Message-ID: <0d9145ee-7f25-429c-934e-7dd9bde31bd5@bootlin.com>
Date: Tue, 2 Sep 2025 16:42:04 +0200
From: Bastien Curutchet <bastien.curutchet@...tlin.com>
To: Lee Jones <lee@...nel.org>
Cc: Thomas Petazzoni <thomas.petazzoni@...tlin.com>,
 Miquèl Raynal <miquel.raynal@...tlin.com>,
 Cheng Ming Lin <linchengming884@...il.com>,
 Cheng Ming Lin <chengminglin@...c.com.tw>, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] mfd: core: Increment of_node's refcount before linking it
 to the platform device

Hi,

On 9/2/25 3:29 PM, Lee Jones wrote:
> On Wed, 20 Aug 2025, Bastien Curutchet wrote:
> 
>> When an MFD device is added, a platform_device is allocated. If this
>> device is linked to a DT description, the corresponding OF node is linked
>> to the new platform device but the OF node's refcount isn't incremented.
>> As of_node_put() is called during the platform device release, it leads
>> to a refcount underflow.
>>
>> Call of_node_get() to increment the OF node's refcount when the node is
>> linked to the newly created platform device.
>>
>> Signed-off-by: Bastien Curutchet <bastien.curutchet@...tlin.com>
>> ---
>> Hi all,
>>
>> I'm currently working on a new MFD driver and I encountered some
>> underflow errors with the of_node refcount. As you can see in the logs
>> below, I reproduced the issue on a mainline driver (atmel-hlcdc):
>>
>>> # modprobe atmel-hlcdc
>>> # modprobe -r atmel-hlcdc
>>> # modprobe atmel-hlcdc
>>> # modprobe -r atmel-hlcdc
>>> [   22.932128] OF: ERROR: of_node_release() detected bad of_node_put() on /amba_pl/atmel_sama5@...00000/dc
>>> [   22.941586] CPU: 1 UID: 0 PID: 103 Comm: modprobe Not tainted 6.17.0-rc2-00053-gb19a97d57c15-dirty #81 NONE
>>> [   22.941608] Hardware name: Xilinx Zynq Platform
>>> [   22.941615] Call trace:
>>> [   22.941626]  unwind_backtrace from show_stack+0x10/0x14
>>> [   22.941660]  show_stack from dump_stack_lvl+0x54/0x68
>>> [   22.941680]  dump_stack_lvl from of_node_release+0x140/0x16c
>>> [   22.941707]  of_node_release from kobject_put+0x110/0x130
>>> [   22.941745]  kobject_put from platform_device_release+0x10/0x3c
>>> [   22.941782]  platform_device_release from device_release+0x30/0xa0
>>> [   22.941814]  device_release from kobject_put+0x88/0x130
>>> [   22.941845]  kobject_put from klist_prev+0xd4/0x16c
>>> [   22.941879]  klist_prev from device_for_each_child_reverse+0x88/0xc8
>>> [   22.941911]  device_for_each_child_reverse from devm_mfd_dev_release+0x30/0x54
>>> [   22.941941]  devm_mfd_dev_release from devres_release_all+0xb0/0x114
>>> [   22.941974]  devres_release_all from device_unbind_cleanup+0xc/0x58
>>> [   22.942003]  device_unbind_cleanup from device_release_driver_internal+0x190/0x1c4
>>> [   22.942025]  device_release_driver_internal from driver_detach+0x54/0xa0
>>> [   22.942046]  driver_detach from bus_remove_driver+0x58/0xa4
>>> [   22.942066]  bus_remove_driver from sys_delete_module+0x178/0x25c
>>> [   22.942094]  sys_delete_module from ret_fast_syscall+0x0/0x54
>>> [   22.942116] Exception stack(0xf0a1dfa8 to 0xf0a1dff0)
>>> [   22.942130] dfa0:                   004ec438 005a0870 005a0d40 00000080 00000000 005a0d18
>>> [   22.942144] dfc0: 004ec438 005a0870 005a0190 00000081 005a0d60 005a0870 00000001 0059f6bc
>>> [   22.942155] dfe0: beccdb20 beccdb10 004ed1f4 b6e9eb40
>>> [   22.942163] OF: ERROR: next of_node_put() on this node will result in a kobject warning 'refcount_t: underflow; use-after-free.'
>>> [   23.098617] OF: ERROR: of_node_release() detected bad of_node_put() on /amba_pl/atmel_sama5@...00000/pwm
>>> [   23.108137] CPU: 1 UID: 0 PID: 103 Comm: modprobe Not tainted 6.17.0-rc2-00053-gb19a97d57c15-dirty #81 NONE
>>> [   23.108159] Hardware name: Xilinx Zynq Platform
>>> [   23.108166] Call trace:
>>> [   23.108173]  unwind_backtrace from show_stack+0x10/0x14
>>> [   23.108206]  show_stack from dump_stack_lvl+0x54/0x68
>>> [   23.108227]  dump_stack_lvl from of_node_release+0x140/0x16c
>>> [   23.108252]  of_node_release from kobject_put+0x110/0x130
>>> [   23.108288]  kobject_put from platform_device_release+0x10/0x3c
>>> [   23.108324]  platform_device_release from device_release+0x30/0xa0
>>> [   23.108354]  device_release from kobject_put+0x88/0x130
>>> [   23.108384]  kobject_put from klist_prev+0xd4/0x16c
>>> [   23.108418]  klist_prev from device_for_each_child_reverse+0x88/0xc8
>>> [   23.108450]  device_for_each_child_reverse from devm_mfd_dev_release+0x30/0x54
>>> [   23.108479]  devm_mfd_dev_release from devres_release_all+0xb0/0x114
>>> [   23.108513]  devres_release_all from device_unbind_cleanup+0xc/0x58
>>> [   23.108541]  device_unbind_cleanup from device_release_driver_internal+0x190/0x1c4
>>> [   23.108563]  device_release_driver_internal from driver_detach+0x54/0xa0
>>> [   23.108585]  driver_detach from bus_remove_driver+0x58/0xa4
>>> [   23.108605]  bus_remove_driver from sys_delete_module+0x178/0x25c
>>> [   23.108631]  sys_delete_module from ret_fast_syscall+0x0/0x54
>>> [   23.108653] Exception stack(0xf0a1dfa8 to 0xf0a1dff0)
>>> [   23.108667] dfa0:                   004ec438 005a0870 005a0d40 00000080 00000000 005a0d18
>>> [   23.108681] dfc0: 004ec438 005a0870 005a0190 00000081 005a0d60 005a0870 00000001 0059f6bc
>>> [   23.108691] dfe0: beccdb20 beccdb10 004ed1f4 b6e9eb40
>>> [   23.108698] OF: ERROR: next of_node_put() on this node will result in a kobject warning 'refcount_t: underflow; use-after-free.'
>> ---
>>   drivers/mfd/mfd-core.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/mfd/mfd-core.c b/drivers/mfd/mfd-core.c
>> index 76bd316a50afc5c07ff2a3303c4363b16d0bc023..7d14a1e7631ee8d5e91b228a07b2d05695e41b6e 100644
>> --- a/drivers/mfd/mfd-core.c
>> +++ b/drivers/mfd/mfd-core.c
>> @@ -131,6 +131,7 @@ static int mfd_match_of_node_to_dev(struct platform_device *pdev,
>>   	of_entry->np = np;
>>   	list_add_tail(&of_entry->list, &mfd_of_node_list);
>>   
>> +	of_node_get(np);
> 
> Looks okay at first blush.
> 
> My question would be, why isn't this required for all calls to device_set_node()?
> 

Tough question, I've found drivers that do call of_node_get() before 
device_set_node() and others that don't. I guess it depends on what 
happens during release, but it's not always trivial to follow all the 
redirections that occur at that stage ..


Best regards,
-- 
Bastien Curutchet, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

Powered by blists - more mailing lists