Re: [PATCH 06/10] watchdog: ebc-c384_wdt: Utilize the ISA bus driver

On Wed, May 11, 2016 at 01:04:34PM -0400, Sasha Levin wrote:

On 04/07/2016 10:47 AM, William Breathitt Gray wrote:

quoted

The WinSystems EBC-C384 watchdog timer is controlled via ISA bus
communication. As such, the ISA bus driver is more appropriate than the
platform driver for the WinSystems EBC-C384 watchdog timer driver.

Signed-off-by: William Breathitt Gray <redacted>

Hey William,

I'm seeing this on boot:

kernel BUG at drivers/base/driver.c:153!
invalid opcode: 0000 [#1] PREEMPT SMP KASAN
Modules linked in:
CPU: 1 PID: 1 Comm: swapper/0 Not tainted 4.6.0-rc7-next-20160511-sasha-00024-g13dfe33 #3081
task: ffff88005b2f8000 ti: ffff88005b300000 task.ti: ffff88005b300000
RIP: driver_register (drivers/base/driver.c:153 (discriminator 1))
RSP: 0000:ffff88005b307c68  EFLAGS: 00010282
RAX: 0000000000000000 RBX: ffffffffb3987c30 RCX: 1ffffffff68acf26
RDX: 0000000000000000 RSI: dffffc0000000000 RDI: ffffffffb4567930
RBP: ffff88005b307c88 R08: 1ffffffff606e3dc R09: dffffc0000000000
R10: 0000000080000000 R11: 1ffffffff79c42e2 R12: ffffffffb45678a0
R13: ffffffffb3987c38 R14: ffffffffb3987c30 R15: 0000000000000000
FS:  0000000000000000(0000) GS:ffff880063e40000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 0000000030023000 CR4: 00000000000406a0
Stack:
1ffff1000b660fa4 dffffc0000000000 0000000000000000 ffffffffb3987c00
ffff88005b307cf0 ffffffffa5bf826d ffff88005b307dc0 ffffffffb3987c30
ffffffffb3987ca8 ffffffffb39847e0 ffffffffb39847e0 00000000f673090a
Call Trace:
isa_register_driver (drivers/base/isa.c:123)
dio48e_driver_init (drivers/gpio/gpio-104-dio-48e.c:396)
do_one_initcall (init/main.c:770)
kernel_init_freeable (init/main.c:834 init/main.c:843 init/main.c:861 init/main.c:1008)
kernel_init (init/main.c:936)
ret_from_fork (arch/x86/entry/entry_64.S:390)
Code: be 00 00 00 00 00 fc ff df 48 89 f9 48 c1 e9 03 80 3c 31 00 74 05 e8 4a 18 be fd 49 83 bc 24 90 00 00 00 00 75 13 e8 2a 89 a0 fd <0f> 0b 48 c7 c7 40 21 51 b4 e8 6f 78 58 ff e8 17 89 a0 fd 49 8d
All code
========
  0:   be 00 00 00 00          mov    $0x0,%esi
  5:   00 fc                   add    %bh,%ah
  7:   ff df                   lcallq *<internal disassembler error>
  9:   48 89 f9                mov    %rdi,%rcx
  c:   48 c1 e9 03             shr    $0x3,%rcx
 10:   80 3c 31 00             cmpb   $0x0,(%rcx,%rsi,1)
 14:   74 05                   je     0x1b
 16:   e8 4a 18 be fd          callq  0xfffffffffdbe1865
 1b:   49 83 bc 24 90 00 00    cmpq   $0x0,0x90(%r12)
 22:   00 00
 24:   75 13                   jne    0x39
 26:   e8 2a 89 a0 fd          callq  0xfffffffffda08955
 2b:*  0f 0b                   ud2             <-- trapping instruction
 2d:   48 c7 c7 40 21 51 b4    mov    $0xffffffffb4512140,%rdi
 34:   e8 6f 78 58 ff          callq  0xffffffffff5878a8
 39:   e8 17 89 a0 fd          callq  0xfffffffffda08955
 3e:   49 8d 00                lea    (%r8),%rax

Code starting with the faulting instruction
===========================================
  0:   0f 0b                   ud2
  2:   48 c7 c7 40 21 51 b4    mov    $0xffffffffb4512140,%rdi
  9:   e8 6f 78 58 ff          callq  0xffffffffff58787d
  e:   e8 17 89 a0 fd          callq  0xfffffffffda0892a
 13:   49 8d 00                lea    (%r8),%rax
RIP driver_register (drivers/base/driver.c:153 (discriminator 1))
RSP <ffff88005b307c68>
---[ end trace 96103422392be10c ]---

Hi Sasha,

I believe this bug arose from a BUG_ON inside driver_register
(specifically line 153 of drivers/base/driver.c):

    BUG_ON(!drv->bus->p)

The 'p' member of the struct bus_type has not been initialized by the
time driver_register was called for the gpio-104-dio-48e driver. This
member is typically initialized by bus_register, which is called inside
isa_bus_init (drivers/base/isa.c).

The isa_bus_init function address is set to be called by the
device_initcall macro. This causes a race condition between the
isa_bus_init function and isa drivers init functions (which use
module_init).

I believe this can be resolved by changing device_initcall to
postcore_initcall, thus ensuring that isa_bus_init is called before any
isa drivers are registered. I will implement a fix and test it out, then
submit the patch if I don't encounter any errors.

Thanks,

William Breathitt Gray