i2c: dev: Fix the race between the release of i2c_dev and cdev
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): 20191125193204.GA14257@kunai
Quoted 75 lines · Wolfram Sang
Hi Kevin,
Quoted 72 lines · Kevin Hao
The struct cdev is embedded in the struct
i2c_dev. In the current code,
we would free thei2c_devstruct directly input_i2c_dev(), but the
cdev is manged by a kobject, and the release of it is not predictable.
So it is very possible that thei2c_devis freed before the cdev is
entirely released. We can easily get the following call trace withCONFIG_DEBUG_KOBJECT_RELEASEandCONFIG_DEBUG_OBJECTS_TIMERSenabled.
ODEBUG: free active (active state 0) object type:timer_listhint:delayed_work_timer_fn+0x0/0x38WARNING: CPU: 19 PID: 1 at lib/debugobjects.c:325 debug_print_object+0xb0/0xf0 Modules linked in: CPU: 19 PID: 1 Comm: swapper/0 Tainted: G W 5.2.20-yocto-standard+ #120 Hardware name: Marvell OcteonTX CN96XX board (DT) pstate: 80c00089 (Nzcv daIf +PAN +UAO) pc : debug_print_object+0xb0/0xf0 lr : debug_print_object+0xb0/0xf0 sp : ffff00001292f7d0 x29: ffff00001292f7d0 x28: ffff800b82151788 x27: 0000000000000001 x26: ffff800b892c0000 x25: ffff0000124a2558 x24: 0000000000000000 x23: ffff00001107a1d8 x22: ffff0000116b5088 x21: ffff800bdc6afca8 x20: ffff000012471ae8 x19: ffff00001168f2c8 x18: 0000000000000010 x17: 00000000fd6f304b x16: 00000000ee79de43 x15: ffff800bc0e80568 x14: 79616c6564203a74 x13: 6e6968207473696c x12: 5f72656d6974203a x11: ffff0000113f0018 x10: 0000000000000000 x9 : 000000000000001f x8 : 0000000000000000 x7 : ffff0000101294cc x6 : 0000000000000000 x5 : 0000000000000000 x4 : 0000000000000001 x3 : 00000000ffffffff x2 : 0000000000000000 x1 : 387fc15c8ec0f200 x0 : 0000000000000000 Call trace: debug_print_object+0xb0/0xf0 __debug_check_no_obj_freed+0x19c/0x228 debug_check_no_obj_freed+0x1c/0x28 kfree+0x250/0x440 put_i2c_dev+0x68/0x78 i2cdev_detach_adapter+0x60/0xc8 i2cdev_notifier_call+0x3c/0x70 notifier_call_chain+0x8c/0xe8 blocking_notifier_call_chain+0x64/0x88 device_del+0x74/0x380 device_unregister+0x54/0x78 i2c_del_adapter+0x278/0x2d0 unittest_i2c_bus_remove+0x3c/0x80 platform_drv_remove+0x30/0x50 device_release_driver_internal+0xf4/0x1c0 driver_detach+0x58/0xa0 bus_remove_driver+0x84/0xd8 driver_unregister+0x34/0x60 platform_driver_unregister+0x20/0x30 of_unittest_overlay+0x8d4/0xbe0 of_unittest+0xae8/0xb3c do_one_initcall+0xac/0x450 do_initcall_level+0x208/0x224 kernel_init_freeable+0x2d8/0x36c kernel_init+0x18/0x108 ret_from_fork+0x10/0x1c irq event stamp: 3934661 hardirqs last enabled at (3934661): [<ffff00001009fa04>] debug_exception_exit+0x4c/0x58 hardirqs last disabled at (3934660): [<ffff00001009fb14>] debug_exception_enter+0xa4/0xe0 softirqs last enabled at (3934654): [<ffff000010081d94>] __do_softirq+0x46c/0x628 softirqs last disabled at (3934649): [<ffff0000100b4a1c>] irq_exit+0x104/0x118This is a common issue when using cdev embedded in a struct. Fortunately, we already have a mechanism to solve this kind of issue. Please see commit
233ed09d7fda("chardev: add helper function to register char devs with a struct device") for more detail.In this patch, we choose to embed the struct device into the
i2c_dev, and use the API provided by the commit233ed09d7fdato make sure that the release ofi2c_devand cdev are in sequence.
Quoted 8 lines
Quoted 1 line
Signed-off-by Kevin Hao <[email protected]>Very good patch description! Thank you for the patch, lifecycle issues are subtle, so I am looking forward to get this fixed.
The patch looks good to me, sadly I didn't have the time to test it properly yet. If other people could give their Tested-by or Rev-by, then I'll try to bring it into v5.5. Otherwise it will get priority for 5.6.
Hi Wolfram,
It seems that this doesn't go to v5.6, do you still have any concerns about this?
Thanks,
Kevin
Quoted 5 lines
Kind regards,
Wolfram
Hi Kevin,
I am very sorry for not making it for 5.6. I work on it now to make it go into v5.7.
Quoted 2 lines
entirely released. We can easily get the following call trace with
CONFIG_DEBUG_KOBJECT_RELEASEandCONFIG_DEBUG_OBJECTS_TIMERSenabled.
I didn't get this trace on my ARM32 board (Renesas Lager). Anything more I need to do besides activating those options and using i2c-dev?
Quoted 1 line
+#include <
linux/cdev.h>
This line is not needed. We include this already.
Other than that, the patch looks good. Although I trust you, I like to verify the test results.
Kind regards,
Wolfram
Quoted 4 lines · Wolfram Sang
Hi Kevin,
I am very sorry for not making it for 5.6. I work on it now to make it go into v5.7.
No worries.
Quoted 6 lines
Quoted 3 lines
entirely released. We can easily get the following call trace with
CONFIG_DEBUG_KOBJECT_RELEASEandCONFIG_DEBUG_OBJECTS_TIMERSenabled.I didn't get this trace on my ARM32 board (Renesas Lager). Anything more I need to do besides activating those options and using i2c-dev?
CONFIG_DEBUG_OBJECTS_FREE is also needed. I got this bug warning when running the OF unittest, so you may need to enable the following kernel options if you also want to reproduce this by using that.
CONFIG_OF_UNITTEST
CONFIG_OF_OVERLAY
I have attached the config I used.
Quoted 4 lines
Quoted 2 lines
+#include <
linux/cdev.h>This line is not needed. We include this already.
Yes, there is a conflict with the latest kernel. Do I need to resend a new version?
Thanks,
Kevin
Quoted 8 lines
Other than that, the patch looks good. Although I trust you, I like to verify the test results.
Kind regards,
Wolfram
Quoted 9 lines
Quoted 3 lines
I didn't get this trace on my ARM32 board (Renesas Lager). Anything more I need to do besides activating those options and using i2c-dev?
CONFIG_DEBUG_OBJECTS_FREEis also needed. I got this bug warning when running the OF unittest, so you may need to enable the following kernel options if you also want to reproduce this by using that.
CONFIG_OF_UNITTEST
CONFIG_OF_OVERLAY
Great! With all these config switches I could reproduce the error and see it is gone with your patch (still two backtraces left, but this is something else). Thanks for the heads up!
Quoted 2 lines
Yes, there is a conflict with the latest kernel. Do I need to resend a new version?
Nope, it was trivial and I fixed it anyhow for my testing.
Thanks for doing it and for your patience!
Wolfram Sang
AppliedQuoted 68 lines · Kevin Hao
The struct cdev is embedded in the struct
i2c_dev. In the current code,
we would free thei2c_devstruct directly input_i2c_dev(), but the
cdev is manged by a kobject, and the release of it is not predictable.
So it is very possible that thei2c_devis freed before the cdev is
entirely released. We can easily get the following call trace withCONFIG_DEBUG_KOBJECT_RELEASEandCONFIG_DEBUG_OBJECTS_TIMERSenabled.
ODEBUG: free active (active state 0) object type:timer_listhint:delayed_work_timer_fn+0x0/0x38WARNING: CPU: 19 PID: 1 at lib/debugobjects.c:325 debug_print_object+0xb0/0xf0 Modules linked in: CPU: 19 PID: 1 Comm: swapper/0 Tainted: G W 5.2.20-yocto-standard+ #120 Hardware name: Marvell OcteonTX CN96XX board (DT) pstate: 80c00089 (Nzcv daIf +PAN +UAO) pc : debug_print_object+0xb0/0xf0 lr : debug_print_object+0xb0/0xf0 sp : ffff00001292f7d0 x29: ffff00001292f7d0 x28: ffff800b82151788 x27: 0000000000000001 x26: ffff800b892c0000 x25: ffff0000124a2558 x24: 0000000000000000 x23: ffff00001107a1d8 x22: ffff0000116b5088 x21: ffff800bdc6afca8 x20: ffff000012471ae8 x19: ffff00001168f2c8 x18: 0000000000000010 x17: 00000000fd6f304b x16: 00000000ee79de43 x15: ffff800bc0e80568 x14: 79616c6564203a74 x13: 6e6968207473696c x12: 5f72656d6974203a x11: ffff0000113f0018 x10: 0000000000000000 x9 : 000000000000001f x8 : 0000000000000000 x7 : ffff0000101294cc x6 : 0000000000000000 x5 : 0000000000000000 x4 : 0000000000000001 x3 : 00000000ffffffff x2 : 0000000000000000 x1 : 387fc15c8ec0f200 x0 : 0000000000000000 Call trace: debug_print_object+0xb0/0xf0 __debug_check_no_obj_freed+0x19c/0x228 debug_check_no_obj_freed+0x1c/0x28 kfree+0x250/0x440 put_i2c_dev+0x68/0x78 i2cdev_detach_adapter+0x60/0xc8 i2cdev_notifier_call+0x3c/0x70 notifier_call_chain+0x8c/0xe8 blocking_notifier_call_chain+0x64/0x88 device_del+0x74/0x380 device_unregister+0x54/0x78 i2c_del_adapter+0x278/0x2d0 unittest_i2c_bus_remove+0x3c/0x80 platform_drv_remove+0x30/0x50 device_release_driver_internal+0xf4/0x1c0 driver_detach+0x58/0xa0 bus_remove_driver+0x84/0xd8 driver_unregister+0x34/0x60 platform_driver_unregister+0x20/0x30 of_unittest_overlay+0x8d4/0xbe0 of_unittest+0xae8/0xb3c do_one_initcall+0xac/0x450 do_initcall_level+0x208/0x224 kernel_init_freeable+0x2d8/0x36c kernel_init+0x18/0x108 ret_from_fork+0x10/0x1c irq event stamp: 3934661 hardirqs last enabled at (3934661): [<ffff00001009fa04>] debug_exception_exit+0x4c/0x58 hardirqs last disabled at (3934660): [<ffff00001009fb14>] debug_exception_enter+0xa4/0xe0 softirqs last enabled at (3934654): [<ffff000010081d94>] __do_softirq+0x46c/0x628 softirqs last disabled at (3934649): [<ffff0000100b4a1c>] irq_exit+0x104/0x118This is a common issue when using cdev embedded in a struct. Fortunately, we already have a mechanism to solve this kind of issue. Please see commit
233ed09d7fda("chardev: add helper function to register char devs with a struct device") for more detail.
Quoted 5 lines
In this patch, we choose to embed the struct device into the
i2c_dev, and use the API provided by the commit233ed09d7fdato make sure that the release ofi2c_devand cdev are in sequence.Signed-off-by Kevin Hao <[email protected]>
Applied to for-next with stable added, thanks!