2008-10-29 08:26:49

by Huang, Ying

[permalink] [raw]
Subject: [PATH -mm -v2] Fix a race condtion of oops_in_progress

Fix a race condition accessing oops_in_progress. Which may be changed on
multiple CPU simultaneously, but it is changed via non-atomic operation
++/--. This patch changes the definition of oops_in_process from int to
atomic_t, and accessing method to atomic operations.


ChangeLog

v2:

- Includes fixes from Andrew Moton.

- Re-based on Matthew Wilcox's new atomic_t patch.


Signed-off-by: Huang Ying <[email protected]>

---

arch/blackfin/kernel/traps.c | 16 ++++++++--------
arch/cris/arch-v32/kernel/time.c | 4 ++--
arch/cris/kernel/traps.c | 6 +++---
arch/cris/mm/fault.c | 6 +++---
arch/ia64/kernel/mca.c | 6 +++---
arch/mn10300/mm/fault.c | 4 ++--
arch/parisc/kernel/traps.c | 4 ++--
arch/s390/kernel/setup.c | 6 +++---
arch/s390/mm/fault.c | 4 ++--
drivers/char/vt.c | 2 +-
drivers/mtd/mtdoops.c | 2 +-
drivers/parisc/led.c | 2 +-
drivers/serial/8250.c | 2 +-
drivers/serial/cpm_uart/cpm_uart_core.c | 2 +-
drivers/serial/sunhv.c | 4 ++--
drivers/serial/sunsab.c | 2 +-
drivers/serial/sunsu.c | 2 +-
drivers/serial/sunzilog.c | 2 +-
drivers/serial/uartlite.c | 4 ++--
drivers/video/aty/radeonfb.h | 2 +-
include/linux/console.h | 3 ++-
include/linux/debug_locks.h | 2 +-
include/linux/kernel.h | 3 ++-
kernel/printk.c | 6 +++---
kernel/sched.c | 3 ++-
lib/bust_spinlocks.c | 4 ++--
lib/debug_locks.c | 2 +-
27 files changed, 54 insertions(+), 51 deletions(-)

--- a/arch/blackfin/kernel/traps.c
+++ b/arch/blackfin/kernel/traps.c
@@ -203,7 +203,7 @@ done:
asmlinkage void double_fault_c(struct pt_regs *fp)
{
console_verbose();
- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);
#ifdef CONFIG_DEBUG_VERBOSE
printk(KERN_EMERG "\n" KERN_EMERG "Double Fault\n");
#ifdef CONFIG_DEBUG_DOUBLEFAULT_PRINT
@@ -261,11 +261,11 @@ asmlinkage void trap_c(struct pt_regs *f
#endif
){
console_verbose();
- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);
} else if (current) {
if (current->mm == NULL) {
console_verbose();
- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);
}
}

@@ -534,7 +534,7 @@ asmlinkage void trap_c(struct pt_regs *f
* if we get here we hit a reserved one, so panic
*/
default:
- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);
info.si_code = ILL_ILLPARAOP;
sig = SIGILL;
verbose_printk(KERN_EMERG "Caught Unhandled Exception, code = %08lx\n",
@@ -560,7 +560,7 @@ asmlinkage void trap_c(struct pt_regs *f
#endif
dump_bfin_trace_buffer();

- if (oops_in_progress) {
+ if (atomic_read(&oops_in_progress)) {
/* Dump the current kernel stack */
verbose_printk(KERN_NOTICE "\n" KERN_NOTICE "Kernel Stack\n");
show_stack(current, NULL);
@@ -931,7 +931,7 @@ void dump_bfin_process(struct pt_regs *f
* stack all the time, so do this until we fix that */
unsigned int context = bfin_read_IPEND();

- if (oops_in_progress)
+ if (atomic_read(&oops_in_progress))
verbose_printk(KERN_EMERG "Kernel OOPS in progress\n");

if (context & 0x0020 && (fp->seqstat & SEQSTAT_EXCAUSE) == VEC_HWERR)
@@ -1015,7 +1015,7 @@ void dump_bfin_mem(struct pt_regs *fp)

/* Hardware error interrupts can be deferred */
if (unlikely(sti && (fp->seqstat & SEQSTAT_EXCAUSE) == VEC_HWERR &&
- oops_in_progress)){
+ atomic_read(&oops_in_progress))) {
verbose_printk(KERN_NOTICE "Looks like this was a deferred error - sorry\n");
#ifndef CONFIG_DEBUG_HWERR
verbose_printk(KERN_NOTICE "The remaining message may be meaningless\n"
@@ -1218,7 +1218,7 @@ void panic_cplb_error(int cplb_panic, st
break;
}

- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);

dump_bfin_process(fp);
dump_bfin_mem(fp);
--- a/arch/cris/arch-v32/kernel/time.c
+++ b/arch/cris/arch-v32/kernel/time.c
@@ -170,7 +170,7 @@ handle_watchdog_bite(struct pt_regs* reg
#if defined(CONFIG_ETRAX_WATCHDOG)
extern int cause_of_death;

- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);
printk(KERN_WARNING "Watchdog bite\n");

/* Check if forced restart or unexpected watchdog */
@@ -191,7 +191,7 @@ handle_watchdog_bite(struct pt_regs* reg
stop_watchdog();
printk(KERN_WARNING "Oops: bitten by watchdog\n");
show_registers(regs);
- oops_in_progress = 0;
+ atomic_set(&oops_in_progress, 0);
#ifndef CONFIG_ETRAX_WATCHDOG_NICE_DOGGY
reset_watchdog();
#endif
--- a/arch/cris/kernel/traps.c
+++ b/arch/cris/kernel/traps.c
@@ -164,10 +164,10 @@ void
oops_nmi_handler(struct pt_regs *regs)
{
stop_watchdog();
- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);
printk("NMI!\n");
show_registers(regs);
- oops_in_progress = 0;
+ atomic_set(&oops_in_progress, 0);
}

static int __init
@@ -223,7 +223,7 @@ die_if_kernel(const char *str, struct pt

show_registers(regs);

- oops_in_progress = 0;
+ atomic_set(&oops_in_progress, 0);

#ifdef CONFIG_ETRAX_WATCHDOG_NICE_DOGGY
reset_watchdog();
--- a/arch/cris/mm/fault.c
+++ b/arch/cris/mm/fault.c
@@ -223,8 +223,8 @@ do_page_fault(unsigned long address, str
* terminate things with extreme prejudice.
*/

- if (!oops_in_progress) {
- oops_in_progress = 1;
+ if (!atomic_read(&oops_in_progress)) {
+ atomic_set(&oops_in_progress, 1);
if ((unsigned long) (address) < PAGE_SIZE)
printk(KERN_ALERT "Unable to handle kernel NULL "
"pointer dereference");
@@ -233,7 +233,7 @@ do_page_fault(unsigned long address, str
" at virtual address %08lx\n", address);

die_if_kernel("Oops", regs, (writeaccess << 1) | protection);
- oops_in_progress = 0;
+ atomic_set(&oops_in_progress, 0);
}

do_exit(SIGKILL);
--- a/arch/ia64/kernel/mca.c
+++ b/arch/ia64/kernel/mca.c
@@ -187,7 +187,7 @@ static unsigned long mlogbuf_timestamp =

static int loglevel_save = -1;
#define BREAK_LOGLEVEL(__console_loglevel) \
- oops_in_progress = 1; \
+ atomic_set(&oops_in_progress, 1); \
if (loglevel_save < 0) \
loglevel_save = __console_loglevel; \
__console_loglevel = 15;
@@ -198,7 +198,7 @@ static int loglevel_save = -1;
loglevel_save = -1; \
} \
mlogbuf_finished = 0; \
- oops_in_progress = 0;
+ atomic_set(&oops_in_progress, 0);

/*
* Push messages into buffer, print them later if not urgent.
@@ -215,7 +215,7 @@ void ia64_mca_printk(const char *fmt, ..
va_end(args);

/* Copy the output into mlogbuf */
- if (oops_in_progress) {
+ if (atomic_read(&oops_in_progress)) {
/* mlogbuf was abandoned, use printk directly instead. */
printk(temp_buf);
} else {
--- a/arch/mn10300/mm/fault.c
+++ b/arch/mn10300/mm/fault.c
@@ -39,7 +39,7 @@
void bust_spinlocks(int yes)
{
if (yes) {
- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);
#ifdef CONFIG_SMP
/* Many serial drivers do __global_cli() */
global_irq_lock = 0;
@@ -49,7 +49,7 @@ void bust_spinlocks(int yes)
#ifdef CONFIG_VT
unblank_screen();
#endif
- oops_in_progress = 0;
+ atomic_set(&oops_in_progress, 0);
/*
* OK, the message is on the console. Now we call printk()
* without oops_in_progress set so that printk will give klogd
--- a/arch/parisc/kernel/traps.c
+++ b/arch/parisc/kernel/traps.c
@@ -246,7 +246,7 @@ void die_if_kernel(char *str, struct pt_
return;
}

- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);

/* Amuse the user in a SPARC fashion */
if (err) printk(
@@ -438,7 +438,7 @@ void parisc_terminate(char *msg, struct
{
static DEFINE_SPINLOCK(terminate_lock);

- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);

set_eiem(0);
local_irq_disable();
--- a/arch/s390/kernel/setup.c
+++ b/arch/s390/kernel/setup.c
@@ -241,7 +241,7 @@ static inline void setup_zfcpdump(unsign

void machine_restart(char *command)
{
- if ((!in_interrupt() && !in_atomic()) || oops_in_progress)
+ if ((!in_interrupt() && !in_atomic()) || atomic_read(&oops_in_progress))
/*
* Only unblank the console if we are called in enabled
* context or a bust_spinlocks cleared the way for us.
@@ -252,7 +252,7 @@ void machine_restart(char *command)

void machine_halt(void)
{
- if (!in_interrupt() || oops_in_progress)
+ if (!in_interrupt() || atomic_read(&oops_in_progress))
/*
* Only unblank the console if we are called in enabled
* context or a bust_spinlocks cleared the way for us.
@@ -263,7 +263,7 @@ void machine_halt(void)

void machine_power_off(void)
{
- if (!in_interrupt() || oops_in_progress)
+ if (!in_interrupt() || atomic_read(&oops_in_progress))
/*
* Only unblank the console if we are called in enabled
* context or a bust_spinlocks cleared the way for us.
--- a/arch/s390/mm/fault.c
+++ b/arch/s390/mm/fault.c
@@ -81,11 +81,11 @@ static inline int notify_page_fault(stru
void bust_spinlocks(int yes)
{
if (yes) {
- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);
} else {
int loglevel_save = console_loglevel;
console_unblank();
- oops_in_progress = 0;
+ atomic_set(&oops_in_progress, 0);
/*
* OK, the message is on the console. Now we call printk()
* without oops_in_progress set so that printk will give klogd
--- a/drivers/char/vt.c
+++ b/drivers/char/vt.c
@@ -3665,7 +3665,7 @@ void do_unblank_screen(int leaving_gfx)
* context for the sake of the low level drivers, except in the special
* case of oops_in_progress
*/
- if (!oops_in_progress)
+ if (!atomic_read(&oops_in_progress))
might_sleep();

WARN_CONSOLE_UNLOCKED();
--- a/drivers/mtd/mtdoops.c
+++ b/drivers/mtd/mtdoops.c
@@ -342,7 +342,7 @@ mtdoops_console_write(struct console *co
struct mtd_info *mtd = cxt->mtd;
unsigned long flags;

- if (!oops_in_progress) {
+ if (!atomic_read(&oops_in_progress)) {
mtdoops_console_sync();
return;
}
--- a/drivers/parisc/led.c
+++ b/drivers/parisc/led.c
@@ -464,7 +464,7 @@ static void led_work_func (struct work_s
if (likely(led_diskio)) currentleds |= led_get_diskio_activity();

/* blink all LEDs twice a second if we got an Oops (HPMC) */
- if (unlikely(oops_in_progress))
+ if (unlikely(atomic_read(&oops_in_progress)))
currentleds = (count_HZ<=(HZ/2)) ? 0 : 0xff;

if (currentleds != lastleds)
--- a/drivers/serial/8250.c
+++ b/drivers/serial/8250.c
@@ -2626,7 +2626,7 @@ serial8250_console_write(struct console
if (up->port.sysrq) {
/* serial8250_handle_port() already took the lock */
locked = 0;
- } else if (oops_in_progress) {
+ } else if (atomic_read(&oops_in_progress)) {
locked = spin_trylock(&up->port.lock);
} else
spin_lock(&up->port.lock);
--- a/drivers/serial/cpm_uart/cpm_uart_core.c
+++ b/drivers/serial/cpm_uart/cpm_uart_core.c
@@ -1120,7 +1120,7 @@ static void cpm_uart_console_write(struc
cbd_t __iomem *bdp, *bdbase;
unsigned char *cp;
unsigned long flags;
- int nolock = oops_in_progress;
+ int nolock = atomic_read(&oops_in_progress);

if (unlikely(nolock)) {
local_irq_save(flags);
--- a/drivers/serial/sunhv.c
+++ b/drivers/serial/sunhv.c
@@ -435,7 +435,7 @@ static void sunhv_console_write_paged(st
local_irq_save(flags);
if (port->sysrq) {
locked = 0;
- } else if (oops_in_progress) {
+ } else if (atomic_read(&oops_in_progress)) {
locked = spin_trylock(&port->lock);
} else
spin_lock(&port->lock);
@@ -494,7 +494,7 @@ static void sunhv_console_write_bychar(s
local_irq_save(flags);
if (port->sysrq) {
locked = 0;
- } else if (oops_in_progress) {
+ } else if (atomic_read(&oops_in_progress)) {
locked = spin_trylock(&port->lock);
} else
spin_lock(&port->lock);
--- a/drivers/serial/sunsab.c
+++ b/drivers/serial/sunsab.c
@@ -852,7 +852,7 @@ static void sunsab_console_write(struct
local_irq_save(flags);
if (up->port.sysrq) {
locked = 0;
- } else if (oops_in_progress) {
+ } else if (atomic_read(&oops_in_progress)) {
locked = spin_trylock(&up->port.lock);
} else
spin_lock(&up->port.lock);
--- a/drivers/serial/sunsu.c
+++ b/drivers/serial/sunsu.c
@@ -1296,7 +1296,7 @@ static void sunsu_console_write(struct c
local_irq_save(flags);
if (up->port.sysrq) {
locked = 0;
- } else if (oops_in_progress) {
+ } else if (atomic_read(&oops_in_progress)) {
locked = spin_trylock(&up->port.lock);
} else
spin_lock(&up->port.lock);
--- a/drivers/serial/sunzilog.c
+++ b/drivers/serial/sunzilog.c
@@ -1154,7 +1154,7 @@ sunzilog_console_write(struct console *c
local_irq_save(flags);
if (up->port.sysrq) {
locked = 0;
- } else if (oops_in_progress) {
+ } else if (atomic_read(&oops_in_progress)) {
locked = spin_trylock(&up->port.lock);
} else
spin_lock(&up->port.lock);
--- a/drivers/serial/uartlite.c
+++ b/drivers/serial/uartlite.c
@@ -368,9 +368,9 @@ static void ulite_console_write(struct c
unsigned int ier;
int locked = 1;

- if (oops_in_progress) {
+ if (atomic_read(&oops_in_progress))
locked = spin_trylock_irqsave(&port->lock, flags);
- } else
+ else
spin_lock_irqsave(&port->lock, flags);

/* save and disable interrupt */
--- a/drivers/video/aty/radeonfb.h
+++ b/drivers/video/aty/radeonfb.h
@@ -390,7 +390,7 @@ struct radeonfb_info {
*/
static inline void _radeon_msleep(struct radeonfb_info *rinfo, unsigned long ms)
{
- if (rinfo->no_schedule || oops_in_progress)
+ if (rinfo->no_schedule || atomic_read(&oops_in_progress))
mdelay(ms);
else
msleep(ms);
--- a/include/linux/console.h
+++ b/include/linux/console.h
@@ -142,7 +142,8 @@ void vcs_remove_sysfs(struct tty_struct

/* Some debug stub to catch some of the obvious races in the VT code */
#if 1
-#define WARN_CONSOLE_UNLOCKED() WARN_ON(!is_console_locked() && !oops_in_progress)
+#define WARN_CONSOLE_UNLOCKED() \
+ WARN_ON(!is_console_locked() && !atomic_read(&oops_in_progress))
#else
#define WARN_CONSOLE_UNLOCKED()
#endif
--- a/include/linux/kernel.h
+++ b/include/linux/kernel.h
@@ -270,7 +270,8 @@ static inline void console_verbose(void)

extern void bust_spinlocks(int yes);
extern void wake_up_klogd(void);
-extern int oops_in_progress; /* If set, an oops, panic(), BUG() or die() is in progress */
+/* If set, an oops, panic(), BUG() or die() is in progress */
+extern atomic_t oops_in_progress;
extern int panic_timeout;
extern int panic_on_oops;
extern int panic_on_unrecovered_nmi;
--- a/kernel/printk.c
+++ b/kernel/printk.c
@@ -64,7 +64,7 @@ int console_printk[4] = {
* Low level drivers may need that to know if they can schedule in
* their unblank() callback or not. So let's export it.
*/
-int oops_in_progress;
+atomic_t oops_in_progress;
EXPORT_SYMBOL(oops_in_progress);

/*
@@ -648,7 +648,7 @@ asmlinkage int vprintk(const char *fmt,
* recursion and return - but flag the recursion so that
* it can be printed at the next appropriate moment:
*/
- if (!oops_in_progress) {
+ if (!atomic_read(&oops_in_progress)) {
recursion_bug = 1;
goto out_restore_irqs;
}
@@ -1042,7 +1042,7 @@ void console_unblank(void)
* console_unblank can no longer be called in interrupt context unless
* oops_in_progress is set to 1..
*/
- if (oops_in_progress) {
+ if (atomic_read(&oops_in_progress)) {
if (down_trylock(&console_sem) != 0)
return;
} else
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -8315,7 +8315,8 @@ void __might_sleep(char *file, int line)
static unsigned long prev_jiffy; /* ratelimiting */

if ((!in_atomic() && !irqs_disabled()) ||
- system_state != SYSTEM_RUNNING || oops_in_progress)
+ system_state != SYSTEM_RUNNING ||
+ atomic_read(&oops_in_progress))
return;
if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
return;
--- a/lib/bust_spinlocks.c
+++ b/lib/bust_spinlocks.c
@@ -17,12 +17,12 @@
void __attribute__((weak)) bust_spinlocks(int yes)
{
if (yes) {
- ++oops_in_progress;
+ atomic_inc(&oops_in_progress);
} else {
#ifdef CONFIG_VT
unblank_screen();
#endif
- if (--oops_in_progress == 0)
+ if (atomic_dec_and_test(&oops_in_progress))
wake_up_klogd();
}
}
--- a/lib/debug_locks.c
+++ b/lib/debug_locks.c
@@ -38,7 +38,7 @@ int debug_locks_off(void)
{
if (xchg(&debug_locks, 0)) {
if (!debug_locks_silent) {
- oops_in_progress = 1;
+ atomic_set(&oops_in_progress, 1);
console_verbose();
return 1;
}
--- a/include/linux/debug_locks.h
+++ b/include/linux/debug_locks.h
@@ -17,7 +17,7 @@ extern int debug_locks_off(void);
({ \
int __ret = 0; \
\
- if (!oops_in_progress && unlikely(c)) { \
+ if (!atomic_read(&oops_in_progress) && unlikely(c)) { \
if (debug_locks_off() && !debug_locks_silent) \
WARN_ON(1); \
__ret = 1; \


Attachments:
signature.asc (197.00 B)
This is a digitally signed message part

2008-10-29 08:35:00

by Ingo Molnar

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress


* Huang Ying <[email protected]> wrote:

> Fix a race condition accessing oops_in_progress. Which may be
> changed on multiple CPU simultaneously, but it is changed via
> non-atomic operation ++/--. This patch changes the definition of
> oops_in_process from int to atomic_t, and accessing method to atomic
> operations.
>
>
> ChangeLog
>
> v2:
>
> - Includes fixes from Andrew Moton.
>
> - Re-based on Matthew Wilcox's new atomic_t patch.

hm, there are a couple of places now that do atomic_set(,1) - they
should be atomic_inc(), correct?

Ingo

2008-10-29 08:42:46

by Huang, Ying

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

On Wed, 2008-10-29 at 02:34 -0600, Ingo Molnar wrote:
> * Huang Ying <[email protected]> wrote:
>
> > Fix a race condition accessing oops_in_progress. Which may be
> > changed on multiple CPU simultaneously, but it is changed via
> > non-atomic operation ++/--. This patch changes the definition of
> > oops_in_process from int to atomic_t, and accessing method to atomic
> > operations.
> >
> >
> > ChangeLog
> >
> > v2:
> >
> > - Includes fixes from Andrew Moton.
> >
> > - Re-based on Matthew Wilcox's new atomic_t patch.
>
> hm, there are a couple of places now that do atomic_set(,1) - they
> should be atomic_inc(), correct?

I just translate "oops_in_progress = 1" to
"atomic_set(&oops_in_progress, 1)". I think this is the safest method to
do the translation.

Best Regards,
Huang Ying


Attachments:
signature.asc (197.00 B)
This is a digitally signed message part

2008-10-29 14:53:00

by Chris Snook

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

Huang Ying wrote:
> Fix a race condition accessing oops_in_progress. Which may be changed on
> multiple CPU simultaneously, but it is changed via non-atomic operation
> ++/--. This patch changes the definition of oops_in_process from int to
> atomic_t, and accessing method to atomic operations.

You also need barriers. I believe rmb() before atomic_read() and wmb() after
atomic_set() should suffice.

-- Chris

2008-10-30 02:03:18

by Huang, Ying

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

Hi, Chris,

On Wed, 2008-10-29 at 08:51 -0600, Chris Snook wrote:
> Huang Ying wrote:
> > Fix a race condition accessing oops_in_progress. Which may be changed on
> > multiple CPU simultaneously, but it is changed via non-atomic operation
> > ++/--. This patch changes the definition of oops_in_process from int to
> > atomic_t, and accessing method to atomic operations.
>
> You also need barriers. I believe rmb() before atomic_read() and wmb() after
> atomic_set() should suffice.

I don't think that is necessary. I haven't found there is particular
consistent requirement about oops_in_progress.

Best Regards,
Huang Ying


Attachments:
signature.asc (197.00 B)
This is a digitally signed message part

2008-10-31 16:42:29

by Chris Snook

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

Huang Ying wrote:
> Hi, Chris,
>
> On Wed, 2008-10-29 at 08:51 -0600, Chris Snook wrote:
>> Huang Ying wrote:
>>> Fix a race condition accessing oops_in_progress. Which may be changed on
>>> multiple CPU simultaneously, but it is changed via non-atomic operation
>>> ++/--. This patch changes the definition of oops_in_process from int to
>>> atomic_t, and accessing method to atomic operations.
>> You also need barriers. I believe rmb() before atomic_read() and wmb() after
>> atomic_set() should suffice.
>
> I don't think that is necessary. I haven't found there is particular
> consistent requirement about oops_in_progress.

atomic_read() and atomic_set() don't inherently cause changes to be visible on
other CPUs any faster than ++ and -- operators. Sometimes it happens to work
out that way as a result of how the compiler and the CPU order operations, but
there's no semantic guarantee, and it could even take arbitrarily long under
some circumstances. If you want to use atomic ops to close the race, you need
to use barriers.

-- Chris

2008-11-03 01:52:53

by Huang, Ying

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

On Sat, 2008-11-01 at 00:42 +0800, Chris Snook wrote:
> Huang Ying wrote:
> > Hi, Chris,
> >
> > On Wed, 2008-10-29 at 08:51 -0600, Chris Snook wrote:
> >> Huang Ying wrote:
> >>> Fix a race condition accessing oops_in_progress. Which may be changed on
> >>> multiple CPU simultaneously, but it is changed via non-atomic operation
> >>> ++/--. This patch changes the definition of oops_in_process from int to
> >>> atomic_t, and accessing method to atomic operations.
> >> You also need barriers. I believe rmb() before atomic_read() and wmb() after
> >> atomic_set() should suffice.
> >
> > I don't think that is necessary. I haven't found there is particular
> > consistent requirement about oops_in_progress.
>
> atomic_read() and atomic_set() don't inherently cause changes to be visible on
> other CPUs any faster than ++ and -- operators. Sometimes it happens to work
> out that way as a result of how the compiler and the CPU order operations, but
> there's no semantic guarantee, and it could even take arbitrarily long under
> some circumstances. If you want to use atomic ops to close the race, you need
> to use barriers.

As far as I know, barriers don't cause changes to be visible on other
CPUs faster too. It just guarantees corresponding operations after will
not get executed until that before have finished. And, I don't think we
need make changes to be visible on other CPUs faster.

Best Regards,
Huang Ying


Attachments:
signature.asc (197.00 B)
This is a digitally signed message part

2008-11-03 18:43:56

by Chris Snook

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

Huang Ying wrote:
> On Sat, 2008-11-01 at 00:42 +0800, Chris Snook wrote:
>> Huang Ying wrote:
>>> Hi, Chris,
>>>
>>> On Wed, 2008-10-29 at 08:51 -0600, Chris Snook wrote:
>>>> Huang Ying wrote:
>>>>> Fix a race condition accessing oops_in_progress. Which may be changed on
>>>>> multiple CPU simultaneously, but it is changed via non-atomic operation
>>>>> ++/--. This patch changes the definition of oops_in_process from int to
>>>>> atomic_t, and accessing method to atomic operations.
>>>> You also need barriers. I believe rmb() before atomic_read() and wmb() after
>>>> atomic_set() should suffice.
>>> I don't think that is necessary. I haven't found there is particular
>>> consistent requirement about oops_in_progress.
>> atomic_read() and atomic_set() don't inherently cause changes to be visible on
>> other CPUs any faster than ++ and -- operators. Sometimes it happens to work
>> out that way as a result of how the compiler and the CPU order operations, but
>> there's no semantic guarantee, and it could even take arbitrarily long under
>> some circumstances. If you want to use atomic ops to close the race, you need
>> to use barriers.
>
> As far as I know, barriers don't cause changes to be visible on other
> CPUs faster too. It just guarantees corresponding operations after will
> not get executed until that before have finished. And, I don't think we
> need make changes to be visible on other CPUs faster.

You're correct that barrier() has no impact on other CPUs. wmb() and rmb() do.
If we don't need to make changes visible any faster, what's the point in using
atomic_set()? It's not any less racy. atomic_inc() and atomic_dec() would be
less racy, but you're not using those.

-- Chris

2008-11-04 01:41:45

by Huang, Ying

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

On Tue, 2008-11-04 at 02:44 +0800, Chris Snook wrote:
> Huang Ying wrote:
> > On Sat, 2008-11-01 at 00:42 +0800, Chris Snook wrote:
> >> Huang Ying wrote:
> >>> Hi, Chris,
> >>>
> >>> On Wed, 2008-10-29 at 08:51 -0600, Chris Snook wrote:
> >>>> Huang Ying wrote:
> >>>>> Fix a race condition accessing oops_in_progress. Which may be changed on
> >>>>> multiple CPU simultaneously, but it is changed via non-atomic operation
> >>>>> ++/--. This patch changes the definition of oops_in_process from int to
> >>>>> atomic_t, and accessing method to atomic operations.
> >>>> You also need barriers. I believe rmb() before atomic_read() and wmb() after
> >>>> atomic_set() should suffice.
> >>> I don't think that is necessary. I haven't found there is particular
> >>> consistent requirement about oops_in_progress.
> >> atomic_read() and atomic_set() don't inherently cause changes to be visible on
> >> other CPUs any faster than ++ and -- operators. Sometimes it happens to work
> >> out that way as a result of how the compiler and the CPU order operations, but
> >> there's no semantic guarantee, and it could even take arbitrarily long under
> >> some circumstances. If you want to use atomic ops to close the race, you need
> >> to use barriers.
> >
> > As far as I know, barriers don't cause changes to be visible on other
> > CPUs faster too. It just guarantees corresponding operations after will
> > not get executed until that before have finished. And, I don't think we
> > need make changes to be visible on other CPUs faster.
>
> You're correct that barrier() has no impact on other CPUs. wmb() and rmb() do.
> If we don't need to make changes visible any faster, what's the point in using
> atomic_set()? It's not any less racy. atomic_inc() and atomic_dec() would be
> less racy, but you're not using those.

In default bust_spinlocks() implementation in lib/bust_spinlocks.c,
atomic_inc() and atomic_dec_and_test() is used. Which is used by x86
too. In some other architecture, atomic_set() is used to replace
"oops_in_progress = <xxx>". So this patch fixes architectures which use
default bust_spinlocks(), other architectures can be fixed by
corresponding architecture developers.

Best Regards,
Huang Ying


Attachments:
signature.asc (197.00 B)
This is a digitally signed message part

2008-11-10 07:35:27

by KOSAKI Motohiro

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

> > > As far as I know, barriers don't cause changes to be visible on other
> > > CPUs faster too. It just guarantees corresponding operations after will
> > > not get executed until that before have finished. And, I don't think we
> > > need make changes to be visible on other CPUs faster.
> >
> > You're correct that barrier() has no impact on other CPUs. wmb() and rmb() do.
> > If we don't need to make changes visible any faster, what's the point in using
> > atomic_set()? It's not any less racy. atomic_inc() and atomic_dec() would be
> > less racy, but you're not using those.
>
> In default bust_spinlocks() implementation in lib/bust_spinlocks.c,
> atomic_inc() and atomic_dec_and_test() is used. Which is used by x86
> too. In some other architecture, atomic_set() is used to replace
> "oops_in_progress = <xxx>". So this patch fixes architectures which use
> default bust_spinlocks(), other architectures can be fixed by
> corresponding architecture developers.

I think Chris is right.
So, I reccomend to read Documentation/memory-barriers.txt

Almost architecture gurantee atomic_inc cause barrier implicitly.
but not _all_ architecture.


2008-11-10 18:45:51

by Chris Snook

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

KOSAKI Motohiro wrote:
>>>> As far as I know, barriers don't cause changes to be visible on other
>>>> CPUs faster too. It just guarantees corresponding operations after will
>>>> not get executed until that before have finished. And, I don't think we
>>>> need make changes to be visible on other CPUs faster.
>>> You're correct that barrier() has no impact on other CPUs. wmb() and rmb() do.
>>> If we don't need to make changes visible any faster, what's the point in using
>>> atomic_set()? It's not any less racy. atomic_inc() and atomic_dec() would be
>>> less racy, but you're not using those.
>> In default bust_spinlocks() implementation in lib/bust_spinlocks.c,
>> atomic_inc() and atomic_dec_and_test() is used. Which is used by x86
>> too. In some other architecture, atomic_set() is used to replace
>> "oops_in_progress = <xxx>". So this patch fixes architectures which use
>> default bust_spinlocks(), other architectures can be fixed by
>> corresponding architecture developers.
>
> I think Chris is right.
> So, I reccomend to read Documentation/memory-barriers.txt
>
> Almost architecture gurantee atomic_inc cause barrier implicitly.
> but not _all_ architecture.

The rmb() before atomic_read() is even more critical, since that's a
non-barrier operation on nearly all platforms.

-- Chris

2008-11-11 01:05:29

by Huang, Ying

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

On Mon, 2008-11-10 at 15:35 +0800, KOSAKI Motohiro wrote:
> > > > As far as I know, barriers don't cause changes to be visible on other
> > > > CPUs faster too. It just guarantees corresponding operations after will
> > > > not get executed until that before have finished. And, I don't think we
> > > > need make changes to be visible on other CPUs faster.
> > >
> > > You're correct that barrier() has no impact on other CPUs. wmb() and rmb() do.
> > > If we don't need to make changes visible any faster, what's the point in using
> > > atomic_set()? It's not any less racy. atomic_inc() and atomic_dec() would be
> > > less racy, but you're not using those.
> >
> > In default bust_spinlocks() implementation in lib/bust_spinlocks.c,
> > atomic_inc() and atomic_dec_and_test() is used. Which is used by x86
> > too. In some other architecture, atomic_set() is used to replace
> > "oops_in_progress = <xxx>". So this patch fixes architectures which use
> > default bust_spinlocks(), other architectures can be fixed by
> > corresponding architecture developers.
>
> I think Chris is right.
> So, I reccomend to read Documentation/memory-barriers.txt
>
> Almost architecture gurantee atomic_inc cause barrier implicitly.
> but not _all_ architecture.

Yes. atomic_inc() doesn't imply barrier on all architecture. But we
should not add barriers before all atomic_inc(), just ones needed. Can
you figure out which ones in the patch should has barrier added?

Best Regards,
Huang Ying


Attachments:
signature.asc (197.00 B)
This is a digitally signed message part

2008-11-11 01:12:18

by Chris Snook

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

Huang Ying wrote:
> On Mon, 2008-11-10 at 15:35 +0800, KOSAKI Motohiro wrote:
>>>>> As far as I know, barriers don't cause changes to be visible on other
>>>>> CPUs faster too. It just guarantees corresponding operations after will
>>>>> not get executed until that before have finished. And, I don't think we
>>>>> need make changes to be visible on other CPUs faster.
>>>> You're correct that barrier() has no impact on other CPUs. wmb() and rmb() do.
>>>> If we don't need to make changes visible any faster, what's the point in using
>>>> atomic_set()? It's not any less racy. atomic_inc() and atomic_dec() would be
>>>> less racy, but you're not using those.
>>> In default bust_spinlocks() implementation in lib/bust_spinlocks.c,
>>> atomic_inc() and atomic_dec_and_test() is used. Which is used by x86
>>> too. In some other architecture, atomic_set() is used to replace
>>> "oops_in_progress = <xxx>". So this patch fixes architectures which use
>>> default bust_spinlocks(), other architectures can be fixed by
>>> corresponding architecture developers.
>> I think Chris is right.
>> So, I reccomend to read Documentation/memory-barriers.txt
>>
>> Almost architecture gurantee atomic_inc cause barrier implicitly.
>> but not _all_ architecture.
>
> Yes. atomic_inc() doesn't imply barrier on all architecture. But we
> should not add barriers before all atomic_inc(), just ones needed. Can
> you figure out which ones in the patch should has barrier added?

You need barriers *after* writes, and *before* reads. Adding barriers to the
oops path should be extremely cheap for performance, unless oopsing is a common
occurrence, in which case we have bigger problems.

-- Chris

2008-11-11 01:19:37

by Huang, Ying

[permalink] [raw]
Subject: Re: [PATH -mm -v2] Fix a race condtion of oops_in_progress

On Tue, 2008-11-11 at 09:10 +0800, Chris Snook wrote:
> Huang Ying wrote:
> > On Mon, 2008-11-10 at 15:35 +0800, KOSAKI Motohiro wrote:
> >>>>> As far as I know, barriers don't cause changes to be visible on other
> >>>>> CPUs faster too. It just guarantees corresponding operations after will
> >>>>> not get executed until that before have finished. And, I don't think we
> >>>>> need make changes to be visible on other CPUs faster.
> >>>> You're correct that barrier() has no impact on other CPUs. wmb() and rmb() do.
> >>>> If we don't need to make changes visible any faster, what's the point in using
> >>>> atomic_set()? It's not any less racy. atomic_inc() and atomic_dec() would be
> >>>> less racy, but you're not using those.
> >>> In default bust_spinlocks() implementation in lib/bust_spinlocks.c,
> >>> atomic_inc() and atomic_dec_and_test() is used. Which is used by x86
> >>> too. In some other architecture, atomic_set() is used to replace
> >>> "oops_in_progress = <xxx>". So this patch fixes architectures which use
> >>> default bust_spinlocks(), other architectures can be fixed by
> >>> corresponding architecture developers.
> >> I think Chris is right.
> >> So, I reccomend to read Documentation/memory-barriers.txt
> >>
> >> Almost architecture gurantee atomic_inc cause barrier implicitly.
> >> but not _all_ architecture.
> >
> > Yes. atomic_inc() doesn't imply barrier on all architecture. But we
> > should not add barriers before all atomic_inc(), just ones needed. Can
> > you figure out which ones in the patch should has barrier added?
>
> You need barriers *after* writes, and *before* reads. Adding barriers to the
> oops path should be extremely cheap for performance, unless oopsing is a common
> occurrence, in which case we have bigger problems.

I just suspect why we need these barriers. Do we have some memory must
to be written after oops_in_progress? Or some memory must to be read
before oops_in_progress?

Best Regards,
Huang Ying


Attachments:
signature.asc (197.00 B)
This is a digitally signed message part