Re: [BUG] Revert 0b05e2d671c4 'powerpc/32: cacheable_memcpy becomes memcpy'

[Steven Rostedt <rostedt@goodmis.org>]

On Wed, 16 Sep 2015 22:01:06 +0200 (CEST)
Thomas Gleixner <tglx@linutronix.de> wrote:

> On Wed, 16 Sep 2015, Scott Wood wrote:
> 
> > On Wed, 2015-09-16 at 18:23 +0200, Thomas Gleixner wrote:
> > > With this commit applied, my MPC5200 board fails to boot 4.3-rc1. The
> > > boot failure is silent.
> > > 
> > > Bisection pointed straight to this commit. Reverting it on top of
> > > 4.3-rc1 works like a charm.
> > 
> > Yes, we've been iterating on a fix for this, and I think the patches posted 
> > earlier today are ready to go in:
> > 
> > http://patchwork.ozlabs.org/patch/518321/
> > http://patchwork.ozlabs.org/patch/518323/
> 
> Tested-by: Thomas Gleixner <tglx@linutronix.de>
> 
> So now I have to chase down that one:
> 
> [    0.230210] ftrace: allocating 16560 entries in 49 pages
> [    0.273313] ------------[ cut here ]------------
> [    0.278048] WARNING: at /home/tglx/work/kernel/tip/tip/kernel/trace/ftrace.c:1974

OK, so this is where ftrace converts the mcount to nops.

Just to be clear, there's nothing in the command line that enables any
function tracing, is there?

> [    0.285653] Modules linked in:
> [    0.288781] CPU: 0 PID: 0 Comm: swapper Not tainted 4.2.0-50528-gfac33bf-dirty #29
> [    0.296485] task: c06563e0 ti: c06a2000 task.ti: c06a2000
> [    0.301976] NIP: c008df9c LR: c008e440 CTR: 00000000
> [    0.307027] REGS: c06a3e80 TRAP: 0700   Not tainted  (4.2.0-50528-gfac33bf-dirty)
> [    0.314631] MSR: 00021032 <ME,IR,DR,RI>  CR: 22028022  XER: 00000000
> [    0.321126]
> GPR00: 00000001 c06a3f30 c06563e0 ffffffea c3888c88 00000004 c06a3ef8 764d9364
> GPR08: 00407584 c069df92 00000000 00000001 22022028 ffffffff 03fb9000 007fff00
> GPR16: c065debc 00009032 c06b8aa0 c0011184 00000000 c06a711c 00000bb6 c38021a0
> GPR24: 00000000 00008c88 00000bb6 c06a7120 c3888c88 101d6120 c3888c88 c00e1f2c
> [    0.351444] NIP [c008df9c] ftrace_bug+0xbc/0x2d4
> [    0.356160] LR [c008e440] ftrace_process_locs+0x28c/0x6cc
> [    0.361652] Call Trace:
> [    0.364169] [c06a3f30] [c00971a0] buffer_ftrace_now.isra.50+0x28/0x64 (unreliable)
> [    0.371888] [c06a3f60] [c008e440] ftrace_process_locs+0x28c/0x6cc
> [    0.378119] [c06a3fb0] [c0610848] ftrace_init+0xac/0x14c
> [    0.383559] [c06a3fc0] [c06051fc] start_kernel+0x390/0x3a8
> [    0.389147] [c06a3ff0] [00003438] 0x3438
> [    0.393136] Instruction dump:
> [    0.396154] 483f9951 3c60c059 7fe4fb78 3863b768 7fe5fb78 483f993d 41b20214 48000144
> [    0.404060] 3d20c06a 3929df92 88090002 68000001 <0f000000> 2f800000 409e01b4 3d20c06a
> [    0.412203] ---[ end trace dc8fa200cb88537f ]---
> [    0.416912] ftrace failed to modify [<c00e1f2c>] change_protection+0x8/0x2c
> [    0.424015]  actual: 74:66:74:70
> [    0.427324] ftrace record flags: 0
> [    0.430785]  (0)   expected tramp: c0011198
> 
> Happens somewhere between the powerpc merge and rc1. System boots up
> to userspace and then locks hard ....

This could be a symptom and not the problem. What the above shows is
that ftrace tried to convert the mcount at change_protection but what
it expected was there wasn't. Unfortunately, it doesn't state exactly
what it wants (that would take a arch specific function to do that, and
this is in generic code). But what it found was "74 66 74 70", which I
have no idea what type of command that is.

-- Steve