cs and eflags of the destination
039  * context in that order (see Intel x86 specs vol 3, figure 5-4).
040  *
041  * @note IMPORTANT: This definition MUST be consistent with the way
042  * the registers are stored on the stack in
043  * irq_wrappers.S/exception_wrappers.S !!! Hence the constraint above.
044  */
045 struct sos_cpu_state {
046   /* (Lower addresses) */
047 
048   /* These are SOS convention */
049   sos_ui16_t  gs;
050   sos_ui16_t  fs;
051   sos_ui16_t  es;
052   sos_ui16_t  ds;
053   sos_ui16_t  cpl0_ss; /* This is ALWAYS the Stack Segment of the
054                           Kernel context (CPL0) of the interrupted
055                           thread, even for a user thread */
056   sos_ui16_t  alignment_padding; /* unused */
057   sos_ui32_t  eax;
058   sos_ui32_t  ebx;
059   sos_ui32_t  ecx;
060   sos_ui32_t  edx;
061   sos_ui32_t  esi;
062   sos_ui32_t  edi;
063   sos_ui32_t  ebp;
064 
065   /* MUST NEVER CHANGE (dependent on the IA32 iret instruction) */
066   sos_ui32_t  error_code;
067   sos_vaddr_t eip;
068   sos_ui32_t  cs; /* 32bits according to the specs ! However, the CS
069                      register is really 16bits long */
070   sos_ui32_t  eflags;
071 
072   /* (Higher addresses) */
073 } __attribute__((packed));
074 
075 
076 /**
077  * The CS value pushed on the stack by the CPU upon interrupt, and
078  * needed by the iret instruction, is 32bits long while the real CPU
079  * CS register is 16bits only: this macro simply retrieves the CPU
080  * "CS" register value from the CS value pushed on the stack by the
081  * CPU upon interrupt.
082  *
083  * The remaining 16bits pushed by the CPU should be considered
084  * "reserved" and architecture dependent. IMHO, the specs don't say
085  * anything about them. Considering that some architectures generate
086  * non-zero values for these 16bits (at least Cyrix), we'd better
087  * ignore them.
088  */
089 #define GET_CPU_CS_REGISTER_VALUE(pushed_ui32_cs_value) \
090   ( (pushed_ui32_cs_value) & 0xffff )
091 
092 
093 /**
094  * Structure of an interrupted Kernel thread's context
095  */
096 struct sos_cpu_kstate
097 {
098   struct sos_cpu_state regs;
099 } __attribute__((packed));
100 
101 
102 /**
103  * Structure of an interrupted User thread's context. This is almost
104  * the same as a kernel context, except that 2 additional values are
105  * pushed on the stack before the eflags/cs/eip of the interrupted
106  * context: the stack configuration of the interrupted user context.
107  *
108  * @see Section 6.4.1 of Intel x86 vol 1
109  */
110 struct sos_cpu_ustate
111 {
112   struct sos_cpu_state regs;
113   struct
114   {
115     sos_ui32_t cpl3_esp;
116     sos_ui16_t cpl3_ss;
117   };
118 } __attribute__((packed));
119 
120 
121 /*
122  * Structure of a Task State Segment on the x86 Architecture.
123  *
124  * @see Intel x86 spec vol 3, figure 6-2
125  *
126  * @note Such a data structure should not cross any page boundary (see
127  * end of section 6.2.1 of Intel spec vol 3). This is the reason why
128  * we tell gcc to align it on a 128B boundary (its size is 104B, which
129  * is <= 128).
130  */
131 struct x86_tss {
132 
133   /**
134    * Intel provides a way for a task to switch to another in an
135    * automatic way (call gates). In this case, the back_link field
136    * stores the source TSS of the context switch. This allows to
137    * easily implement coroutines, task backtracking, ... In SOS we
138    * don't use TSS for the context switch purpouse, so we always
139    * ignore this field.
140    * (+0)
141    */
142   sos_ui16_t back_link;
143 
144   sos_ui16_t reserved1;
145 
146   /* CPL0 saved context. (+4) */
147   sos_vaddr_t esp0;
148   sos_ui16_t ss0;
149 
150   sos_ui16_t reserved2;
151 
152   /* CPL1 saved context. (+12) */
153   sos_vaddr_t esp1;
154   sos_ui16_t ss1;
155 
156   sos_ui16_t reserved3;
157 
158   /* CPL2 saved context. (+20) */
159   sos_vaddr_t esp2;
160   sos_ui16_t ss2;
161 
162   sos_ui16_t reserved4;
163 
164   /* Interrupted context's saved registers. (+28) */
165   sos_vaddr_t cr3;
166   sos_vaddr_t eip;
167   sos_ui32_t eflags;
168   sos_ui32_t eax;
169   sos_ui32_t ecx;
170   sos_ui32_t edx;
171   sos_ui32_t ebx;
172   sos_ui32_t esp;
173   sos_ui32_t ebp;
174   sos_ui32_t esi;
175   sos_ui32_t edi;
176 
177   /* +72 */
178   sos_ui16_t es;
179   sos_ui16_t reserved5;
180 
181   /* +76 */
182   sos_ui16_t cs;
183   sos_ui16_t reserved6;
184 
185   /* +80 */
186   sos_ui16_t ss;
187   sos_ui16_t reserved7;
188 
189   /* +84 */
190   sos_ui16_t ds;
191   sos_ui16_t reserved8;
192 
193   /* +88 */
194   sos_ui16_t fs;
195   sos_ui16_t reserved9;
196 
197   /* +92 */
198   sos_ui16_t gs;
199   sos_ui16_t reserved10;
200 
201   /* +96 */
202   sos_ui16_t ldtr;
203   sos_ui16_t reserved11;
204 
205   /* +100 */
206   sos_ui16_t debug_trap_flag :1;
207   sos_ui16_t reserved12      :15;
208   sos_ui16_t iomap_base_addr;
209 
210   /* 104 */
211 } __attribute__((packed, aligned(128)));
212 
213 
214 static struct x86_tss kernel_tss;
215 
216 
217 sos_ret_t sos_cpu_context_subsystem_setup()
218 {
219   /* Reset the kernel TSS */
220   memset(&kernel_tss, 0x0, sizeof(kernel_tss));
221 
222   /**
223    * Now setup the kernel TSS.
224    *
225    * Considering the privilege change method we choose (cpl3 -> cpl0
226    * through a software interrupt), we don't need to initialize a
227    * full-fledged TSS. See section 6.4.1 of Intel x86 vol 1. Actually,
228    * only a correct value for the kernel esp and ss are required (aka
229    * "ss0" and "esp0" fields). Since the esp0 will have to be updated
230    * at privilege change time, we don't have to set it up now.
231    */
232   kernel_tss.ss0 = SOS_BUILD_SEGMENT_REG_VALUE(0, FALSE, SOS_SEG_KDATA);
233 
234   /* Register this TSS into the gdt */
235   sos_gdt_register_kernel_tss((sos_vaddr_t) &kernel_tss);
236 
237   return SOS_OK;
238 }
239 
240 
241 /**
242  * THE main operation of a kernel thread. This routine calls the
243  * kernel thread function start_func and calls exit_func when
244  * start_func returns.
245  */
246 static void core_routine (sos_cpu_kstate_function_arg1_t *start_func,
247                           sos_ui32_t start_arg,
248                           sos_cpu_kstate_function_arg1_t *exit_func,
249                           sos_ui32_t exit_arg)
250      __attribute__((noreturn));
251 
252 static void core_routine (sos_cpu_kstate_function_arg1_t *start_func,
253                           sos_ui32_t start_arg,
254                           sos_cpu_kstate_function_arg1_t *exit_func,
255                           sos_ui32_t exit_arg)
256 {
257   start_func(start_arg);
258   exit_func(exit_arg);
259 
260   SOS_ASSERT_FATAL(! "The exit function of the thread should NOT return !");
261   for(;;);
262 }
263 
264 
265 sos_ret_t sos_cpu_kstate_init(struct sos_cpu_state **ctxt,
266                               sos_cpu_kstate_function_arg1_t *start_func,
267                               sos_ui32_t  start_arg,
268                               sos_vaddr_t stack_bottom,
269                               sos_size_t  stack_size,
270                               sos_cpu_kstate_function_arg1_t *exit_func,
271                               sos_ui32_t  exit_arg)
272 {
273   /* We are initializing a Kernel thread's context */
274   struct sos_cpu_kstate *kctxt;
275 
276   /* This is a critical internal function, so that it is assumed that
277      the caller knows what he does: we legitimally assume that values
278      for ctxt, start_func, stack_* and exit_func are allways VALID ! */
279 
280   /* Setup the stack.
281    *
282    * On x86, the stack goes downward. Each frame is configured this
283    * way (higher addresses first):
284    *
285    *  - (optional unused space. As of gcc 3.3, this space is 24 bytes)
286    *  - arg n
287    *  - arg n-1
288    *  - ...
289    *  - arg 1
290    *  - return instruction address: The address the function returns to
291    *    once finished
292    *  - local variables
293    *
294    * The remaining of the code should be read from the end upward to
295    * understand how the processor will handle it.
296    */
297 
298   sos_vaddr_t tmp_vaddr = stack_bottom + stack_size;
299   sos_ui32_t *stack = (sos_ui32_t*)tmp_vaddr;
300 
301   /* If needed, poison the stack */
302 #ifdef SOS_CPU_STATE_DETECT_UNINIT_KERNEL_VARS
303   memset((void*)stack_bottom, SOS_CPU_STATE_STACK_POISON, stack_size);
304 #elif defined(SOS_CPU_STATE_DETECT_KERNEL_STACK_OVERFLOW)
305   sos_cpu_state_prepare_detect_kernel_stack_overflow(stack_bottom, stack_size);
306 #endif
307 
308   /* Simulate a call to the core_routine() function: prepare its
309      arguments */
310   *(--stack) = exit_arg;
311   *(--stack) = (sos_ui32_t)exit_func;
312   *(--stack) = start_arg;
313   *(--stack) = (sos_ui32_t)start_func;
314   *(--stack) = 0; /* Return address of core_routine => force page fault */
315 
316   /*
317    * Setup the initial context structure, so that the CPU will execute
318    * the function core_routine() once this new context has been
319    * restored on CPU
320    */
321 
322   /* Compute the base address of the structure, which must be located
323      below the previous elements */
324   tmp_vaddr  = ((sos_vaddr_t)stack) - sizeof(struct sos_cpu_kstate);
325   kctxt = (struct sos_cpu_kstate*)tmp_vaddr;
326 
327   /* Initialize the CPU context structure */
328   memset(kctxt, 0x0, sizeof(struct sos_cpu_kstate));
329 
330   /* Tell the CPU context structure that the first instruction to
331      execute will be that of the core_routine() function */
332   kctxt->regs.eip = (sos_ui32_t)core_routine;
333 
334   /* Setup the segment registers */
335   kctxt->regs.cs
336     = SOS_BUILD_SEGMENT_REG_VALUE(0, FALSE, SOS_SEG_KCODE); /* Code */
337   kctxt->regs.ds
338     = SOS_BUILD_SEGMENT_REG_VALUE(0, FALSE, SOS_SEG_KDATA); /* Data */
339   kctxt->regs.es
340     = SOS_BUILD_SEGMENT_REG_VALUE(0, FALSE, SOS_SEG_KDATA); /* Data */
341   kctxt->regs.cpl0_ss
342     = SOS_BUILD_SEGMENT_REG_VALUE(0, FALSE, SOS_SEG_KDATA); /* Stack */
343   /* fs and gs unused for the moment. */
344 
345   /* The newly created context is initially interruptible */
346   kctxt->regs.eflags = (1 << 9); /* set IF bit */
347 
348   /* Finally, update the generic kernel/user thread context */
349   *ctxt = (struct sos_cpu_state*) kctxt;
350 
351   return SOS_OK;
352 }
353 
354 
355 sos_ret_t sos_cpu_ustate_init(struct sos_cpu_state **ctxt,
356                               sos_uaddr_t  user_start_PC,
357                               sos_ui32_t   user_start_arg1,
358                               sos_ui32_t   user_start_arg2,
359                               sos_uaddr_t  user_initial_SP,
360                               sos_vaddr_t  kernel_stack_bottom,
361                               sos_size_t   kernel_stack_size)
362 {
363   /* We are initializing a User thread's context */
364   struct sos_cpu_ustate *uctxt;
365 
366   /* This is a critical internal function, so that it is assumed that
367      the caller knows what he does: we legitimally assume that values
368      for ctxt, etc. are allways VALID ! */
369 
370   /* Compute the address of the CPU state to restore on CPU when
371      switching to this new user thread */
372   sos_vaddr_t uctxt_vaddr = kernel_stack_bottom
373                              + kernel_stack_size
374                              - sizeof(struct sos_cpu_ustate);
375   uctxt = (struct sos_cpu_ustate*)uctxt_vaddr;
376 
377   /* If needed, poison the kernel stack */
378 #ifdef SOS_CPU_STATE_DETECT_UNINIT_KERNEL_VARS
379   memset((void*)kernel_stack_bottom,
380          SOS_CPU_STATE_STACK_POISON,
381          kernel_stack_size);
382 #elif defined(SOS_CPU_STATE_DETECT_KERNEL_STACK_OVERFLOW)
383   sos_cpu_state_prepare_detect_kernel_stack_overflow(kernel_stack_bottom,
384                                                      kernel_stack_size);
385 #endif
386 
387   /*
388    * Setup the initial context structure, so that the CPU will restore
389    * the initial registers' value for the user thread. The
390    * user thread argument is passed in the EAX register.
391    */
392 
393   memset(uctxt, 0x0, sizeof(struct sos_cpu_ustate));
394   
395   /* Tell the CPU context structure that the first instruction to
396      execute will be located at user_start_PC (in user space) */
397   uctxt->regs.eip = (sos_ui32_t)user_start_PC;
398   
399   /* Tell the CPU where will be the user stack */
400   uctxt->cpl3_esp = user_initial_SP;
401 
402   /* The parameter to the start function is not passed by the stack to
403      avoid a possible page fault */
404   uctxt->regs.eax = user_start_arg1;
405   uctxt->regs.ebx = user_start_arg2;
406 
407   /* Setup the segment registers */
408   uctxt->regs.cs
409     = SOS_BUILD_SEGMENT_REG_VALUE(3, FALSE, SOS_SEG_UCODE); /* Code */
410   uctxt->regs.ds
411     = SOS_BUILD_SEGMENT_REG_VALUE(3, FALSE, SOS_SEG_UDATA); /* Data */
412   uctxt->regs.es
413     = SOS_BUILD_SEGMENT_REG_VALUE(3, FALSE, SOS_SEG_UDATA); /* Data */
414   uctxt->cpl3_ss
415     = SOS_BUILD_SEGMENT_REG_VALUE(3, FALSE, SOS_SEG_UDATA); /* User Stack */
416 
417   /* We need also to update the segment for the kernel stack
418      segment. It will be used when this context will be restored on
419      CPU: initially it will be executing in kernel mode and will
420      switch immediatly to user mode */
421   uctxt->regs.cpl0_ss
422     = SOS_BUILD_SEGMENT_REG_VALUE(0, FALSE, SOS_SEG_KDATA); /* Kernel Stack */
423 
424   /* fs and gs unused for the moment. */
425 
426   /* The newly created context is initially interruptible */
427   uctxt->regs.eflags = (1 << 9); /* set IF bit */
428 
429   /* Finally, update the generic kernel/user thread context */
430   *ctxt = (struct sos_cpu_state*) uctxt;
431 
432   return SOS_OK;
433 }
434 
435 
436 sos_ret_t
437 sos_cpu_context_is_in_user_mode(const struct sos_cpu_state *ctxt)
438 {
439   /* An interrupted user thread has its CS register set to that of the
440      User code segment */
441   switch (GET_CPU_CS_REGISTER_VALUE(ctxt->cs))
442     {
443     case SOS_BUILD_SEGMENT_REG_VALUE(3, FALSE, SOS_SEG_UCODE):
444       return TRUE;
445       break;
446 
447     case SOS_BUILD_SEGMENT_REG_VALUE(0, FALSE, SOS_SEG_KCODE):
448       return FALSE;
449       break;
450 
451     default:
452       SOS_FATAL_ERROR("Invalid saved context Code segment register: 0x%x (k=%x, u=%x) !",
453                       (unsigned) GET_CPU_CS_REGISTER_VALUE(ctxt->cs),
454                       SOS_BUILD_SEGMENT_REG_VALUE(0, FALSE, SOS_SEG_KCODE),
455                       SOS_BUILD_SEGMENT_REG_VALUE(3, FALSE, SOS_SEG_UCODE));
456       break;
457     }
458 
459   /* Should never get here */
460   return -SOS_EFATAL;
461 }
462 
463 
464 #if defined(SOS_CPU_STATE_DETECT_KERNEL_STACK_OVERFLOW)
465 void
466 sos_cpu_state_prepare_detect_kernel_stack_overflow(const struct sos_cpu_state *ctxt,
467                                                    sos_vaddr_t stack_bottom,
468                                                    sos_size_t stack_size)
469 {
470   sos_size_t poison_size = SOS_CPU_STATE_DETECT_KERNEL_STACK_OVERFLOW;
471   if (poison_size > stack_size)
472     poison_size = stack_size;
473 
474   memset((void*)stack_bottom, SOS_CPU_STATE_STACK_POISON, poison_size);
475 }
476 
477 
478 void
479 sos_cpu_state_detect_kernel_stack_overflow(const struct sos_cpu_state *ctxt,
480                                            sos_vaddr_t stack_bottom,
481                                            sos_size_t stack_size)
482 {
483   unsigned char *c;
484   int i;
485 
486   /* On SOS, "ctxt" corresponds to the address of the esp register of
487      the saved context in Kernel mode (always, even for the interrupted
488      context of a user thread). Here we make sure that this stack
489      pointer is within the allowed stack area */
490   SOS_ASSERT_FATAL(((sos_vaddr_t)ctxt) >= stack_bottom);
491   SOS_ASSERT_FATAL(((sos_vaddr_t)ctxt) + sizeof(struct sos_cpu_kstate)
492                    <= stack_bottom + stack_size);
493 
494   /* Check that the bottom of the stack has not been altered */
495   for (c = (unsigned char*) stack_bottom, i = 0 ;
496        (i < SOS_CPU_STATE_DETECT_KERNEL_STACK_OVERFLOW) && (i < stack_size) ;
497        c++, i++)
498     {
499       SOS_ASSERT_FATAL(SOS_CPU_STATE_STACK_POISON == *c);
500     }
501 }
502 #endif
503 
504 
505 /* =======================================================================
506  * Public Accessor functions
507  */
508 
509 
510 sos_vaddr_t sos_cpu_context_get_PC(const struct sos_cpu_state *ctxt)
511 {
512   SOS_ASSERT_FATAL(NULL != ctxt);
513 
514   /* This is the PC of the interrupted context (ie kernel or user
515      context). */
516   return ctxt->eip;
517 }
518 
519 
520 sos_vaddr_t sos_cpu_context_get_SP(const struct sos_cpu_state *ctxt)
521 {
522   SOS_ASSERT_FATAL(NULL != ctxt);
523 
524   /* 'ctxt' corresponds to the SP of the interrupted context, in Kernel
525      mode. We have to test whether the original interrupted context
526      was that of a kernel or user thread */
527   if (TRUE == sos_cpu_context_is_in_user_mode(ctxt))
528     {
529       struct sos_cpu_ustate * uctxt = (struct sos_cpu_ustate*)ctxt;
530       return uctxt->cpl3_esp;
531     }
532 
533   /* On SOS, "ctxt" corresponds to the address of the esp register of
534      the saved context in Kernel mode (always, even for the interrupted
535      context of a user thread). */
536   return (sos_vaddr_t)ctxt;
537 }
538 
539 
540 sos_ret_t
541 sos_cpu_context_set_EX_return_address(struct sos_cpu_state *ctxt,
542                                       sos_vaddr_t ret_vaddr)
543 {
544   ctxt->eip = ret_vaddr;
545   return SOS_OK;
546 }
547 
548 
549 void sos_cpu_context_dump(const struct sos_cpu_state *ctxt)
550 {
551   char buf[128];
552 
553   snprintf(buf, sizeof(buf),
554            "CPU: eip=%x esp0=%x eflags=%x cs=%x ds=%x ss0=%x err=%x",
555            (unsigned)ctxt->eip, (unsigned)ctxt, (unsigned)ctxt->eflags,
556            (unsigned)GET_CPU_CS_REGISTER_VALUE(ctxt->cs), (unsigned)ctxt->ds,
557            (unsigned)ctxt->cpl0_ss,
558            (unsigned)ctxt->error_code);
559   if (TRUE == sos_cpu_context_is_in_user_mode(ctxt))
560     {
561       struct sos_cpu_ustate * uctxt = (struct sos_cpu_ustate*)ctxt;
562       snprintf(buf, sizeof(buf),
563                "%s esp3=%x ss3=%x",
564                buf, (unsigned)uctxt->cpl3_esp, (unsigned)uctxt->cpl3_ss);
565     }
566   else
567     snprintf(buf, sizeof(buf), "%s [KERNEL MODE]", buf);
568 
569   sos_bochs_putstring(buf); sos_bochs_putstring("\n");
570   sos_x86_videomem_putstring(23, 0,
571                              SOS_X86_VIDEO_FG_BLACK | SOS_X86_VIDEO_BG_LTGRAY,
572                              buf);
573 }
574 
575 
576 /* =======================================================================
577  * Public Accessor functions TO BE USED ONLY BY Exception handlers
578  */
579 
580 
581 sos_ui32_t sos_cpu_context_get_EX_info(const struct sos_cpu_state *ctxt)
582 {
583   SOS_ASSERT_FATAL(NULL != ctxt);
584   return ctxt->error_code;
585 }
586 
587 
588 sos_vaddr_t
589 sos_cpu_context_get_EX_faulting_vaddr(const struct sos_cpu_state *ctxt)
590 {
591   sos_ui32_t cr2;
592 
593   /*
594    * See Intel Vol 3 (section 5.14): the address of the faulting
595    * virtual address of a page fault is stored in the cr2
596    * register.
597    *
598    * Actually, we do not store the cr2 register in a saved
599    * kernel thread's context. So we retrieve the cr2's value directly
600    * from the processor. The value we retrieve in an exception handler
601    * is actually the correct one because an exception is synchronous
602    * with the code causing the fault, and cannot be interrupted since
603    * the IDT entries in SOS are "interrupt gates" (ie IRQ are
604    * disabled).
605    */
606   asm volatile ("movl %%cr2, %0"
607                 :"=r"(cr2)
608                 : );
609 
610   return cr2;
611 }
612 
613 
614 /* =======================================================================
615  * Public Accessor functions TO BE USED ONLY BY the SYSCALL handler
616  */
617 
618 
619 /*
620  * By convention, the USER SOS programs always pass 4 arguments to the
621  * kernel syscall handler: in eax/../edx. For less arguments, the
622  * unused registers are filled with 0s. For more arguments, the 4th
623  * syscall parameter gives the address of the array containing the
624  * remaining arguments. In any case, eax corresponds to the syscall
625  * IDentifier.
626  */
627 
628 
629 inline
630 sos_ret_t sos_syscall_get3args(const struct sos_cpu_state *user_ctxt,
631                                /* out */unsigned int *arg1,
632                                /* out */unsigned int *arg2,
633                                /* out */unsigned int *arg3)
634 {
635   *arg1 = user_ctxt->ebx;
636   *arg2 = user_ctxt->ecx;
637   *arg3 = user_ctxt->edx; 
638   return SOS_OK;
639 }
640 
641 
642 sos_ret_t sos_syscall_get1arg(const struct sos_cpu_state *user_ctxt,
643                               /* out */unsigned int *arg1)
644 {
645   unsigned int unused;
646   return sos_syscall_get3args(user_ctxt, arg1, & unused, & unused);
647 }
648 
649 
650 sos_ret_t sos_syscall_get2args(const struct sos_cpu_state *user_ctxt,
651                                /* out */unsigned int *arg1,
652                                /* out */unsigned int *arg2)
653 {
654   unsigned int unused;
655   return sos_syscall_get3args(user_ctxt, arg1, arg2, & unused);
656 }
657 
658 
659 /*
660  * sos_syscall_get3args() is defined in cpu_context.c because it needs
661  * to know the structure of a struct spu_state
662  */
663 
664 sos_ret_t sos_syscall_get4args(const struct sos_cpu_state *user_ctxt,
665                                /* out */unsigned int *arg1,
666                                /* out */unsigned int *arg2,
667                                /* out */unsigned int *arg3,
668                                /* out */unsigned int *arg4)
669 {
670   sos_uaddr_t  uaddr_other_args;
671   unsigned int other_args[2];
672   sos_ret_t    retval;
673 
674   /* Retrieve the 3 arguments. The last one is an array containing the
675      remaining arguments */
676   retval = sos_syscall_get3args(user_ctxt, arg1, arg2,
677                                 (unsigned int *)& uaddr_other_args);
678   if (SOS_OK != retval)
679     return retval;
680   
681   /* Copy the array containing the remaining arguments from user
682      space */
683   retval = sos_memcpy_from_user((sos_vaddr_t)other_args,
684                                 (sos_uaddr_t)uaddr_other_args,
685                                 sizeof(other_args));
686   if (sizeof(other_args) != retval)
687     return -SOS_EFAULT;
688 
689   *arg3 = other_args[0];
690   *arg4 = other_args[1];
691   return SOS_OK;
692 }
693 
694 
695 sos_ret_t sos_syscall_get5args(const struct sos_cpu_state *user_ctxt,
696                                /* out */unsigned int *arg1,
697                                /* out */unsigned int *arg2,
698                                /* out */unsigned int *arg3,
699                                /* out */unsigned int *arg4,
700                                /* out */unsigned int *arg5)
701 {
702   sos_uaddr_t  uaddr_other_args;
703   unsigned int other_args[3];
704   sos_ret_t    retval;
705 
706   /* Retrieve the 3 arguments. The last one is an array containing the
707      remaining arguments */
708   retval = sos_syscall_get3args(user_ctxt, arg1, arg2,
709                                 (unsigned int *)& uaddr_other_args);
710   if (SOS_OK != retval)
711     return retval;
712   
713   /* Copy the array containing the remaining arguments from user
714      space */
715   retval = sos_memcpy_from_user((sos_vaddr_t)other_args,
716                                 (sos_uaddr_t)uaddr_other_args,
717                                 sizeof(other_args));
718   if (sizeof(other_args) != retval)
719     return -SOS_EFAULT;
720 
721   *arg3 = other_args[0];
722   *arg4 = other_args[1];
723   *arg5 = other_args[2];
724   return SOS_OK;
725 }
726 
727 
728 sos_ret_t sos_syscall_get6args(const struct sos_cpu_state *user_ctxt,
729                                /* out */unsigned int *arg1,
730                                /* out */unsigned int *arg2,
731                                /* out */unsigned int *arg3,
732                                /* out */unsigned int *arg4,
733                                /* out */unsigned int *arg5,
734                                /* out */unsigned int *arg6)
735 {
736   sos_uaddr_t  uaddr_other_args;
737   unsigned int other_args[4];
738   sos_ret_t    retval;
739 
740   /* Retrieve the 3 arguments. The last one is an array containing the
741      remaining arguments */
742   retval = sos_syscall_get3args(user_ctxt, arg1, arg2,
743                                 (unsigned int *)& uaddr_other_args);
744   if (SOS_OK != retval)
745     return retval;
746   
747   /* Copy the array containing the remaining arguments from user
748      space */
749   retval = sos_memcpy_from_user((sos_vaddr_t)other_args,
750                                 (sos_uaddr_t)uaddr_other_args,
751                                 sizeof(other_args));
752   if (sizeof(other_args) != retval)
753     return -SOS_EFAULT;
754 
755   *arg3 = other_args[0];
756   *arg4 = other_args[1];
757   *arg5 = other_args[2];
758   *arg6 = other_args[3];
759   return SOS_OK;
760 }
761 
762 
763 sos_ret_t sos_syscall_get7args(const struct sos_cpu_state *user_ctxt,
764                                /* out */unsigned int *arg1,
765                                /* out */unsigned int *arg2,
766                                /* out */unsigned int *arg3,
767                                /* out */unsigned int *arg4,
768                                /* out */unsigned int *arg5,
769                                /* out */unsigned int *arg6,
770                                /* out */unsigned int *arg7)
771 {
772   sos_uaddr_t  uaddr_other_args;
773   unsigned int other_args[5];
774   sos_ret_t    retval;
775 
776   /* Retrieve the 3 arguments. The last one is an array containing the
777      remaining arguments */
778   retval = sos_syscall_get3args(user_ctxt, arg1, arg2,
779                                 (unsigned int *)& uaddr_other_args);
780   if (SOS_OK != retval)
781     return retval;
782   
783   /* Copy the array containing the remaining arguments from user
784      space */
785   retval = sos_memcpy_from_user((sos_vaddr_t)other_args,
786                                 (sos_uaddr_t)uaddr_other_args,
787                                 sizeof(other_args));
788   if (sizeof(other_args) != retval)
789     return -SOS_EFAULT;
790 
791   *arg3 = other_args[0];
792   *arg4 = other_args[1];
793   *arg5 = other_args[2];
794   *arg6 = other_args[3];
795   *arg7 = other_args[4];
796   return SOS_OK;
797 }
798 
799 
800 sos_ret_t sos_syscall_get8args(const struct sos_cpu_state *user_ctxt,
801                                /* out */unsigned int *arg1,
802                                /* out */unsigned int *arg2,
803                                /* out */unsigned int *arg3,
804                                /* out */unsigned int *arg4,
805                                /* out */unsigned int *arg5,
806                                /* out */unsigned int *arg6,
807                                /* out */unsigned int *arg7,
808                                /* out */unsigned int *arg8)
809 {
810   sos_uaddr_t  uaddr_other_args;
811   unsigned int other_args[6];
812   sos_ret_t    retval;
813 
814   /* Retrieve the 3 arguments. The last one is an array containing the
815      remaining arguments */
816   retval = sos_syscall_get3args(user_ctxt, arg1, arg2,
817                                 (unsigned int *)& uaddr_other_args);
818   if (SOS_OK != retval)
819     return retval;
820   
821   /* Copy the array containing the remaining arguments from user
822      space */
823   retval = sos_memcpy_from_user((sos_vaddr_t)other_args,
824                                 (sos_uaddr_t)uaddr_other_args,
825                                 sizeof(other_args));
826   if (sizeof(other_args) != retval)
827     return -SOS_EFAULT;
828 
829   *arg3 = other_args[0];
830   *arg4 = other_args[1];
831   *arg5 = other_args[2];
832   *arg6 = other_args[3];
833   *arg7 = other_args[4];
834   *arg8 = other_args[5];
835   return SOS_OK;
836 }
837 
838 
839 /* =======================================================================
840  * Backtrace facility. To be used for DEBUGging purpose ONLY.
841  */
842 
843 
844 sos_ui32_t sos_backtrace(const struct sos_cpu_state *cpu_state,
845                          sos_ui32_t max_depth,
846                          sos_vaddr_t stack_bottom,
847                          sos_size_t stack_size,
848                          sos_backtrace_callback_t * backtracer,
849                          void *custom_arg)
850 {
851   int depth;
852   sos_vaddr_t callee_PC, caller_frame;
853 
854   /* Cannot backtrace an interrupted user thread ! */
855   if ((NULL != cpu_state)
856       &&
857       (TRUE == sos_cpu_context_is_in_user_mode(cpu_state)))
858     {
859       return 0;
860     }
861   
862   /*
863    * Layout of a frame on the x86 (compiler=gcc):
864    *
865    * funcA calls funcB calls funcC
866    *
867    *         ....
868    *         funcB Argument 2
869    *         funcB Argument 1
870    *         funcA Return eip
871    * frameB: funcA ebp (ie previous stack frame)
872    *         ....
873    *         (funcB local variables)
874    *         ....
875    *         funcC Argument 2
876    *         funcC Argument 1
877    *         funcB Return eip
878    * frameC: funcB ebp (ie previous stack frame == A0) <---- a frame address
879    *         ....
880    *         (funcC local variables)
881    *         ....
882    *
883    * The presence of "ebp" on the stack depends on 2 things:
884    *   + the compiler is gcc
885    *   + the source is compiled WITHOUT the -fomit-frame-pointer option
886    * In the absence of "ebp", chances are high that the value pushed
887    * at that address is outside the stack boundaries, meaning that the
888    * function will return -SOS_ENOSUP.
889    */
890 
891   if (cpu_state)
892     {
893       callee_PC    = cpu_state->eip;
894       caller_frame = cpu_state->ebp;
895     }
896   else
897     {
898       /* Skip the sos_backtrace() frame */
899       callee_PC    = (sos_vaddr_t)__builtin_return_address(0);
900       caller_frame = (sos_vaddr_t)__builtin_frame_address(1);
901     }
902 
903   for(depth=0 ; depth < max_depth ; depth ++)
904     {
905       /* Call the callback */
906       backtracer(callee_PC, caller_frame + 8, depth, custom_arg);
907 
908       /* If the frame address is funky, don't go further */
909       if ( (caller_frame < stack_bottom)
910            || (caller_frame + 4 >= stack_bottom + stack_size) )
911         return depth;
912 
913       /* Go to caller frame */
914       callee_PC    = *((sos_vaddr_t*) (caller_frame + 4));
915       caller_frame = *((sos_vaddr_t*) caller_frame);
916     }
917   
918   return depth;
919 }
920 
921 
922 /* *************************************************************
923  * Function to manage the TSS.  This function is not really "public":
924  * it is reserved to the assembler routines defined in
925  * cpu_context_switch.S
926  *
927  * Update the kernel stack address so that the IRQ, syscalls and
928  * exception return in a correct stack location when coming back into
929  * kernel mode.
930  */
931 void
932 sos_cpu_context_update_kernel_tss(struct sos_cpu_state *next_ctxt)
933 {
934   /* next_ctxt corresponds to an interrupted user thread ? */
935   if (sos_cpu_context_is_in_user_mode(next_ctxt))
936     {
937       /*
938        * Yes: "next_ctxt" is an interrupted user thread => we are
939        * going to switch to user mode ! Setup the stack address so
940        * that the user thread "next_ctxt" can come back to the correct
941        * stack location when returning in kernel mode.
942        *
943        * This stack location corresponds to the SP of the next user
944        * thread once its context has been transferred on the CPU, ie
945        * once the CPU has executed all the pop/iret instruction of the
946        * context switch with privilege change.
947        */
948       kernel_tss.esp0 = ((sos_vaddr_t)next_ctxt)
949                         + sizeof(struct sos_cpu_ustate);
950       /* Note: no need to protect this agains IRQ because IRQs are not
951          allowed to update it by themselves, and they are not allowed
952          to block */
953     }
954   else
955     {
956       /* No: No need to update kernel TSS when we stay in kernel
957          mode */
958     }
959 }

source navigation ] diff markup ] identifier search ] general search ]