| Message ID | CAKvkGKcLU4ehteAqozw6+dCodSoQvJRSkpnu85um0C62c7RzSg@mail.gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi, Ohad,
Thank you. The kernel version is
3.4.-rc6(android-3.4-compat kernel version). I have
checkout 93b465c2e186d96fb90012ba0f9372eb9952e732 and that is good
solution using mutex protect these interface.
steve.
2012/12/20 Ohad Ben-Cohen <ohad@wizery.com>
> Hello Steve,
>
> On Thu, Dec 20, 2012 at 2:44 PM, steve.zhan <zhanzhenbo@gmail.com> wrote:
> > When enable this config to check kernel, kernel will
> > complain panic at slab.c(kmem_cache_alloc function): "BUG: sleeping
> function
> > called from invalid context".
>
> Which kernel version are you using ?
>
> Specifically, do you have commit
> 93b465c2e186d96fb90012ba0f9372eb9952e732 "hwspinlock/core: use a mutex
> to protect the radix tree" ?
>
> Thanks,
> Ohad.
>
diff --git a/drivers/hwspinlock/hwspinlock_core.c b/drivers/hwspinlock/hwspinlock_core.c index db713c0..49ff54a 100644 --- a/drivers/hwspinlock/hwspinlock_core.c +++ b/drivers/hwspinlock/hwspinlock_core.c @@ -50,7 +50,7 @@ * tree, looking for an unused hwspinlock instance, is now reduced to a * single radix tree API call. */ -static RADIX_TREE(hwspinlock_tree, GFP_KERNEL); +static RADIX_TREE(hwspinlock_tree, GFP_ATOMIC); /* * Synchronization of access to the tree is achieved using this mutex, @@ -413,7 +413,7 @@ static int __hwspin_lock_request(struct hwspinlock *hwlock) } /* notify PM core that power is now needed */ - ret = pm_runtime_get_sync(dev); + ret = pm_runtime_get(dev); if (ret < 0) { dev_err(dev, "%s: can't power on device\n", __func__);
Dear Ohad, When enable this config to check kernel, kernel will complain panic at slab.c(kmem_cache_alloc function): "BUG: sleeping function called from invalid context". The Backtrace is: hwspin_lock_register -> radix_tree_insert -> kmem_cache_alloc -> __might_sleep -> BUG Hwspinlock has been inited in postcore_initcall phase, but kernel sleep check has enable in early_initcall phase(__might_sleep_init_called ==1 in sched.c) I think the the same problem is in __hwspin_lock_request function, this function use pm_runtime_get _sync interface, so might seep that flag don't have RPM_ASYNC value. Do you hae some good suggestions Or comments about this? return ret;