lkml.org 
[lkml]   [2006]   [May]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH] kmap tracking
On Tue, 23 May 2006 11:42:41 -0700 Randy.Dunlap wrote:

> On Mon, 22 May 2006 09:43:24 -0700 Zach Brown wrote:
>
> > Randy.Dunlap wrote:
> > > From: Randy Dunlap <rdunlap@xenotime.net>
> > >
> > > Track kmap/kunmap call history, storing caller function address,
> > > action, and time (jiffies), if CONFIG_DEBUG_KMAP is enabled.
> > > Based on a patch to 2.4.21 by Zach Brown that was used successfully
> > > at Oracle to track down some kmap/kunmap problems.
> >
> > Thanks for bringing this to 2.6.. sorry for the lag in reviewing.
> >
> > > +enum {
> > > + KMAP_FIRST = 1,
> > > + KMAP_ADDREF,
> > > + KMAP_DECREF,
> > > + KMAP_LAST,
> > > +};
> >
> > I trust you got rid of these in the most recent version :)
>
> Yes.
Gone.

> > > +#else
> > > +#define kmap_record_action(nr, action, refcount, retaddr) do {} while (0)
> > > +#endif
> >
> > Make this an inline, please, so that we don't introduce unused var warnings.
>
> What warnings? I built the config option =y and =n and didn't see
> any warnings. (not that I mind changing it)

I got syntax errors when I changed it to inline, so it's back to
being a macro.

> > > +#define kmap(page) __kmap(page, __builtin_return_address(0))
> > > +#define kunmap(page) __kunmap(page, __builtin_return_address(0))
> >
> > Hmm, I was hoping we wouldn't have to do this. Can we use
> > __builtin_return_address(1) from within the debug paths instead of
> > passing down (0)? Then we wouldn't have to ifdef around the declarations..

I have little confidence in the reliability of __builtin_return_address(1).
I see emails about __b_a_r() not working unless using CONFIG_FRAME_POINTER
and sometimes not even then.

Here's an updated kmap tracking patch. All that I have done is
eliminated the unused enums that you mentioned. And I don't plan
to add support for kmap_atomic variants unless someone finds a need
for that.

---
From: Randy Dunlap <rdunlap@xenotime.net>

Track kmap/kunmap call history, storing caller function address,
action, and time (jiffies), if CONFIG_DEBUG_KMAP is enabled.
Based on a patch to 2.4.21 by Zach Brown.
Built with CONFIG_DEBUG_KMAP =y and =n.

Creates "kmap-history" in debugfs.
Any write to kmap-history clears the history table.

Output format is (all tab-separated):
table_index action_index next_flag caller_function+offset/size refcount jiffies

Example:
99 0 - do_execve+0x133/0x1bf 2 39340
99 1 - do_execve+0x133/0x1bf 2 39340
99 2 + proc_pid_cmdline+0x53/0xeb 2 4294946079
99 3 - proc_pid_cmdline+0x53/0xeb 2 4294946079
99 4 - copy_strings_kernel+0x24/0x2b 2 39340
99 5 - copy_strings_kernel+0x24/0x2b 2 39340
99 6 - do_execve+0x11d/0x1bf 2 39340
99 7 - do_execve+0x11d/0x1bf 2 39340

+ = next; next entry for this table_index goes here.

Each history table row tracks up to 8 actions on it. This is a
circular history table per kmap address row, so the 8 most recent
actions are stored and older actions on each row are lost.

I don't plan to add tracking for kmap* variants (atomic, pfn, etc.)
unless a need arises for that.

Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
---
arch/i386/Kconfig.debug | 8 ++
arch/i386/mm/highmem.c | 19 ----
include/asm-i386/highmem.h | 49 +++++++++++-
lib/Kconfig.debug | 2
mm/highmem.c | 174 +++++++++++++++++++++++++++++++++++++++++++++
5 files changed, 229 insertions(+), 23 deletions(-)

--- linux-2617-rc4.orig/mm/highmem.c
+++ linux-2617-rc4/mm/highmem.c
@@ -27,8 +27,41 @@
#include <linux/hash.h>
#include <linux/highmem.h>
#include <linux/blktrace_api.h>
+#include <linux/jiffies.h>
+#include <linux/kallsyms.h>
#include <asm/tlbflush.h>

+#ifdef CONFIG_DEBUG_KMAP
+
+#define NUM_ACTIONS 8
+#define NUM_ACTIONS_MASK (NUM_ACTIONS - 1)
+
+struct kmap_action {
+ unsigned long caller; /* caller's return address */
+ unsigned long jiffies;
+ int count; /* reference count */
+};
+
+struct kmap_hist {
+ unsigned int next;
+ struct kmap_action acts[NUM_ACTIONS];
+} kh_running[LAST_PKMAP];
+
+static inline void kmap_record_action(unsigned int nr,
+ int refcount, void *retaddr)
+{
+ struct kmap_hist *kh = &kh_running[nr];
+ struct kmap_action *act = &kh->acts[kh->next];
+
+ act->caller = (unsigned long)retaddr;
+ act->count = refcount;
+ act->jiffies = jiffies;
+ kh->next = (kh->next + 1) & NUM_ACTIONS_MASK;
+}
+#else
+#define kmap_record_action(nr, refcount, retaddr) do { } while(0)
+#endif
+
static mempool_t *page_pool, *isa_page_pool;

static void *mempool_alloc_pages_isa(gfp_t gfp_mask, void *data)
@@ -142,7 +175,11 @@ start:
return vaddr;
}

+#ifdef CONFIG_DEBUG_KMAP
+void fastcall *kmap_high(struct page *page, void *retaddr)
+#else
void fastcall *kmap_high(struct page *page)
+#endif
{
unsigned long vaddr;

@@ -157,6 +194,8 @@ void fastcall *kmap_high(struct page *pa
if (!vaddr)
vaddr = map_new_virtual(page);
pkmap_count[PKMAP_NR(vaddr)]++;
+ kmap_record_action(PKMAP_NR(vaddr),
+ pkmap_count[PKMAP_NR(vaddr)], retaddr);
BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 2);
spin_unlock(&kmap_lock);
return (void*) vaddr;
@@ -164,7 +203,11 @@ void fastcall *kmap_high(struct page *pa

EXPORT_SYMBOL(kmap_high);

+#ifdef CONFIG_DEBUG_KMAP
+void fastcall kunmap_high(struct page *page, void *retaddr)
+#else
void fastcall kunmap_high(struct page *page)
+#endif
{
unsigned long vaddr;
unsigned long nr;
@@ -175,6 +218,9 @@ void fastcall kunmap_high(struct page *p
BUG_ON(!vaddr);
nr = PKMAP_NR(vaddr);

+ kmap_record_action(PKMAP_NR(vaddr),
+ pkmap_count[nr], retaddr);
+
/*
* A count must never go down to zero
* without a TLB flush!
@@ -600,3 +646,131 @@ void __init page_address_init(void)
}

#endif /* defined(CONFIG_HIGHMEM) && !defined(WANT_PAGE_VIRTUAL) */
+
+/* --------------------------------------------------- */
+
+#ifdef CONFIG_DEBUG_KMAP
+
+#include <linux/debugfs.h>
+#include <linux/seq_file.h>
+
+static void *kmap_seq_start(struct seq_file *seq, loff_t *pos)
+{
+ struct kmap_hist *kha = seq->private;
+
+ if (*pos < 0 || *pos >= LAST_PKMAP)
+ return NULL;
+ return &kha[*pos];
+}
+
+static void *kmap_seq_next(struct seq_file *seq, void *v, loff_t *pos)
+{
+ (*pos)++;
+ return kmap_seq_start(seq, pos);
+}
+
+static int kmap_seq_show(struct seq_file *seq, void *v)
+{
+ struct kmap_hist *khbase = seq->private;
+ struct kmap_hist *kh = v;
+ struct kmap_action *act;
+ unsigned int ind;
+ const char *name;
+ char *modname;
+ unsigned long offset, size;
+ char namebuf[KSYM_NAME_LEN + 1];
+ char caller[256];
+
+ spin_lock(&kmap_lock);
+ for (ind = 0; ind < NUM_ACTIONS; ind++) {
+ act = &kh->acts[ind];
+ name = kallsyms_lookup(act->caller, &size, &offset,
+ &modname, namebuf);
+ if (!name)
+ sprintf(caller, "0x%lx", act->caller);
+ else if (modname)
+ sprintf(caller, "%s+%#lx/%#lx [%s]", name, offset,
+ size, modname);
+ else
+ sprintf(caller, "%s+%#lx/%#lx", name, offset, size);
+ seq_printf(seq, "%u\t%u\t%c\t%s\t%d\t%lu\n",
+ kh - &khbase[0], ind,
+ ind == kh->next ? '+' : '-',
+ caller, act->count,
+ act->jiffies);
+ }
+ spin_unlock(&kmap_lock);
+
+ return 0;
+}
+
+static void kmap_seq_stop(struct seq_file *seq, void *v)
+{
+}
+
+static struct seq_operations kmap_seq_ops = {
+ .start = kmap_seq_start,
+ .next = kmap_seq_next,
+ .stop = kmap_seq_stop,
+ .show = kmap_seq_show,
+};
+
+static int kmap_shared_open(struct inode *inode, struct file *file,
+ struct kmap_hist *source)
+{
+ int ret;
+
+ ret = seq_open(file, &kmap_seq_ops);
+ if (!ret) {
+ struct seq_file *seq = file->private_data;
+ seq->private = source;
+ }
+ return ret;
+}
+
+/* *any* write here clears the kmap history tables */
+static ssize_t kmap_reset_write(struct file *file, const char __user *buf,
+ size_t count, loff_t *pos)
+{
+ spin_lock(&kmap_lock);
+ memset(kh_running, 0, sizeof(kh_running));
+ spin_unlock(&kmap_lock);
+ return count;
+}
+
+static int kmap_running_fop_open(struct inode *inode, struct file *file)
+{
+ return kmap_shared_open(inode, file, kh_running);
+}
+
+struct file_operations kmap_running_seq_fops = {
+ .open = kmap_running_fop_open,
+ .read = seq_read,
+ .write = kmap_reset_write,
+ .llseek = seq_lseek,
+ .release = seq_release,
+};
+
+static struct dentry *kmap_history_file;
+
+static __init int kmap_history_init(void)
+{
+ kmap_history_file = debugfs_create_file("kmap-history", 0644, NULL,
+ kh_running, &kmap_running_seq_fops);
+ if (!kmap_history_file)
+ goto out1;
+
+ return 0;
+
+out1:
+ return -ENOMEM;
+}
+
+static void kmap_history_exit(void)
+{
+ debugfs_remove(kmap_history_file);
+}
+
+subsys_initcall(kmap_history_init);
+module_exit(kmap_history_exit);
+#endif /* CONFIG_DEBUG_KMAP */
--- linux-2617-rc4.orig/arch/i386/mm/highmem.c
+++ linux-2617-rc4/arch/i386/mm/highmem.c
@@ -1,23 +1,6 @@
#include <linux/highmem.h>
#include <linux/module.h>

-void *kmap(struct page *page)
-{
- might_sleep();
- if (!PageHighMem(page))
- return page_address(page);
- return kmap_high(page);
-}
-
-void kunmap(struct page *page)
-{
- if (in_interrupt())
- BUG();
- if (!PageHighMem(page))
- return;
- kunmap_high(page);
-}
-
/*
* kmap_atomic/kunmap_atomic is significantly faster than kmap/kunmap because
* no global lock is needed and because the kmap code must perform a global TLB
@@ -106,8 +89,6 @@ struct page *kmap_atomic_to_page(void *p
return pte_page(*pte);
}

-EXPORT_SYMBOL(kmap);
-EXPORT_SYMBOL(kunmap);
EXPORT_SYMBOL(kmap_atomic);
EXPORT_SYMBOL(kunmap_atomic);
EXPORT_SYMBOL(kmap_atomic_to_page);
--- linux-2617-rc4.orig/include/asm-i386/highmem.h
+++ linux-2617-rc4/include/asm-i386/highmem.h
@@ -63,11 +63,54 @@ extern pte_t *pkmap_page_table;
#define PKMAP_NR(virt) ((virt-PKMAP_BASE) >> PAGE_SHIFT)
#define PKMAP_ADDR(nr) (PKMAP_BASE + ((nr) << PAGE_SHIFT))

-extern void * FASTCALL(kmap_high(struct page *page));
+#ifndef CONFIG_DEBUG_KMAP
+extern void *FASTCALL(kmap_high(struct page *page));
extern void FASTCALL(kunmap_high(struct page *page));

-void *kmap(struct page *page);
-void kunmap(struct page *page);
+#define kmap(page) __kmap(page)
+#define kunmap(page) __kunmap(page)
+
+static inline void *__kmap(struct page *page)
+{
+ might_sleep();
+ if (!PageHighMem(page))
+ return page_address(page);
+ return kmap_high(page);
+}
+
+static inline void __kunmap(struct page *page)
+{
+ if (in_interrupt())
+ BUG();
+ if (!PageHighMem(page))
+ return;
+ kunmap_high(page);
+}
+#else
+extern void *FASTCALL(kmap_high(struct page *page, void *retaddr));
+extern void FASTCALL(kunmap_high(struct page *page, void *retaddr));
+
+#define kmap(page) __kmap(page, __builtin_return_address(0))
+#define kunmap(page) __kunmap(page, __builtin_return_address(0))
+
+static inline void *__kmap(struct page *page, void *retaddr)
+{
+ might_sleep();
+ if (!PageHighMem(page))
+ return page_address(page);
+ return kmap_high(page, retaddr);
+}
+
+static inline void __kunmap(struct page *page, void *retaddr)
+{
+ if (in_interrupt())
+ BUG();
+ if (!PageHighMem(page))
+ return;
+ kunmap_high(page, retaddr);
+}
+#endif
+
void *kmap_atomic(struct page *page, enum km_type type);
void kunmap_atomic(void *kvaddr, enum km_type type);
void *kmap_atomic_pfn(unsigned long pfn, enum km_type type);
--- linux-2617-rc4.orig/arch/i386/Kconfig.debug
+++ linux-2617-rc4/arch/i386/Kconfig.debug
@@ -51,6 +51,14 @@ config DEBUG_PAGEALLOC
This results in a large slowdown, but helps to find certain types
of memory corruptions.

+config DEBUG_KMAP
+ bool "Keep history of kmap/kunmap actions"
+ depends on DEBUG_KERNEL && HIGHMEM
+ help
+ Keep a table of kmap/kunmap actions for developer debugging.
+ This table is accessible via debugfs, file "kmap-history".
+ The data in it can be cleared (reset) by any write to it.
+
config DEBUG_RODATA
bool "Write protect kernel read-only data structures"
depends on DEBUG_KERNEL
--- linux-2617-rc4.orig/lib/Kconfig.debug
+++ linux-2617-rc4/lib/Kconfig.debug
@@ -134,7 +134,7 @@ config DEBUG_HIGHMEM
bool "Highmem debugging"
depends on DEBUG_KERNEL && HIGHMEM
help
- This options enables addition error checking for high memory systems.
+ Enables additional error checking for high memory systems.
Disable for production systems.

config DEBUG_BUGVERBOSE
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
\
 
 \ /
  Last update: 2006-05-24 03:59    [W:0.085 / U:0.080 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site