Re: HP Jornada 600-series bisected

[Paul Mundt <lethal@linux-sh.org>]

On Thu, Nov 20, 2008 at 01:16:00AM +0100, Kristoffer Ericson wrote:
> HD64461 configured at 0xb0000000 on irq 36(mapped into 64 to 79)
> HD64461: enabling PCMCIA devices
[snip]
> scsi0 : pata_platform
> ata1: PATA max PIO0 mmio cmd 0x150001f0 ctl 0x150001fe irq 77
> Unable to handle kernel NULL pointer dereference at virtual address 00000000
> pc = 00000000
> *pde = 00000000
> Oops: 0000 [#1]
> Modules linked in:
> 
> Pid : 0, Comm:              swapper
> PC is at 0x0
> PC  : 00000000 SP  : 8d291f54 SR  : 400001f1 TEA : 00000000    Not tainted
> R0  : 0000004d R1  : 00000000 R2  : 8d2c1a98 R3  : 0000000f
> R4  : 0000004d R5  : 8d29970c R6  : 00000019 R7  : ffffffff
> R8  : 8d2cec0c R9  : 00000000 R10 : 00000034 R11 : 8d2c009c
> R12 : 8d21bb80 R13 : 8d2c005c R14 : 8d013cc0
> MACH: 000071c6 MACL: 000010d8 GBR : 8c009800 PR  : 8d0036c4
> 
> Call trace: 
> [<8d0070e8>] ret_from_irq+0x0/0x10
> [<8d066130>] quicklist_trim+0x0/0x120
> [<8d003690>] do_IRQ+0x0/0x60
> [<8d003730>] default_idle+0x0/0xd0
> [<8d066130>] quicklist_trim+0x0/0x120
> [<8d21bb80>] __sched_text_start+0x0/0x4b0
> [<8d013cc0>] printk+0x0/0x20
> [<8d003782>] default_idle+0x52/0xd0
> [<8d003854>] cpu_idle+0x54/0xb0
> [<8d2a994e>] start_kernel+0x4ee/0x560
> [<8d2ae4f0>] __alloc_bootmem+0x0/0x10
> [<8d0e7810>] strlen+0x0/0x60
> [<8d002024>] _stext+0x24/0x30

On Thu, Nov 20, 2008 at 06:04:48PM +0100, Kristoffer Ericson wrote:
> 5093c9a4e41518425d42c0bb5bb92f514ec77b1d is first bad commit
> commit 5093c9a4e41518425d42c0bb5bb92f514ec77b1d
> Author: Paul Mundt <lethal@linux-sh.org>
> Date:   Mon Aug 4 14:17:13 2008 +0900
> 
>     sh: define GENERIC_HARDIRQS_NO__DO_IRQ.
> 
>     We haven't called in to __do_IRQ() in a long time, so it seems like a
>     reasonable time to switch this on by default.
> 
>     Signed-off-by: Paul Mundt <lethal@linux-sh.org>
> 
> :040000 040000 cf3a9f428a81b93dcdc7d8da101ef722b6e0159e 563fcba3d9482fd80e8f4f9b02178eb17567bbb7 M  arch
> 
> Paul, thoughts about this?
> 
The only relevant change here is the dispatch in include/linux/irq.h.
With this patch reverted, the call goes through __do_IRQ() if there is no
desc->handle_irq(). Given that the hd64461 code maps the vector that
blows up, the implication is that the hd64461 interrupt code is broken.
Indeed, looking at the code, I suspect it is the demux that is where the
problem comes from, given that the hd64461 irq code does its own special
thing and completely bypasses the generic hardirq layer for chained
handlers.

So, pending a rewrite of hd64461, we should probably just leave the
__do_IRQ() dispatch as an option there, until someone gets around to
rewriting that mess. I will conditionalize it on !HD64461 for now.