[PATCH] staging: unisys: visorbus: fix kernel BUG discovered by day0 testing

[Greg KH]

On Thu, Mar 30, 2017 at 05:43:17PM -0400, David Kershner wrote:
> Fixes: 5b6f9b95f7ae ("staging: unisys: visorbus: get rid of create_bus_type.")
> 
> Kernel day0 testing robot reported a kernel BUG at drivers/base/driver.c!
> with the following call stack:
> 
> [   14.963563] ------------[ cut here ]------------
> [   14.967298] kernel BUG at drivers/base/driver.c:153!
> [   14.970948] invalid opcode: 0000 [#1] SMP
> [   14.974013] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.11.0-rc4-00790-g0789e2c #1
> [   14.978221] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.9.3-20161025_171302-gandalf 04/01/2014
> [   14.983417] task: ffff88001ea46040 task.stack: ffffc90000008000
> [   14.987315] RIP: 0010:driver_register+0xa1/0xd0
> [   14.990044] RSP: 0000:ffffc9000000be60 EFLAGS: 00010246
> [   14.993039] RAX: 0000000000000000 RBX: ffffffff831d4c20 RCX: 0000000000000000
> [   14.997040] RDX: 000000000000004d RSI: ffffffff831d47c0 RDI: ffffffff831d4c20
> [   15.001511] RBP: ffffc9000000be78 R08: ffffc9000000be78 R09: ffffc9000000be7c
> [   15.006163] R10: 0000000000000000 R11: 0000000000000001 R12: 0000000000000000
> [   15.010068] R13: 00000000ffffffff R14: ffffffff832f3923 R15: 0000000000000000
> [   15.013715] FS:  0000000000000000(0000) GS:ffff88001fa00000(0000) knlGS:0000000000000000
> [   15.017460] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [   15.021268] CR2: 0000000000000000 CR3: 0000000003009000 CR4: 00000000000006b0
> [   15.025633] Call Trace:
> [   15.028069]  ? visorbus_register_visor_driver+0x3f/0x60
> [   15.031065]  ? init_unisys+0x3a/0x90
> [   15.033562]  ? device_resume_response+0x50/0x50
> [   15.036083]  visorinput_init+0x10/0x20
> [   15.038937]  do_one_initcall+0x9a/0x164
> [   15.041838]  ? set_debug_rodata+0x12/0x12
> [   15.045333]  kernel_init_freeable+0x11e/0x1a1
> [   15.048369]  ? rest_init+0x80/0x80
> [   15.050813]  kernel_init+0x9/0x100
> [   15.053353]  ret_from_fork+0x2c/0x40
> [   15.056009] Code: ff 85 c0 41 89 c4 75 13 48 8b 7b 70 31 f6 e8 97 16 be ff 44 89 e0 5b 41 5c 5d c3 48 89 df e8 57 e1 ff ff 44 89 e0 5b 41 5c 5d c3 <0f> 0b 48 8b 33 48 c7 c7 a0 dd d5 82 e8 ec f0 6f ff 48 8b 73 08
> [   15.065144] RIP: driver_register+0xa1/0xd0 RSP: ffffc9000000be60
> [   15.068360] ---[ end trace 7d13369c38d80a8f ]---
> 
> This bug will occur if the visorbus driver is built-in to the kernel, and
> the resulting kernel is run in an environment where visorbus devices are
> NOT supported, and an attempt is made to load any of the drivers: visorhba,
> visornic, or visorinput.
> 
> This reverts commit 5b6f9b95f7ae ("staging: unisys: visorbus: get rid of
> create_bus_type.")
> 
> Signed-off-by: David Kershner <david.kershner at unisys.com>
> Reviewed-by: Tim Sell <timothy.sell at unisys.com>
> ---
>  drivers/staging/unisys/visorbus/visorbus_main.c | 32 ++++++++++++++++++++++---
>  1 file changed, 29 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/staging/unisys/visorbus/visorbus_main.c b/drivers/staging/unisys/visorbus/visorbus_main.c
> index 4348072..9f698ab 100644
> --- a/drivers/staging/unisys/visorbus/visorbus_main.c
> +++ b/drivers/staging/unisys/visorbus/visorbus_main.c
> @@ -29,6 +29,7 @@
>  #define CURRENT_FILE_PC VISOR_BUS_PC_visorbus_main_c
>  #define POLLJIFFIES_NORMALCHANNEL 10
>  
> +static int busreg_rc = -ENODEV; /* stores the result from bus registration */
>  static struct dentry *visorbus_debugfs_dir;
>  
>  /*
> @@ -959,6 +960,9 @@ static void bus_device_info_init(
>   */
>  int visorbus_register_visor_driver(struct visor_driver *drv)
>  {
> +	if (busreg_rc < 0)
> +		return -ENODEV; /*can't register on a nonexistent bus*/

Why no extra ' ' at the beginning and end of the comment?

> +
>  	drv->driver.name = drv->name;
>  	drv->driver.bus = &visorbus_type;
>  	drv->driver.probe = visordriver_probe_device;
> @@ -1067,6 +1071,29 @@ int visorbus_register_visor_driver(struct visor_driver *drv)
>  }
>  
>  /*
> + * create_bus_type() - create and register the one-and-only one instance of
> + *                     the visor bus type (visorbus_type)
> + * Return: 0 for success, otherwise negative errno value returned by
> + *         bus_register() indicating the reason for failure
> + */
> +static int
> +create_bus_type(void)
> +{
> +	busreg_rc = bus_register(&visorbus_type);
> +	return busreg_rc;
> +}
> +
> +/*
> + * remove_bus_type() - remove the one-and-only one instance of the visor bus
> + *                     type (visorbus_type)
> + */
> +static void
> +remove_bus_type(void)
> +{
> +	bus_unregister(&visorbus_type);
> +}

Why are these two separate functions?  Just include the logic where the
functions were called, a one-line function that is only called one place
is not needed.

Also, why not set "busreg_rc" when you unregister the bus?

thanks,

greg k-h