[PATCH net] dcbnl : Fix lock initialization

[PATCH net] dcbnl : Fix lock initialization

[Anish Bhatt]

dcb_lock was being used uninitialized in dcbnl and is infact missing
 initialization code. Fixed

Signed-off-by: Anish Bhatt <anish@chelsio.com>
---
 net/dcb/dcbnl.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/net/dcb/dcbnl.c b/net/dcb/dcbnl.c
index ca11d28..7bc44e1 100644
--- a/net/dcb/dcbnl.c
+++ b/net/dcb/dcbnl.c
@@ -1914,6 +1914,8 @@ static int __init dcbnl_init(void)
 {
 	INIT_LIST_HEAD(&dcb_app_list);
 
+	spin_lock_init(&dcb_lock);
+
 	rtnl_register(PF_UNSPEC, RTM_GETDCB, dcb_doit, NULL, NULL);
 	rtnl_register(PF_UNSPEC, RTM_SETDCB, dcb_doit, NULL, NULL);
 
-- 
2.1.3

Re: [PATCH net] dcbnl : Fix lock initialization

[John Fastabend]

On 11/06/2014 10:09 AM, Anish Bhatt wrote:
> dcb_lock was being used uninitialized in dcbnl and is infact missing
>   initialization code. Fixed
>

Are you trying to resolve a bug? It is initialized with

static DEFINE_SPINLOCK(dcb_lock);

and if you follow the code far enough you get to this in
spinlock_types.h:


  #ifdef CONFIG_DEBUG_SPINLOCK
  # define SPIN_DEBUG_INIT(lockname)      \
      .magic = SPINLOCK_MAGIC,        \
      .owner_cpu = -1,            \
      .owner = SPINLOCK_OWNER_INIT,
  #else
  # define SPIN_DEBUG_INIT(lockname)
  #endif

  #define __RAW_SPIN_LOCK_INITIALIZER(lockname)   \
      {                   \
      .raw_lock = __ARCH_SPIN_LOCK_UNLOCKED,  \
      SPIN_DEBUG_INIT(lockname)       \
      SPIN_DEP_MAP_INIT(lockname) }

[...]



-- 
John Fastabend         Intel Corporation

RE: [PATCH net] dcbnl : Fix lock initialization

[Anish Bhatt]

Yes, without this kernel is complaining about inconsitent lock state when lock debugging is enabled. Unfortunately I do not have the trace lying around right now.

If you wish, you can reject this patch, I'll resend it when I get the trace again, with trace included.
-Anish
________________________________________
From: John Fastabend [john.fastabend@gmail.com]
Sent: Thursday, November 06, 2014 11:03 AM
To: Anish Bhatt
Cc: netdev@vger.kernel.org; davem@davemloft.net; john.r.fastabend@intel.com; ying.xue@windriver.com; jeffrey.t.kirsher@intel.com; ebiederm@xmission.com
Subject: Re: [PATCH net] dcbnl : Fix lock initialization

On 11/06/2014 10:09 AM, Anish Bhatt wrote:
> dcb_lock was being used uninitialized in dcbnl and is infact missing
>   initialization code. Fixed
>

Are you trying to resolve a bug? It is initialized with

static DEFINE_SPINLOCK(dcb_lock);

and if you follow the code far enough you get to this in
spinlock_types.h:


  #ifdef CONFIG_DEBUG_SPINLOCK
  # define SPIN_DEBUG_INIT(lockname)      \
      .magic = SPINLOCK_MAGIC,        \
      .owner_cpu = -1,            \
      .owner = SPINLOCK_OWNER_INIT,
  #else
  # define SPIN_DEBUG_INIT(lockname)
  #endif

  #define __RAW_SPIN_LOCK_INITIALIZER(lockname)   \
      {                   \
      .raw_lock = __ARCH_SPIN_LOCK_UNLOCKED,  \
      SPIN_DEBUG_INIT(lockname)       \
      SPIN_DEP_MAP_INIT(lockname) }

[...]



--
John Fastabend         Intel Corporation

Re: [PATCH net] dcbnl : Fix lock initialization

[John Fastabend]

On 11/06/2014 11:12 AM, Anish Bhatt wrote:
> Yes, without this kernel is complaining about inconsitent lock state
> when lock debugging is enabled. Unfortunately I do not have the trace
> lying around right now.
>

If you have the trace that might help. I can't recall seeing any splats
in these code paths. Also as far as I can tell you shouldn't need to do
an init after the define. There are lots of examples in ./net/core where
this is done.

So we need to sort out why the init resolves the issue.

> If you wish, you can reject this patch, I'll resend it when I get the trace again, with trace included.
> -Anish
> ________________________________________
> From: John Fastabend [john.fastabend@gmail.com]
> Sent: Thursday, November 06, 2014 11:03 AM
> To: Anish Bhatt
> Cc: netdev@vger.kernel.org; davem@davemloft.net; john.r.fastabend@intel.com; ying.xue@windriver.com; jeffrey.t.kirsher@intel.com; ebiederm@xmission.com
> Subject: Re: [PATCH net] dcbnl : Fix lock initialization
>
> On 11/06/2014 10:09 AM, Anish Bhatt wrote:
>> dcb_lock was being used uninitialized in dcbnl and is infact missing
>>    initialization code. Fixed
>>
>
> Are you trying to resolve a bug? It is initialized with
>
> static DEFINE_SPINLOCK(dcb_lock);
>
> and if you follow the code far enough you get to this in
> spinlock_types.h:
>
>
>    #ifdef CONFIG_DEBUG_SPINLOCK
>    # define SPIN_DEBUG_INIT(lockname)      \
>        .magic = SPINLOCK_MAGIC,        \
>        .owner_cpu = -1,            \
>        .owner = SPINLOCK_OWNER_INIT,
>    #else
>    # define SPIN_DEBUG_INIT(lockname)
>    #endif
>
>    #define __RAW_SPIN_LOCK_INITIALIZER(lockname)   \
>        {                   \
>        .raw_lock = __ARCH_SPIN_LOCK_UNLOCKED,  \
>        SPIN_DEBUG_INIT(lockname)       \
>        SPIN_DEP_MAP_INIT(lockname) }
>
> [...]
>
>
>
> --
> John Fastabend         Intel Corporation
>


-- 
John Fastabend         Intel Corporation

RE: [PATCH net] dcbnl : Fix lock initialization

[Anish Bhatt]

Dave,
   Please do not apply this patch then. 
John, 
I will try to recreate this again and investigate. My setup is currently being used for other purposes so might take a few days.
-Anish
________________________________________
From: John Fastabend [john.fastabend@gmail.com]
Sent: Thursday, November 06, 2014 12:16 PM
To: Anish Bhatt
Cc: netdev@vger.kernel.org; davem@davemloft.net; john.r.fastabend@intel.com; ying.xue@windriver.com; jeffrey.t.kirsher@intel.com; ebiederm@xmission.com
Subject: Re: [PATCH net] dcbnl : Fix lock initialization

On 11/06/2014 11:12 AM, Anish Bhatt wrote:
> Yes, without this kernel is complaining about inconsitent lock state
> when lock debugging is enabled. Unfortunately I do not have the trace
> lying around right now.
>

If you have the trace that might help. I can't recall seeing any splats
in these code paths. Also as far as I can tell you shouldn't need to do
an init after the define. There are lots of examples in ./net/core where
this is done.

So we need to sort out why the init resolves the issue.

> If you wish, you can reject this patch, I'll resend it when I get the trace again, with trace included.
> -Anish
> ________________________________________
> From: John Fastabend [john.fastabend@gmail.com]
> Sent: Thursday, November 06, 2014 11:03 AM
> To: Anish Bhatt
> Cc: netdev@vger.kernel.org; davem@davemloft.net; john.r.fastabend@intel.com; ying.xue@windriver.com; jeffrey.t.kirsher@intel.com; ebiederm@xmission.com
> Subject: Re: [PATCH net] dcbnl : Fix lock initialization
>
> On 11/06/2014 10:09 AM, Anish Bhatt wrote:
>> dcb_lock was being used uninitialized in dcbnl and is infact missing
>>    initialization code. Fixed
>>
>
> Are you trying to resolve a bug? It is initialized with
>
> static DEFINE_SPINLOCK(dcb_lock);
>
> and if you follow the code far enough you get to this in
> spinlock_types.h:
>
>
>    #ifdef CONFIG_DEBUG_SPINLOCK
>    # define SPIN_DEBUG_INIT(lockname)      \
>        .magic = SPINLOCK_MAGIC,        \
>        .owner_cpu = -1,            \
>        .owner = SPINLOCK_OWNER_INIT,
>    #else
>    # define SPIN_DEBUG_INIT(lockname)
>    #endif
>
>    #define __RAW_SPIN_LOCK_INITIALIZER(lockname)   \
>        {                   \
>        .raw_lock = __ARCH_SPIN_LOCK_UNLOCKED,  \
>        SPIN_DEBUG_INIT(lockname)       \
>        SPIN_DEP_MAP_INIT(lockname) }
>
> [...]
>
>
>
> --
> John Fastabend         Intel Corporation
>


--
John Fastabend         Intel Corporation