*/
#include <linux/list.h>
+#include <linux/rculist.h>
#include <linux/spinlock.h>
#include <linux/hash.h>
#include <linux/init.h>
struct list_head list;
struct kmmio_fault_page *release_next;
unsigned long page; /* location of the fault page */
+ pteval_t old_presence; /* page presence prior to arming */
+ bool armed;
/*
* Number of times this page has been registered as a part
* of a probe. If zero, page is disarmed and this may be freed.
- * Used only by writers (RCU).
+ * Used only by writers (RCU) and post_kmmio_handler().
+ * Protected by kmmio_lock, when linked into kmmio_page_table.
*/
int count;
};
{
struct kmmio_probe *p;
list_for_each_entry_rcu(p, &kmmio_probes, list) {
- if (addr >= p->addr && addr <= (p->addr + p->len))
+ if (addr >= p->addr && addr < (p->addr + p->len))
return p;
}
return NULL;
static struct kmmio_fault_page *get_kmmio_fault_page(unsigned long page)
{
struct list_head *head;
- struct kmmio_fault_page *p;
+ struct kmmio_fault_page *f;
page &= PAGE_MASK;
head = kmmio_page_list(page);
- list_for_each_entry_rcu(p, head, list) {
- if (p->page == page)
- return p;
+ list_for_each_entry_rcu(f, head, list) {
+ if (f->page == page)
+ return f;
}
return NULL;
}
-static void set_page_present(unsigned long addr, bool present, int *pglevel)
+static void clear_pmd_presence(pmd_t *pmd, bool clear, pmdval_t *old)
{
- pteval_t pteval;
- pmdval_t pmdval;
- int level;
- pmd_t *pmd;
- pte_t *pte = lookup_address(addr, &level);
+ pmdval_t v = pmd_val(*pmd);
+ if (clear) {
+ *old = v & _PAGE_PRESENT;
+ v &= ~_PAGE_PRESENT;
+ } else /* presume this has been called with clear==true previously */
+ v |= *old;
+ set_pmd(pmd, __pmd(v));
+}
+
+static void clear_pte_presence(pte_t *pte, bool clear, pteval_t *old)
+{
+ pteval_t v = pte_val(*pte);
+ if (clear) {
+ *old = v & _PAGE_PRESENT;
+ v &= ~_PAGE_PRESENT;
+ } else /* presume this has been called with clear==true previously */
+ v |= *old;
+ set_pte_atomic(pte, __pte(v));
+}
+
+static int clear_page_presence(struct kmmio_fault_page *f, bool clear)
+{
+ unsigned int level;
+ pte_t *pte = lookup_address(f->page, &level);
if (!pte) {
- pr_err("kmmio: no pte for page 0x%08lx\n", addr);
- return;
+ pr_err("kmmio: no pte for page 0x%08lx\n", f->page);
+ return -1;
}
- if (pglevel)
- *pglevel = level;
-
switch (level) {
case PG_LEVEL_2M:
- pmd = (pmd_t *)pte;
- pmdval = pmd_val(*pmd) & ~_PAGE_PRESENT;
- if (present)
- pmdval |= _PAGE_PRESENT;
- set_pmd(pmd, __pmd(pmdval));
+ clear_pmd_presence((pmd_t *)pte, clear, &f->old_presence);
break;
-
case PG_LEVEL_4K:
- pteval = pte_val(*pte) & ~_PAGE_PRESENT;
- if (present)
- pteval |= _PAGE_PRESENT;
- set_pte_atomic(pte, __pte(pteval));
+ clear_pte_presence(pte, clear, &f->old_presence);
break;
-
default:
pr_err("kmmio: unexpected page level 0x%x.\n", level);
- return;
+ return -1;
}
- __flush_tlb_one(addr);
+ __flush_tlb_one(f->page);
+ return 0;
}
-/** Mark the given page as not present. Access to it will trigger a fault. */
-static void arm_kmmio_fault_page(unsigned long page, int *page_level)
+/*
+ * Mark the given page as not present. Access to it will trigger a fault.
+ *
+ * Struct kmmio_fault_page is protected by RCU and kmmio_lock, but the
+ * protection is ignored here. RCU read lock is assumed held, so the struct
+ * will not disappear unexpectedly. Furthermore, the caller must guarantee,
+ * that double arming the same virtual address (page) cannot occur.
+ *
+ * Double disarming on the other hand is allowed, and may occur when a fault
+ * and mmiotrace shutdown happen simultaneously.
+ */
+static int arm_kmmio_fault_page(struct kmmio_fault_page *f)
{
- set_page_present(page & PAGE_MASK, false, page_level);
+ int ret;
+ WARN_ONCE(f->armed, KERN_ERR "kmmio page already armed.\n");
+ if (f->armed) {
+ pr_warning("kmmio double-arm: page 0x%08lx, ref %d, old %d\n",
+ f->page, f->count, !!f->old_presence);
+ }
+ ret = clear_page_presence(f, true);
+ WARN_ONCE(ret < 0, KERN_ERR "kmmio arming 0x%08lx failed.\n", f->page);
+ f->armed = true;
+ return ret;
}
-/** Mark the given page as present. */
-static void disarm_kmmio_fault_page(unsigned long page, int *page_level)
+/** Restore the given page to saved presence state. */
+static void disarm_kmmio_fault_page(struct kmmio_fault_page *f)
{
- set_page_present(page & PAGE_MASK, true, page_level);
+ int ret = clear_page_presence(f, false);
+ WARN_ONCE(ret < 0,
+ KERN_ERR "kmmio disarming 0x%08lx failed.\n", f->page);
+ f->armed = false;
}
/*
ctx = &get_cpu_var(kmmio_ctx);
if (ctx->active) {
- disarm_kmmio_fault_page(faultpage->page, NULL);
if (addr == ctx->addr) {
/*
- * On SMP we sometimes get recursive probe hits on the
- * same address. Context is already saved, fall out.
+ * A second fault on the same page means some other
+ * condition needs handling by do_page_fault(), the
+ * page really not being present is the most common.
*/
- pr_debug("kmmio: duplicate probe hit on CPU %d, for "
- "address 0x%08lx.\n",
- smp_processor_id(), addr);
- ret = 1;
- goto no_kmmio_ctx;
- }
- /*
- * Prevent overwriting already in-flight context.
- * This should not happen, let's hope disarming at least
- * prevents a panic.
- */
- pr_emerg("kmmio: recursive probe hit on CPU %d, "
+ pr_debug("kmmio: secondary hit for 0x%08lx CPU %d.\n",
+ addr, smp_processor_id());
+
+ if (!faultpage->old_presence)
+ pr_info("kmmio: unexpected secondary hit for "
+ "address 0x%08lx on CPU %d.\n", addr,
+ smp_processor_id());
+ } else {
+ /*
+ * Prevent overwriting already in-flight context.
+ * This should not happen, let's hope disarming at
+ * least prevents a panic.
+ */
+ pr_emerg("kmmio: recursive probe hit on CPU %d, "
"for address 0x%08lx. Ignoring.\n",
smp_processor_id(), addr);
- pr_emerg("kmmio: previous hit was at 0x%08lx.\n",
- ctx->addr);
+ pr_emerg("kmmio: previous hit was at 0x%08lx.\n",
+ ctx->addr);
+ disarm_kmmio_fault_page(faultpage);
+ }
goto no_kmmio_ctx;
}
ctx->active++;
regs->flags &= ~X86_EFLAGS_IF;
/* Now we set present bit in PTE and single step. */
- disarm_kmmio_fault_page(ctx->fpage->page, NULL);
+ disarm_kmmio_fault_page(ctx->fpage);
/*
* If another cpu accesses the same page while we are stepping,
struct kmmio_context *ctx = &get_cpu_var(kmmio_ctx);
if (!ctx->active) {
- pr_debug("kmmio: spurious debug trap on CPU %d.\n",
+ /*
+ * debug traps without an active context are due to either
+ * something external causing them (f.e. using a debugger while
+ * mmio tracing enabled), or erroneous behaviour
+ */
+ pr_warning("kmmio: unexpected debug trap on CPU %d.\n",
smp_processor_id());
goto out;
}
if (ctx->probe && ctx->probe->post_handler)
ctx->probe->post_handler(ctx->probe, condition, regs);
- arm_kmmio_fault_page(ctx->fpage->page, NULL);
+ /* Prevent racing against release_kmmio_fault_page(). */
+ spin_lock(&kmmio_lock);
+ if (ctx->fpage->count)
+ arm_kmmio_fault_page(ctx->fpage);
+ spin_unlock(&kmmio_lock);
regs->flags &= ~X86_EFLAGS_TF;
regs->flags |= ctx->saved_flags;
f = get_kmmio_fault_page(page);
if (f) {
if (!f->count)
- arm_kmmio_fault_page(f->page, NULL);
+ arm_kmmio_fault_page(f);
f->count++;
return 0;
}
- f = kmalloc(sizeof(*f), GFP_ATOMIC);
+ f = kzalloc(sizeof(*f), GFP_ATOMIC);
if (!f)
return -1;
f->count = 1;
f->page = page;
- list_add_rcu(&f->list, kmmio_page_list(f->page));
- arm_kmmio_fault_page(f->page, NULL);
+ if (arm_kmmio_fault_page(f)) {
+ kfree(f);
+ return -1;
+ }
+
+ list_add_rcu(&f->list, kmmio_page_list(f->page));
return 0;
}
f->count--;
BUG_ON(f->count < 0);
if (!f->count) {
- disarm_kmmio_fault_page(f->page, NULL);
+ disarm_kmmio_fault_page(f);
f->release_next = *release_list;
*release_list = f;
}
}
+/*
+ * With page-unaligned ioremaps, one or two armed pages may contain
+ * addresses from outside the intended mapping. Events for these addresses
+ * are currently silently dropped. The events may result only from programming
+ * mistakes by accessing addresses before the beginning or past the end of a
+ * mapping.
+ */
int register_kmmio_probe(struct kmmio_probe *p)
{
unsigned long flags;
int ret = 0;
unsigned long size = 0;
+ const unsigned long size_lim = p->len + (p->addr & ~PAGE_MASK);
spin_lock_irqsave(&kmmio_lock, flags);
if (get_kmmio_probe(p->addr)) {
}
kmmio_count++;
list_add_rcu(&p->list, &kmmio_probes);
- while (size < p->len) {
+ while (size < size_lim) {
if (add_kmmio_fault_page(p->addr + size))
pr_err("kmmio: Unable to set page fault.\n");
size += PAGE_SIZE;
head,
struct kmmio_delayed_release,
rcu);
- struct kmmio_fault_page *p = dr->release_list;
- while (p) {
- struct kmmio_fault_page *next = p->release_next;
- BUG_ON(p->count);
- kfree(p);
- p = next;
+ struct kmmio_fault_page *f = dr->release_list;
+ while (f) {
+ struct kmmio_fault_page *next = f->release_next;
+ BUG_ON(f->count);
+ kfree(f);
+ f = next;
}
kfree(dr);
}
static void remove_kmmio_fault_pages(struct rcu_head *head)
{
- struct kmmio_delayed_release *dr = container_of(
- head,
- struct kmmio_delayed_release,
- rcu);
- struct kmmio_fault_page *p = dr->release_list;
+ struct kmmio_delayed_release *dr =
+ container_of(head, struct kmmio_delayed_release, rcu);
+ struct kmmio_fault_page *f = dr->release_list;
struct kmmio_fault_page **prevp = &dr->release_list;
unsigned long flags;
+
spin_lock_irqsave(&kmmio_lock, flags);
- while (p) {
- if (!p->count)
- list_del_rcu(&p->list);
- else
- *prevp = p->release_next;
- prevp = &p->release_next;
- p = p->release_next;
+ while (f) {
+ if (!f->count) {
+ list_del_rcu(&f->list);
+ prevp = &f->release_next;
+ } else {
+ *prevp = f->release_next;
+ }
+ f = f->release_next;
}
spin_unlock_irqrestore(&kmmio_lock, flags);
+
/* This is the real RCU destroy call. */
call_rcu(&dr->rcu, rcu_free_kmmio_fault_pages);
}
{
unsigned long flags;
unsigned long size = 0;
+ const unsigned long size_lim = p->len + (p->addr & ~PAGE_MASK);
struct kmmio_fault_page *release_list = NULL;
struct kmmio_delayed_release *drelease;
spin_lock_irqsave(&kmmio_lock, flags);
- while (size < p->len) {
+ while (size < size_lim) {
release_kmmio_fault_page(p->addr + size, &release_list);
size += PAGE_SIZE;
}
}
EXPORT_SYMBOL(unregister_kmmio_probe);
-static int kmmio_die_notifier(struct notifier_block *nb, unsigned long val,
- void *args)
+static int
+kmmio_die_notifier(struct notifier_block *nb, unsigned long val, void *args)
{
struct die_args *arg = args;
.notifier_call = kmmio_die_notifier
};
-static int __init init_kmmio(void)
+int kmmio_init(void)
{
int i;
+
for (i = 0; i < KMMIO_PAGE_TABLE_SIZE; i++)
INIT_LIST_HEAD(&kmmio_page_table[i]);
+
return register_die_notifier(&nb_die);
}
-fs_initcall(init_kmmio); /* should be before device_initcall() */
+
+void kmmio_cleanup(void)
+{
+ int i;
+
+ unregister_die_notifier(&nb_die);
+ for (i = 0; i < KMMIO_PAGE_TABLE_SIZE; i++) {
+ WARN_ONCE(!list_empty(&kmmio_page_table[i]),
+ KERN_ERR "kmmio_page_table not empty at cleanup, any further tracing will leak memory.\n");
+ }
+}
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff