Re: [PATCH] usb: dwc2: Fix parameter type in function pointer prototype

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Fri, Jul 24, 2020 at 11:03:54PM -0700, Nathan Chancellor wrote:
> When booting up on a Raspberry Pi 4 with Control Flow Integrity checking
> enabled, the following warning/panic happens:
> 
> [    1.626435] CFI failure (target: dwc2_set_bcm_params+0x0/0x4):
> [    1.632408] WARNING: CPU: 0 PID: 32 at kernel/cfi.c:30 __cfi_check_fail+0x54/0x5c
> [    1.640021] Modules linked in:
> [    1.643137] CPU: 0 PID: 32 Comm: kworker/0:1 Not tainted 5.8.0-rc6-next-20200724-00051-g89ba619726de #1
> [    1.652693] Hardware name: Raspberry Pi 4 Model B Rev 1.2 (DT)
> [    1.658637] Workqueue: events deferred_probe_work_func
> [    1.663870] pstate: 60000005 (nZCv daif -PAN -UAO BTYPE=--)
> [    1.669542] pc : __cfi_check_fail+0x54/0x5c
> [    1.673798] lr : __cfi_check_fail+0x54/0x5c
> [    1.678050] sp : ffff8000102bbaa0
> [    1.681419] x29: ffff8000102bbaa0 x28: ffffab09e21c7000
> [    1.686829] x27: 0000000000000402 x26: ffff0000f6e7c228
> [    1.692238] x25: 00000000fb7cdb0d x24: 0000000000000005
> [    1.697647] x23: ffffab09e2515000 x22: ffffab09e069a000
> [    1.703055] x21: 4c550309df1cf4c1 x20: ffffab09e2433c60
> [    1.708462] x19: ffffab09e160dc50 x18: ffff0000f6e8cc78
> [    1.713870] x17: 0000000000000041 x16: ffffab09e0bce6f8
> [    1.719278] x15: ffffab09e1c819b7 x14: 0000000000000003
> [    1.724686] x13: 00000000ffffefff x12: 0000000000000000
> [    1.730094] x11: 0000000000000000 x10: 00000000ffffffff
> [    1.735501] x9 : c932f7abfc4bc600 x8 : c932f7abfc4bc600
> [    1.740910] x7 : 077207610770075f x6 : ffff0000f6c38f00
> [    1.746317] x5 : 0000000000000000 x4 : 0000000000000000
> [    1.751723] x3 : 0000000000000000 x2 : 0000000000000000
> [    1.757129] x1 : ffff8000102bb7d8 x0 : 0000000000000032
> [    1.762539] Call trace:
> [    1.765030]  __cfi_check_fail+0x54/0x5c
> [    1.768938]  __cfi_check+0x5fa6c/0x66afc
> [    1.772932]  dwc2_init_params+0xd74/0xd78
> [    1.777012]  dwc2_driver_probe+0x484/0x6ec
> [    1.781180]  platform_drv_probe+0xb4/0x100
> [    1.785350]  really_probe+0x228/0x63c
> [    1.789076]  driver_probe_device+0x80/0xc0
> [    1.793247]  __device_attach_driver+0x114/0x160
> [    1.797857]  bus_for_each_drv+0xa8/0x128
> [    1.801851]  __device_attach.llvm.14901095709067289134+0xc0/0x170
> [    1.808050]  bus_probe_device+0x44/0x100
> [    1.812044]  deferred_probe_work_func+0x78/0xb8
> [    1.816656]  process_one_work+0x204/0x3c4
> [    1.820736]  worker_thread+0x2f0/0x4c4
> [    1.824552]  kthread+0x174/0x184
> [    1.827837]  ret_from_fork+0x10/0x18
> 
> CFI validates that all indirect calls go to a function with the same
> exact function pointer prototype. In this case, dwc2_set_bcm_params
> is the target, which has a parameter of type 'struct dwc2_hsotg *',
> but it is being implicitly cast to have a parameter of type 'void *'
> because that is the set_params function pointer prototype. Make the
> function pointer protoype match the definitions so that there is no
> more violation.
> 
> Cc: stable@vger.kernel.org

Why does this matter for stable kernels, given that CFI is not in any
kernel tree yet?

thanks,

greg k-h