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

[Thomas Gleixner]

On Thu, 17 Sep 2015, Steven Rostedt wrote:

> On Thu, 17 Sep 2015 16:38:52 +0200 (CEST)
> Thomas Gleixner <tglx at linutronix.de> wrote:
> 
> > On Thu, 17 Sep 2015, Steven Rostedt wrote:
> > 
> > > On Thu, 17 Sep 2015 12:13:15 +0200 (CEST)
> > > Thomas Gleixner <tglx at linutronix.de> wrote:
> > > 
> > > > Digging deeper. My assumption that it's a post powerpc merge failure
> > > > turned out to be wrong.
> > > 
> > > Does 4.2 have the problem?
> > 
> > No. Neither does 
> > 
> > 4c92b5bb1422: Merge branch 'pcmcia' of git://ftp.arm.linux.org.uk/~rmk/linux-arm
> 
> What's the significance of that commit?

It's the commit before the merge of the powerpc tree.

ff474e8ca854: Merge tag 'powerpc-4.3-1' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux
 
> > It just results in a different failure mode. Instead of silently
> > hanging I get:
> > 
> > [    2.248275] Oops: Exception in kernel mode, sig: 4 [#1]
> > [    2.253633] PREEMPT lite5200
> > [    2.256584] Modules linked in:
> > [    2.259723] CPU: 0 PID: 1 Comm: swapper Not tainted 4.3.0-rc1-51179-gae80a2f-dirty #75
> > [    2.267815] task: c383c000 ti: c383a000 task.ti: c383a000
> > [    2.273330] NIP: c00e1eec LR: c00df0f4 CTR: 00000000
> > [    2.278405] REGS: c383bcd0 TRAP: 0700   Not tainted  (4.3.0-rc1-51179-gae80a2f-dirty)
> > [    2.286396] MSR: 00089032 <EE,ME,IR,DR,RI>  CR: 44824028  XER: 00000000
> > [    2.293187] 
> > GPR00: c00def84 c383bd80 c383c000 c3084000 bffffff1 00677595 c383bdd8 00000000 
> > GPR08: 00000001 00000001 00000400 00000002 24828022 00000000 c0004254 84822042 
> > GPR16: 20000000 44822042 fffff000 c3086ffc c06ce248 c383a000 c3082000 c06d0000 
> > GPR24: c383a000 00000ffc 00677595 bffffff1 c3084000 c3015bfc 00000017 c3086000 
> > [    2.323656] NIP [c00e1eec] vm_normal_page+0x0/0xdc
> > [    2.328560] LR [c00df0f4] follow_page_mask+0x260/0x4fc
> > [    2.333807] Call Trace:
> > [    2.336321] [c383bd80] [c00def84] follow_page_mask+0xf0/0x4fc (unreliable)
> > [    2.343360] [c383bdd0] [c00df4a4] __get_user_pages.part.28+0x114/0x3e0
> > [    2.350050] [c383be30] [c010e788] copy_strings+0x16c/0x2c8
> > [    2.355668] [c383bea0] [c010e91c] copy_strings_kernel+0x38/0x50
> > [    2.361730] [c383bec0] [c011057c] do_execveat_common+0x440/0x658
> > [    2.367877] [c383bf10] [c01107cc] do_execve+0x38/0x48
> > [    2.373056] [c383bf20] [c00039f0] try_to_run_init_process+0x24/0x64
> > [    2.379469] [c383bf30] [c000430c] kernel_init+0xb8/0x10c
> > [    2.384924] [c383bf40] [c0010c40] ret_from_kernel_thread+0x5c/0x64
> > [    2.391242] --- interrupt: 0 at   (null)
> > [    2.391242]     LR =   (null)
> > [    2.398263] Instruction dump:
> > [    2.401297] 01000000 00037000 00000000 00000000 f0000000 00000001 0a641e09 acde4823 
> > [    2.409237] 000f0000 179a7b00 07de2900 03ef1480 <01f78a40> 0001c200 60000000 9421fff0 
> 
> Can you objdump this and and see what that is suppose to be.

Certainly not the code at NIP [c00e1eec] vm_normal_page:

c00e1eec <vm_normal_page>:
c00e1eec:       7c 08 02 a6     mflr    r0
c00e1ef0:       90 01 00 04     stw     r0,4(r1)
c00e1ef4:       4b f2 f6 05     bl      c00114f8 <_mcount>
c00e1ef8:       94 21 ff f0     stwu    r1,-16(r1)
c00e1efc:       7c 08 02 a6     mflr    r0
c00e1f00:       90 01 00 14     stw     r0,20(r1)
c00e1f04:       70 a9 08 00     andi.   r9,r5,2048

That looks more like random data corruption.
 
> > [    2.417375] ---[ end trace 996fd312ce9c18ce ]---
> > 
> > Again, if I disable CONFIG_TRACER its gone.
> 
> You mean if you disable CONFIG_FUNCTION_TRACER?

I have to disable both to make it boot. Disabling
CONFIG_FUNCTION_TRACER changes the failure mode, but does not make it
go away.
 

> Below is the entire push of ftrace for this merge window. Not much has
> changed. Could using "unsigned long" instead of "long" with the
> MCOUNT_ADDR cause this bug?

No, because the trace merge happened after the powerpc merge. But the
powerpc merge might be a red herring and the whole issue is caused by
something else which just gets unearthed by it.

Thanks,

	tglx