Re: [PATCH] usb: dwc3: dwc3-omap: Fix the crash in module removal

From: Felipe Balbi
Date: Wed Dec 18 2013 - 11:22:57 EST


Hi,

On Wed, Dec 18, 2013 at 06:35:09PM +0530, George Cherian wrote:
> In the wrapper driver the child nodes are populated using
> of_platform_populate(), which does a device_add. So while
> removing the module if platform_device_unregister is called
> it leads to following crash.
>
> [ 57.676757] Unable to handle kernel NULL pointer dereference at virtual address 00000018
> [ 57.685241] pgd = edf7c000
> [ 57.687988] [00000018] *pgd=add17831, *pte=00000000, *ppte=00000000
> [ 57.694488] Internal error: Oops: 17 [#1] PREEMPT SMP ARM
> [ 57.699920] Modules linked in: g_mass_storage usb_f_mass_storage libcomposite configfs wlcore_sdio dwc3_omap(-) pixcir_i2c_ts [last unloaded: dwc3]
> [ 57.713317] CPU: 0 PID: 1580 Comm: rmmod Not tainted 3.12.4-02001-gb1278cc-dirty #21
> [ 57.721099] task: ed9f8bc0 ti: edfb0000 task.ti: edfb0000
> [ 57.726562] PC is at release_resource+0x14/0x7c
> [ 57.731109] LR is at release_resource+0x10/0x7c
> [ 57.735687] pc : [<c004e278>] lr : [<c004e274>] psr: 600f0013
> [ 57.735687] sp : edfb1eb0 ip : 600f0013 fp : 00021008
> [ 57.747192] r10: 00000000 r9 : edfb0000 r8 : c00142e8
> [ 57.752441] r7 : edfb0000 r6 : bf005198 r5 : edd18000 r4 : edfa3e80
> [ 57.759002] r3 : 00000000 r2 : 00000000 r1 : 00000010 r0 : c09bf394
> [ 57.765563] Flags: nZCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment user
> [ 57.772735] Control: 10c53c7d Table: adf7c059 DAC: 00000015
> [ 57.778503] Process rmmod (pid: 1580, stack limit = 0xedfb0248)
> [ 57.784454] Stack: (0xedfb1eb0 to 0xedfb2000)
> [ 57.788848] 1ea0: 00000000 00000001 edd18000 c03a1870
> [ 57.797088] 1ec0: edd18010 edd18000 00000000 c03a1b70 00000000 bf0051a4 edd18010 c039c86c
> [ 57.805297] 1ee0: ed927a40 edd9d9f0 edcc0990 ed97e410 ed97e444 bf00518c bf005120 ed97e410
> [ 57.813507] 1f00: bf005d54 c03a1594 c03a157c c039fe94 bf005d54 ed97e410 bf005d54 c03a0698
> [ 57.821746] 1f20: 00000000 bf005d54 c09e4f90 c039fcbc 00000000 bf005d98 00000800 c00abd2c
> [ 57.829956] 1f40: c09c9920 00000000 bf005d98 00000800 edfb1f44 33637764 616d6f5f 00000070
> [ 57.838165] 1f60: ed9f8bc0 c0014190 00000001 00000000 edfb0000 bee6ee08 00021008 c00a36e0
> [ 57.846405] 1f80: 00021088 00000800 00021088 00000081 80000010 00021088 00000800 00021088
> [ 57.854614] 1fa0: 00000081 c0014100 00021088 00000800 000210b8 00000800 c0514b00 c0514b00
> [ 57.862823] 1fc0: 00021088 00000800 00021088 00000081 00000001 bee6ebf8 bee6ee08 00021008
> [ 57.871032] 1fe0: 43098880 bee6ebb4 b6fd59bc 4309888c 80000010 000210b8 0002130c 00000000
> [ 57.879302] [<c004e278>] (release_resource+0x14/0x7c) from [<c03a1870>] (platform_device_del+0x6c/0x9c)
> [ 57.888732] [<c03a1870>] (platform_device_del+0x6c/0x9c) from [<c03a1b70>] (platform_device_unregister+0xc/0x18)
> [ 57.898986] [<c03a1b70>] (platform_device_unregister+0xc/0x18) from [<bf0051a4>] (dwc3_omap_remove_core+0xc/0x14 [dwc3_omap])
> [ 57.910400] [<bf0051a4>] (dwc3_omap_remove_core+0xc/0x14 [dwc3_omap]) from [<c039c86c>] (device_for_each_child+0x34/0x74)
> [ 57.921417] [<c039c86c>] (device_for_each_child+0x34/0x74) from [<bf00518c>] (dwc3_omap_remove+0x6c/0x78 [dwc3_omap])
> [ 57.932067] [<bf00518c>] (dwc3_omap_remove+0x6c/0x78 [dwc3_omap]) from [<c03a1594>] (platform_drv_remove+0x18/0x1c)
> [ 57.942565] [<c03a1594>] (platform_drv_remove+0x18/0x1c) from [<c039fe94>] (__device_release_driver+0x70/0xc8)
> [ 57.952606] [<c039fe94>] (__device_release_driver+0x70/0xc8) from [<c03a0698>] (driver_detach+0xb4/0xb8)
> [ 57.962127] [<c03a0698>] (driver_detach+0xb4/0xb8) from [<c039fcbc>] (bus_remove_driver+0x8c/0xd0)
> [ 57.971160] [<c039fcbc>] (bus_remove_driver+0x8c/0xd0) from [<c00abd2c>] (SyS_delete_module+0x118/0x22c)
> [ 57.980712] [<c00abd2c>] (SyS_delete_module+0x118/0x22c) from [<c0014100>] (ret_fast_syscall+0x0/0x48)
> [ 57.990051] Code: e1a04000 e59f0068 eb175e43 e5943010 (e5932018)
> [ 57.996368] ---[ end trace 2af5f15c0b475a3d ]---
>
> Replace platform_device_unregister with of_device_unregister.
>
> Signed-off-by: George Cherian <george.cherian@xxxxxx>
> ---
> drivers/usb/dwc3/dwc3-omap.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/usb/dwc3/dwc3-omap.c b/drivers/usb/dwc3/dwc3-omap.c
> index b269dbd..d241ff3 100644
> --- a/drivers/usb/dwc3/dwc3-omap.c
> +++ b/drivers/usb/dwc3/dwc3-omap.c
> @@ -323,7 +323,7 @@ static int dwc3_omap_remove_core(struct device *dev, void *c)
> {
> struct platform_device *pdev = to_platform_device(dev);
>
> - platform_device_unregister(pdev);
> + of_device_unregister(pdev);

causes a build break on x86. Who on earth is handling the OF bus ?
Nobody build tests anymore ? There are quite a few problems with this
one liner (and not all of them are related to your patch, George).

a) of_device_unregister() is *not* analogous to of_platform_populate()

b) of_device_unregister() does *not* provide a no-op in case of
!CONFIG_OF

c) it's insane that of_device_* receives a platform_device as argument
but we can't use standard platform_* API.

d) If you look at drivers/of/platform.c you'll very easily notice that
there are *no* methods for freeing previously allocated
resources. The thing was written without any knowledge of
modules. If you just grep for the internal functions, you'll
get:

$ git grep -e "^static \w\+" drivers/of/platform.c
drivers/of/platform.c:static int of_dev_node_match(struct device *dev, void *data)
drivers/of/platform.c:static struct platform_device *of_platform_device_create_pdata(
drivers/of/platform.c:static struct amba_device *of_amba_device_create(struct device_node *node,
drivers/of/platform.c:static struct amba_device *of_amba_device_create(struct device_node *node,
drivers/of/platform.c:static const struct of_dev_auxdata *of_dev_lookup(const struct of_dev_auxdata *look
drivers/of/platform.c:static int of_platform_bus_create(struct device_node *bus,

Only resource allocations, no free.

I'm sorry George, but I'm not taking this patch since it makes matters
worse. Someone needs to patch drivers/of/platform.c though. Also, make
sure you can modprobe && rmmod in a loop after fixing the build error on
x86.

--
balbi

Attachment: signature.asc
Description: Digital signature