Re: [PATCH v1 3/4] powerpc/code-patching: Use jump_label for testing freed initmem

[Christophe Leroy]

Le 19/05/2022 à 08:27, Christophe Leroy a écrit :
> Le 19/05/2022 à 04:17, Guenter Roeck a écrit :
> > On Tue, Mar 22, 2022 at 04:40:20PM +0100, Christophe Leroy wrote:
> > > Once init is done, initmem is freed forever so no need to
> > > test system_state at every call to patch_instruction().
> > >
> > > Use jump_label.
> > >
> > > This reduces by 2% the time needed to activate ftrace on an 8xx.
> >
> > It also causes the qemu mpc8544ds emulation to crash.
> >
> > BUG: Unable to handle kernel data access on write at 0xc122eb34
> > Faulting instruction address: 0xc001b580
> > Oops: Kernel access of bad area, sig: 11 [#1]
> > BE PAGE_SIZE=4K MPC8544 DS
> > Modules linked in:
> > CPU: 0 PID: 1 Comm: swapper Not tainted 5.18.0-rc7-next-20220518 #1
> > NIP:  c001b580 LR: c001b560 CTR: 00000003
> > REGS: c5107dd0 TRAP: 0300   Not tainted  (5.18.0-rc7-next-20220518)
> > MSR:  00009000 <EE,ME>  CR: 24000882  XER: 00000000
> > DEAR: c122eb34 ESR: 00800000
> > GPR00: c001b560 c5107ec0 c5120020 10000000 00000000 00000078 0c000000 cfffffff
> > GPR08: c001e9ec 00000001 00000007 00000000 44000882 00000000 c0005178 00000000
> > GPR16: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> > GPR24: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 c1230000
> > NIP [c001b580] free_initmem+0x48/0xa8
> > LR [c001b560] free_initmem+0x28/0xa8
> > Call Trace:
> > [c5107ec0] [c001b560] free_initmem+0x28/0xa8 (unreliable)
> > [c5107ee0] [c00051b0] kernel_init+0x38/0x150
> > [c5107f00] [c001626c] ret_from_kernel_thread+0x5c/0x64
> > Instruction dump:
> > 3fe0c123 912a00dc 90010024 48000665 3d20c218 8929fa65 2c090000 41820058
> > 813feb34 2c090000 4082003c 39200001 <913feb34> 80010024 3cc0c114 83e1001c
> >
> > Reverting this patch fixes the problem.
>
> That's strange.
>
> I was able to reproduce the problem.
>
> Removing the __ro_after_init in front of
> DEFINE_STATIC_KEY_FALSE(init_mem_is_free) fixes the problem.
>
> I can't understand why, mark_readonly() is called after free_initmem().

Moving static_branch_enable(&init_mem_is_free) before mark_initmem_nx() 
also solves the problem.

There must be something wrong with mark_initmem_nx().

Christophe