| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-id: <569F2BED.3060504@samsung.com>
Date: Wed, 20 Jan 2016 07:40:45 +0100
From: Marek Szyprowski <m.szyprowski@...sung.com>
To: Ulf Hansson <ulf.hansson@...aro.org>
Cc: linux-samsung-soc <linux-samsung-soc@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Kevin Hilman <khilman@...nel.org>,
"Rafael J. Wysocki" <rjw@...ysocki.net>,
Anand Moon <linux.amoon@...il.com>,
Bartlomiej Zolnierkiewicz <b.zolnierkie@...sung.com>,
"linux-pm@...r.kernel.org" <linux-pm@...r.kernel.org>,
Daniel Kurtz <djkurtz@...omium.org>,
Nicolas Boichat <drinkcat@...omium.org>
Subject: Re: [PATCH] power: genpd: fix lockdep issue for all subdomains
Hello,
On 2016-01-19 16:25, Ulf Hansson wrote:
> +linux-pm, Daniel Kurtz, Nicolas Boichat
>
> On 4 January 2016 at 11:39, Marek Szyprowski <m.szyprowski@...sung.com> wrote:
>> During genpd_poweron, genpd->lock is acquired recursively for each
>> parent (master) domain, which are separate obejcts. This confuses
>> lockdep, which considers every operation on genpd->lock as being done on
>> the same lock class. This leads to the following false positive warning:
>>
>> =============================================
>> [ INFO: possible recursive locking detected ]
>> 4.4.0-rc4-xu3s #32 Not tainted
>> ---------------------------------------------
>> swapper/0/1 is trying to acquire lock:
>> (&genpd->lock){+.+...}, at: [<c0361550>] __genpd_poweron+0x64/0x108
>>
>> but task is already holding lock:
>> (&genpd->lock){+.+...}, at: [<c0361af8>] genpd_dev_pm_attach+0x168/0x1b8
>>
>> other info that might help us debug this:
>> Possible unsafe locking scenario:
>>
>> CPU0
>> ----
>> lock(&genpd->lock);
>> lock(&genpd->lock);
>>
>> *** DEADLOCK ***
>>
>> May be due to missing lock nesting notation
>>
>> 3 locks held by swapper/0/1:
>> #0: (&dev->mutex){......}, at: [<c0350910>] __driver_attach+0x48/0x98
>> #1: (&dev->mutex){......}, at: [<c0350920>] __driver_attach+0x58/0x98
>> #2: (&genpd->lock){+.+...}, at: [<c0361af8>] genpd_dev_pm_attach+0x168/0x1b8
>>
>> stack backtrace:
>> CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.4.0-rc4-xu3s #32
>> Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
>> [<c0016c98>] (unwind_backtrace) from [<c00139c4>] (show_stack+0x10/0x14)
>> [<c00139c4>] (show_stack) from [<c0270df0>] (dump_stack+0x84/0xc4)
>> [<c0270df0>] (dump_stack) from [<c00780b8>] (__lock_acquire+0x1f88/0x215c)
>> [<c00780b8>] (__lock_acquire) from [<c007886c>] (lock_acquire+0xa4/0xd0)
>> [<c007886c>] (lock_acquire) from [<c0641f2c>] (mutex_lock_nested+0x70/0x4d4)
>> [<c0641f2c>] (mutex_lock_nested) from [<c0361550>] (__genpd_poweron+0x64/0x108)
>> [<c0361550>] (__genpd_poweron) from [<c0361b00>] (genpd_dev_pm_attach+0x170/0x1b8)
>> [<c0361b00>] (genpd_dev_pm_attach) from [<c03520a8>] (platform_drv_probe+0x2c/0xac)
>> [<c03520a8>] (platform_drv_probe) from [<c03507d4>] (driver_probe_device+0x208/0x2fc)
>> [<c03507d4>] (driver_probe_device) from [<c035095c>] (__driver_attach+0x94/0x98)
>> [<c035095c>] (__driver_attach) from [<c034ec14>] (bus_for_each_dev+0x68/0x9c)
>> [<c034ec14>] (bus_for_each_dev) from [<c034fec8>] (bus_add_driver+0x1a0/0x218)
>> [<c034fec8>] (bus_add_driver) from [<c035115c>] (driver_register+0x78/0xf8)
>> [<c035115c>] (driver_register) from [<c0338488>] (exynos_drm_register_drivers+0x28/0x74)
>> [<c0338488>] (exynos_drm_register_drivers) from [<c0338594>] (exynos_drm_init+0x6c/0xc4)
>> [<c0338594>] (exynos_drm_init) from [<c00097f4>] (do_one_initcall+0x90/0x1dc)
>> [<c00097f4>] (do_one_initcall) from [<c0895e08>] (kernel_init_freeable+0x158/0x1f8)
>> [<c0895e08>] (kernel_init_freeable) from [<c063ecac>] (kernel_init+0x8/0xe8)
>> [<c063ecac>] (kernel_init) from [<c000f7d0>] (ret_from_fork+0x14/0x24)
>>
>> This patch replaces mutex_lock with mutex_lock_nested() and uses
>> recursion depth to annotate each genpd->lock operation with separate
>> lockdep subclass.
>>
>> Reported-by: Anand Moon <linux.amoon@...il.com>
>> Signed-off-by: Marek Szyprowski <m.szyprowski@...sung.com>
> Thanks for looking into this!
>
> Daniel Kurtz, did also run into this issue [1] and reported it a while ago.
> There where some discussions and Daniel also posted a patch trying to
> solve the issue [2]. That approach didn't work out, and unfortunate I
> haven't yet been able to look closer into the issue. Sorry about that!
>
>> ---
>> drivers/base/power/domain.c | 21 +++++++++++++--------
>> 1 file changed, 13 insertions(+), 8 deletions(-)
>>
>> diff --git a/drivers/base/power/domain.c b/drivers/base/power/domain.c
>> index b803790..e02ddf6 100644
>> --- a/drivers/base/power/domain.c
>> +++ b/drivers/base/power/domain.c
>> @@ -170,16 +170,15 @@ static void genpd_queue_power_off_work(struct generic_pm_domain *genpd)
>> queue_work(pm_wq, &genpd->power_off_work);
>> }
>>
>> -static int genpd_poweron(struct generic_pm_domain *genpd);
>> -
>> /**
>> * __genpd_poweron - Restore power to a given PM domain and its masters.
>> * @genpd: PM domain to power up.
>> + * @depth: nesting count for lockdep.
>> *
>> * Restore power to @genpd and all of its masters so that it is possible to
>> * resume a device belonging to it.
>> */
>> -static int __genpd_poweron(struct generic_pm_domain *genpd)
>> +static int __genpd_poweron(struct generic_pm_domain *genpd, unsigned int depth)
>> {
>> struct gpd_link *link;
>> int ret = 0;
>> @@ -194,11 +193,16 @@ static int __genpd_poweron(struct generic_pm_domain *genpd)
>> * with it.
>> */
>> list_for_each_entry(link, &genpd->slave_links, slave_node) {
>> - genpd_sd_counter_inc(link->master);
>> + struct generic_pm_domain *master = link->master;
>> +
>> + genpd_sd_counter_inc(master);
>> +
>> + mutex_lock_nested(&master->lock, depth + 1);
> Nested locks isn't a solution to a problem, but instead this tricks
> lockdep about what goes on. Right?
The only difference between mutex_lock_nested and mutex_lock is the way
it is interpreted by deplock. The additional argument is deplock subclass
of the lock. The name of this function is imho a bit misleading.
> I am wondering whether there's another option available which better
> can instruct lockdep to not treat this as an error.
Similar solution is already applied in regulators and i2c cores, see
regulator_lock_supply() and i2c_del_adapter() functions.
>> + ret = __genpd_poweron(master, depth + 1);
>> + mutex_unlock(&master->lock);
>>
>> - ret = genpd_poweron(link->master);
>> if (ret) {
>> - genpd_sd_counter_dec(link->master);
>> + genpd_sd_counter_dec(master);
>> goto err;
>> }
>> }
>> @@ -230,11 +234,12 @@ static int genpd_poweron(struct generic_pm_domain *genpd)
>> int ret;
>>
>> mutex_lock(&genpd->lock);
>> - ret = __genpd_poweron(genpd);
>> + ret = __genpd_poweron(genpd, 0);
>> mutex_unlock(&genpd->lock);
>> return ret;
>> }
>>
>> +
>> static int genpd_save_dev(struct generic_pm_domain *genpd, struct device *dev)
>> {
>> return GENPD_DEV_CALLBACK(genpd, int, save_state, dev);
>> @@ -482,7 +487,7 @@ static int pm_genpd_runtime_resume(struct device *dev)
>> }
>>
>> mutex_lock(&genpd->lock);
>> - ret = __genpd_poweron(genpd);
>> + ret = __genpd_poweron(genpd, 0);
>> mutex_unlock(&genpd->lock);
>>
>> if (ret)
>> --
>> 1.9.2
>>
> Kind regards
> Uffe
>
> [1]
> http://www.spinics.net/lists/arm-kernel/msg471025.html
> [2]
> http://www.spinics.net/lists/stable/msg113650.html
>
>
Best regards
--
Marek Szyprowski, PhD
Samsung R&D Institute Poland
Powered by blists - more mailing lists