unwind.c

xen虚拟机源代码安装包

	prev_ip = info->ip;
	prev_sp = info->sp;
	prev_bsp = info->bsp;

	/* restore the ip */
	if (!info->rp_loc) {
		/* FIXME: should really be level 0 but it occurs too often. KAO */
		UNW_DPRINT(1, "unwind.%s: failed to locate return link (ip=0x%lx)!\n",
			   __FUNCTION__, info->ip);
		STAT(unw.stat.api.unwind_time += ia64_get_itc() - start; local_irq_restore(flags));
		return -1;
	}
	ip = info->ip = *info->rp_loc;
#ifndef XEN
	if (ip < GATE_ADDR) {
#else
	if (!is_hypervisor_virt(info->ip)) {
#endif
		UNW_DPRINT(2, "unwind.%s: reached user-space (ip=0x%lx)\n", __FUNCTION__, ip);
		STAT(unw.stat.api.unwind_time += ia64_get_itc() - start; local_irq_restore(flags));
		return -1;
	}

	/* restore the cfm: */
	if (!info->pfs_loc) {
		UNW_DPRINT(0, "unwind.%s: failed to locate ar.pfs!\n", __FUNCTION__);
		STAT(unw.stat.api.unwind_time += ia64_get_itc() - start; local_irq_restore(flags));
		return -1;
	}
	info->cfm_loc = info->pfs_loc;

	/* restore the bsp: */
	pr = info->pr;
	num_regs = 0;
	if ((info->flags & UNW_FLAG_INTERRUPT_FRAME)) {
		info->pt = info->sp + 16;
		if ((pr & (1UL << PRED_NON_SYSCALL)) != 0)
			num_regs = *info->cfm_loc & 0x7f;		/* size of frame */
		info->pfs_loc =
			(unsigned long *) (info->pt + offsetof(struct pt_regs, ar_pfs));
		UNW_DPRINT(3, "unwind.%s: interrupt_frame pt 0x%lx\n", __FUNCTION__, info->pt);
	} else
		num_regs = (*info->cfm_loc >> 7) & 0x7f;	/* size of locals */
	info->bsp = (unsigned long) ia64_rse_skip_regs((unsigned long *) info->bsp, -num_regs);
	if (info->bsp < info->regstk.limit || info->bsp > info->regstk.top) {
		UNW_DPRINT(0, "unwind.%s: bsp (0x%lx) out of range [0x%lx-0x%lx]\n",
			__FUNCTION__, info->bsp, info->regstk.limit, info->regstk.top);
		STAT(unw.stat.api.unwind_time += ia64_get_itc() - start; local_irq_restore(flags));
		return -1;
	}

	/* restore the sp: */
	info->sp = info->psp;
	if (info->sp < info->memstk.top || info->sp > info->memstk.limit) {
		UNW_DPRINT(0, "unwind.%s: sp (0x%lx) out of range [0x%lx-0x%lx]\n",
			__FUNCTION__, info->sp, info->memstk.top, info->memstk.limit);
		STAT(unw.stat.api.unwind_time += ia64_get_itc() - start; local_irq_restore(flags));
		return -1;
	}

	if (info->ip == prev_ip && info->sp == prev_sp && info->bsp == prev_bsp) {
		UNW_DPRINT(0, "unwind.%s: ip, sp, bsp unchanged; stopping here (ip=0x%lx)\n",
			   __FUNCTION__, ip);
		STAT(unw.stat.api.unwind_time += ia64_get_itc() - start; local_irq_restore(flags));
		return -1;
	}

	/* as we unwind, the saved ar.unat becomes the primary unat: */
	info->pri_unat_loc = info->unat_loc;

	/* finally, restore the predicates: */
	unw_get_pr(info, &info->pr);

	retval = find_save_locs(info);
	STAT(unw.stat.api.unwind_time += ia64_get_itc() - start; local_irq_restore(flags));
	return retval;
}
EXPORT_SYMBOL(unw_unwind);

int
unw_unwind_to_user (struct unw_frame_info *info)
{
	unsigned long ip, sp, pr = 0;

	while (unw_unwind(info) >= 0) {
		unw_get_sp(info, &sp);
		if ((long)((unsigned long)info->task + IA64_STK_OFFSET - sp)
		    < IA64_PT_REGS_SIZE) {
			UNW_DPRINT(0, "unwind.%s: ran off the top of the kernel stack\n",
				   __FUNCTION__);
			break;
		}
#ifndef XEN
		if (unw_is_intr_frame(info) &&
		    (pr & (1UL << PRED_USER_STACK)))
			return 0;
#else
		if (unw_is_intr_frame(info) &&
		    !is_hvm_vcpu(info->task) &&
		    (pr & (1UL << PRED_USER_STACK)))
			return 0;
		/*
		 * vmx fault handlers don't vcpu->on_stack and keep
		 * (pr & (1UL << PRED_USER_STACK)) condition untouched.
		 * we need to stop unwinding somehow.
		 */
		if (unw_is_intr_frame(info) &&
		    is_hvm_vcpu(info->task) &&
		    info->pr_loc == &vcpu_regs(info->task)->pr)
			return 0;
#endif
		if (unw_get_pr (info, &pr) < 0) {
			unw_get_rp(info, &ip);
			UNW_DPRINT(0, "unwind.%s: failed to read "
				   "predicate register (ip=0x%lx)\n",
				__FUNCTION__, ip);
			return -1;
		}
	}
	unw_get_ip(info, &ip);
	UNW_DPRINT(0, "unwind.%s: failed to unwind to user-level (ip=0x%lx)\n",
		   __FUNCTION__, ip);
	return -1;
}
EXPORT_SYMBOL(unw_unwind_to_user);

static void
init_frame_info (struct unw_frame_info *info, struct task_struct *t,
		 struct switch_stack *sw, unsigned long stktop)
{
	unsigned long rbslimit, rbstop, stklimit;
	STAT(unsigned long start, flags;)

	STAT(local_irq_save(flags); ++unw.stat.api.inits; start = ia64_get_itc());

	/*
	 * Subtle stuff here: we _could_ unwind through the switch_stack frame but we
	 * don't want to do that because it would be slow as each preserved register would
	 * have to be processed.  Instead, what we do here is zero out the frame info and
	 * start the unwind process at the function that created the switch_stack frame.
	 * When a preserved value in switch_stack needs to be accessed, run_script() will
	 * initialize the appropriate pointer on demand.
	 */
	memset(info, 0, sizeof(*info));

	rbslimit = (unsigned long) t + IA64_RBS_OFFSET;
	rbstop   = sw->ar_bspstore;
	if (rbstop - (unsigned long) t >= IA64_STK_OFFSET)
		rbstop = rbslimit;

	stklimit = (unsigned long) t + IA64_STK_OFFSET;
	if (stktop <= rbstop)
		stktop = rbstop;

	info->regstk.limit = rbslimit;
	info->regstk.top   = rbstop;
	info->memstk.limit = stklimit;
	info->memstk.top   = stktop;
	info->task = t;
	info->sw  = sw;
	info->sp = info->psp = stktop;
	info->pr = sw->pr;
	UNW_DPRINT(3, "unwind.%s:\n"
		   "  task   0x%lx\n"
		   "  rbs = [0x%lx-0x%lx)\n"
		   "  stk = [0x%lx-0x%lx)\n"
		   "  pr     0x%lx\n"
		   "  sw     0x%lx\n"
		   "  sp     0x%lx\n",
		   __FUNCTION__, (unsigned long) t, rbslimit, rbstop, stktop, stklimit,
		   info->pr, (unsigned long) info->sw, info->sp);
	STAT(unw.stat.api.init_time += ia64_get_itc() - start; local_irq_restore(flags));
}

void
unw_init_from_interruption (struct unw_frame_info *info, struct task_struct *t,
			    struct pt_regs *pt, struct switch_stack *sw)
{
	unsigned long sof;

	init_frame_info(info, t, sw, pt->r12);
	info->cfm_loc = &pt->cr_ifs;
	info->unat_loc = &pt->ar_unat;
	info->pfs_loc = &pt->ar_pfs;
	sof = *info->cfm_loc & 0x7f;
	info->bsp = (unsigned long) ia64_rse_skip_regs((unsigned long *) info->regstk.top, -sof);
	info->ip = pt->cr_iip + ia64_psr(pt)->ri;
	info->pt = (unsigned long) pt;
	UNW_DPRINT(3, "unwind.%s:\n"
		   "  bsp    0x%lx\n"
		   "  sof    0x%lx\n"
		   "  ip     0x%lx\n",
		   __FUNCTION__, info->bsp, sof, info->ip);
	find_save_locs(info);
}

void
unw_init_frame_info (struct unw_frame_info *info, struct task_struct *t, struct switch_stack *sw)
{
	unsigned long sol;

	init_frame_info(info, t, sw, (unsigned long) (sw + 1) - 16);
	info->cfm_loc = &sw->ar_pfs;
	sol = (*info->cfm_loc >> 7) & 0x7f;
	info->bsp = (unsigned long) ia64_rse_skip_regs((unsigned long *) info->regstk.top, -sol);
	info->ip = sw->b0;
	UNW_DPRINT(3, "unwind.%s:\n"
		   "  bsp    0x%lx\n"
		   "  sol    0x%lx\n"
		   "  ip     0x%lx\n",
		   __FUNCTION__, info->bsp, sol, info->ip);
	find_save_locs(info);
}

EXPORT_SYMBOL(unw_init_frame_info);

void
unw_init_from_blocked_task (struct unw_frame_info *info, struct task_struct *t)
{
#ifdef XEN
	struct switch_stack *sw = (struct switch_stack *) (t->arch._thread.ksp + 16);
#else
	struct switch_stack *sw = (struct switch_stack *) (t->thread.ksp + 16);
#endif

	UNW_DPRINT(1, "unwind.%s\n", __FUNCTION__);
	unw_init_frame_info(info, t, sw);
}
EXPORT_SYMBOL(unw_init_from_blocked_task);

static void
init_unwind_table (struct unw_table *table, const char *name, unsigned long segment_base,
		   unsigned long gp, const void *table_start, const void *table_end)
{
	const struct unw_table_entry *start = table_start, *end = table_end;

	table->name = name;
	table->segment_base = segment_base;
	table->gp = gp;
	table->start = segment_base + start[0].start_offset;
	table->end = segment_base + end[-1].end_offset;
	table->array = start;
	table->length = end - start;
}

#ifndef XEN
void *
unw_add_unwind_table (const char *name, unsigned long segment_base, unsigned long gp,
		      const void *table_start, const void *table_end)
{
	const struct unw_table_entry *start = table_start, *end = table_end;
	struct unw_table *table;
	unsigned long flags;

	if (end - start <= 0) {
		UNW_DPRINT(0, "unwind.%s: ignoring attempt to insert empty unwind table\n",
			   __FUNCTION__);
		return NULL;
	}

	table = kmalloc(sizeof(*table), GFP_USER);
	if (!table)
		return NULL;

	init_unwind_table(table, name, segment_base, gp, table_start, table_end);

	spin_lock_irqsave(&unw.lock, flags);
	{
		/* keep kernel unwind table at the front (it's searched most commonly): */
		table->next = unw.tables->next;
		unw.tables->next = table;
	}
	spin_unlock_irqrestore(&unw.lock, flags);

	return table;
}

void
unw_remove_unwind_table (void *handle)
{
	struct unw_table *table, *prev;
	struct unw_script *tmp;
	unsigned long flags;
	long index;

	if (!handle) {
		UNW_DPRINT(0, "unwind.%s: ignoring attempt to remove non-existent unwind table\n",
			   __FUNCTION__);
		return;
	}

	table = handle;
	if (table == &unw.kernel_table) {
		UNW_DPRINT(0, "unwind.%s: sorry, freeing the kernel's unwind table is a "
			   "no-can-do!\n", __FUNCTION__);
		return;
	}

	spin_lock_irqsave(&unw.lock, flags);
	{
		/* first, delete the table: */

		for (prev = (struct unw_table *) &unw.tables; prev; prev = prev->next)
			if (prev->next == table)
				break;
		if (!prev) {
			UNW_DPRINT(0, "unwind.%s: failed to find unwind table %p\n",
				   __FUNCTION__, (void *) table);
			spin_unlock_irqrestore(&unw.lock, flags);
			return;
		}
		prev->next = table->next;
	}
	spin_unlock_irqrestore(&unw.lock, flags);

	/* next, remove hash table entries for this table */

	for (index = 0; index <= UNW_HASH_SIZE; ++index) {
		tmp = unw.cache + unw.hash[index];
		if (unw.hash[index] >= UNW_CACHE_SIZE
		    || tmp->ip < table->start || tmp->ip >= table->end)
			continue;

		write_lock(&tmp->lock);
		{
			if (tmp->ip >= table->start && tmp->ip < table->end) {
				unw.hash[index] = tmp->coll_chain;
				tmp->ip = 0;
			}
		}
		write_unlock(&tmp->lock);
	}

	kfree(table);
}

static int __init
create_gate_table (void)
{
	const struct unw_table_entry *entry, *start, *end;
	unsigned long *lp, segbase = GATE_ADDR;
	size_t info_size, size;
	char *info;
	Elf64_Phdr *punw = NULL, *phdr = (Elf64_Phdr *) (GATE_ADDR + GATE_EHDR->e_phoff);
	int i;

	for (i = 0; i < GATE_EHDR->e_phnum; ++i, ++phdr)
		if (phdr->p_type == PT_IA_64_UNWIND) {
			punw = phdr;
			break;
		}

	if (!punw) {
		printk("%s: failed to find gate DSO's unwind table!\n", __FUNCTION__);
		return 0;
	}

	start = (const struct unw_table_entry *) punw->p_vaddr;
	end = (struct unw_table_entry *) ((char *) start + punw->p_memsz);
	size  = 0;

	unw_add_unwind_table("linux-gate.so", segbase, 0, start, end);

	for (entry = start; entry < end; ++entry)
		size += 3*8 + 8 + 8*UNW_LENGTH(*(u64 *) (segbase + entry->info_offset));
	size += 8;	/* reserve space for "end of table" marker */

	unw.gate_table = kmalloc(size, GFP_KERNEL);
	if (!unw.gate_table) {
		unw.gate_table_size = 0;
		printk(KERN_ERR "%s: unable to create unwind data for gate page!\n", __FUNCTION__);
		return 0;
	}
	unw.gate_table_size = size;

	lp = unw.gate_table;
	info = (char *) unw.gate_table + size;

	for (entry = start; entry < end; ++entry, lp += 3) {
		info_size = 8 + 8*UNW_LENGTH(*(u64 *) (segbase + entry->info_offset));
		info -= info_size;
		memcpy(info, (char *) segbase + entry->info_offset, info_size);

		lp[0] = segbase + entry->start_offset;		/* start */
		lp[1] = segbase + entry->end_offset;		/* end */
		lp[2] = info - (char *) unw.gate_table;		/* info */
	}
	*lp = 0;	/* end-of-table marker */
	return 0;
}

__initcall(create_gate_table);
#endif // !XEN

void __init
unw_init (void)
{
	extern char __gp[];
	extern void unw_hash_index_t_is_too_narrow (void);
	long i, off;

	if (8*sizeof(unw_hash_index_t) < UNW_LOG_HASH_SIZE)
		unw_hash_index_t_is_too_narrow();

	unw.sw_off[unw.preg_index[UNW_REG_PRI_UNAT_GR]] = SW(CALLER_UNAT);
	unw.sw_off[unw.preg_index[UNW_REG_BSPSTORE]] = SW(AR_BSPSTORE);
	unw.sw_off[unw.preg_index[UNW_REG_PFS]] = SW(AR_PFS);
	unw.sw_off[unw.preg_index[UNW_REG_RP]] = SW(B0);
	unw.sw_off[unw.preg_index[UNW_REG_UNAT]] = SW(CALLER_UNAT);
	unw.sw_off[unw.preg_index[UNW_REG_PR]] = SW(PR);
	unw.sw_off[unw.preg_index[UNW_REG_LC]] = SW(AR_LC);
	unw.sw_off[unw.preg_index[UNW_REG_FPSR]] = SW(AR_FPSR);
	for (i = UNW_REG_R4, off = SW(R4); i <= UNW_REG_R7; ++i, off += 8)
		unw.sw_off[unw.preg_index[i]] = off;
	for (i = UNW_REG_B1, off = SW(B1); i <= UNW_REG_B5; ++i, off += 8)
		unw.sw_off[unw.preg_index[i]] = off;
	for (i = UNW_REG_F2, off = SW(F2); i <= UNW_REG_F5; ++i, off += 16)
		unw.sw_off[unw.preg_index[i]] = off;
	for (i = UNW_REG_F16, off = SW(F16); i <= UNW_REG_F31; ++i, off += 16)
		unw.sw_off[unw.preg_index[i]] = off;

	for (i = 0; i < UNW_CACHE_SIZE; ++i) {
		if (i > 0)
			unw.cache[i].lru_chain = (i - 1);
		unw.cache[i].coll_chain = -1;
		rwlock_init(&unw.cache[i].lock);
	}
	unw.lru_head = UNW_CACHE_SIZE - 1;
	unw.lru_tail = 0;

	init_unwind_table(&unw.kernel_table, "kernel", KERNEL_START, (unsigned long) __gp,
			  __start_unwind, __end_unwind);
}

#ifndef XEN
/*
 * DEPRECATED DEPRECATED DEPRECATED DEPRECATED DEPRECATED DEPRECATED DEPRECATED
 *
 *	This system call has been deprecated.  The new and improved way to get
 *	at the kernel's unwind info is via the gate DSO.  The address of the
 *	ELF header for this DSO is passed to user-level via AT_SYSINFO_EHDR.
 *
 * DEPRECATED DEPRECATED DEPRECATED DEPRECATED DEPRECATED DEPRECATED DEPRECATED
 *
 * This system call copies the unwind data into the buffer pointed to by BUF and returns
 * the size of the unwind data.  If BUF_SIZE is smaller than the size of the unwind data
 * or if BUF is NULL, nothing is copied, but the system call still returns the size of the
 * unwind data.
 *
 * The first portion of the unwind data contains an unwind table and rest contains the
 * associated unwind info (in no particular order).  The unwind table consists of a table
 * of entries of the form:
 *
 *	u64 start;	(64-bit address of start of function)
 *	u64 end;	(64-bit address of start of function)
 *	u64 info;	(BUF-relative offset to unwind info)
 *
 * The end of the unwind table is indicated by an entry with a START address of zero.
 *
 * Please see the IA-64 Software Conventions and Runtime Architecture manual for details
 * on the format of the unwind info.
 *
 * ERRORS
 *	EFAULT	BUF points outside your accessible address space.
 */
asmlinkage long
sys_getunwind (void __user *buf, size_t buf_size)
{
	if (buf && buf_size >= unw.gate_table_size)
		if (copy_to_user(buf, unw.gate_table, unw.gate_table_size) != 0)
			return -EFAULT;
	return unw.gate_table_size;
}
#endif