2005-11-23 02:04:17

by liyu

[permalink] [raw]
Subject: [Question] I doublt on spin_lock again.

Hi, All.

I come here again.

I have two questions on spin_lock. these are:

1. I found these use spin_lock(&rq->lock) in set_user_nice(), but
not disable interrput ( e.g. when sys_nice() call it ), if the one
timer interrput come before we unlock the spin_lock, Shall
we dead lock here? Since the scheduler_tick() may try to hold the same
lock.

2. I also found some function name in its definition have some
postfix, I show here two classical examples:

static void double_rq_lock(runqueue_t *rq1, runqueue_t *rq2)
__acquires(rq1->lock)
__acquires(rq2->lock)
{ ... }

static void double_rq_unlock(runqueue_t *rq1, runqueue_t *rq2)
__releases(rq1->lock)
__releases(rq2->lock)
{ ... }

In the related header files, they are defined as two preprocess
macroes, are follow:

# define __acquires(x) __attribute__((context(0,1)))
# define __releases(x) __attribute__((context(1,0)))
# define __acquire(x) __context__(1)
# define __release(x) __context__(-1)


I guess they are some extensions of gcc for C language, but I did
not found any information in GCC manual.

Would you like reply these questions? Thank advanced.

-liyu










2005-11-23 02:51:33

by Zhang Le

[permalink] [raw]
Subject: Re: [Question] I doublt on spin_lock again.

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

liyu wrote:

| 1. I found these use spin_lock(&rq->lock) in set_user_nice(), but
| not disable interrput ( e.g. when sys_nice() call it ), if the
| one timer interrput come before we unlock the spin_lock, Shall we
| dead lock here? Since the scheduler_tick() may try to hold the
| same lock.

set_user_nice() -> task_rq_lock -> local_irq_save()

|
| 2. # define __acquires(x) __attribute__((context(0,1))) # define
| __releases(x) __attribute__((context(1,0))) # define
| __acquire(x) __context__(1) # define __release(x)
| __context__(-1)

info gcc

BTW, have you ever read "HOWTO do Linux kernel development" be Greg?
Do it, if you haven't

- --
Zhang Le, Robert
Linux Engineer/Trainer

ThizLinux Laboratory Limited
Address: Unit 1004, 10/F, Tower B,
Hunghom Commercial Centre, 37 Ma Tau Wai Road,
To Kwa Wan, Kowloon, Hong Kong
Telephone: (852) 2735 2725
Mobile:(852) 9845 4336
Fax: (852) 2111 0702
URL: http://www.thizgroup.com
Public key: gpg --keyserver pgp.mit.edu --recv-keys 1E4E2973

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDg9kOvFHICB5OKXMRAg7BAJwIhyW9Qop4YGF9G56nzqImjy8UgQCfUE/g
b8pK2Fk6oW8ScK42krTZdOQ=
=mDfg
-----END PGP SIGNATURE-----

2005-11-23 02:56:18

by liyu

[permalink] [raw]
Subject: Re: [Question] I doublt on spin_lock again.

Zhang Le wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> liyu wrote:
>
> | 1. I found these use spin_lock(&rq->lock) in set_user_nice(), but
> | not disable interrput ( e.g. when sys_nice() call it ), if the
> | one timer interrput come before we unlock the spin_lock, Shall we
> | dead lock here? Since the scheduler_tick() may try to hold the
> | same lock.
>
> set_user_nice() -> task_rq_lock -> local_irq_save()
>
> |
> | 2. # define __acquires(x) __attribute__((context(0,1))) # define
> | __releases(x) __attribute__((context(1,0))) # define
> | __acquire(x) __context__(1) # define __release(x)
> | __context__(-1)
>
> info gcc
>
> BTW, have you ever read "HOWTO do Linux kernel development" be Greg?
> Do it, if you haven't
>
> - --
> Zhang Le, Robert
> Linux Engineer/Trainer
>
> ThizLinux Laboratory Limited
> Address: Unit 1004, 10/F, Tower B,
> Hunghom Commercial Centre, 37 Ma Tau Wai Road,
> To Kwa Wan, Kowloon, Hong Kong
> Telephone: (852) 2735 2725
> Mobile:(852) 9845 4336
> Fax: (852) 2111 0702
> URL: http://www.thizgroup.com
> Public key: gpg --keyserver pgp.mit.edu --recv-keys 1E4E2973
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.2 (GNU/Linux)
> Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org
>
> iD8DBQFDg9kOvFHICB5OKXMRAg7BAJwIhyW9Qop4YGF9G56nzqImjy8UgQCfUE/g
> b8pK2Fk6oW8ScK42krTZdOQ=
> =mDfg
> -----END PGP SIGNATURE-----
>
>
>
Thanks first.

1. I know local_irq_save() in task_rq_lock(), but it only save FLAGS
register to one unsigned long variable, but not disable maskable interrupt.
2. OK, I am going to get it.

-liyu


2005-11-23 04:09:17

by Zhang Le

[permalink] [raw]
Subject: Re: [Question] I doublt on spin_lock again.

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

liyu wrote:

| 1. I know local_irq_save() in task_rq_lock(), but it only save
| FLAGS register to one unsigned long variable, but not disable
| maskable interrupt.

#define local_irq_save(x) __asm__ __volatile__("pushfl ; popl %0 ;
cli":"=g" (x): /* no
~ input */ :"memory")

- --
Zhang Le, Robert
Linux Engineer/Trainer

ThizLinux Laboratory Limited
Address: Unit 1004, 10/F, Tower B,
Hunghom Commercial Centre, 37 Ma Tau Wai Road,
To Kwa Wan, Kowloon, Hong Kong
Telephone: (852) 2735 2725
Mobile:(852) 9845 4336
Fax: (852) 2111 0702
URL: http://www.thizgroup.com
Public key: gpg --keyserver pgp.mit.edu --recv-keys 1E4E2973

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDg+tnvFHICB5OKXMRAq+UAJ9svPra7vq5Q60T89B5mA+KJMFf6wCfXPGS
Lv/nGwQkHaXUpKYfYx+dxxg=
=xjKN
-----END PGP SIGNATURE-----