Re: Crash on FSL Book3E due to pte_pgprot()? (was Re: [PATCH v3 12/24] powerpc/mm: use pte helpers in generic code)

From: Aneesh Kumar K.V
Date: Wed Oct 17 2018 - 05:40:07 EST

Christophe Leroy <christophe.leroy@xxxxxx> writes:

> On 10/17/2018 12:59 AM, Michael Ellerman wrote:
>> Christophe Leroy <christophe.leroy@xxxxxx> writes:
>>
>>> Get rid of platform specific _PAGE_XXXX in powerpc common code and
>>> use helpers instead.
>>>
>>> mm/dump_linuxpagetables.c will be handled separately
>>>
>>> Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@xxxxxxxxxxxxx>
>>> Signed-off-by: Christophe Leroy <christophe.leroy@xxxxxx>
>>> ---
>>> arch/powerpc/include/asm/book3s/32/pgtable.h | 9 +++------
>>> arch/powerpc/include/asm/nohash/32/pgtable.h | 12 ++++++++----
>>> arch/powerpc/include/asm/nohash/pgtable.h | 3 +--
>>> arch/powerpc/mm/pgtable.c | 21 +++++++--------------
>>> arch/powerpc/mm/pgtable_32.c | 15 ++++++++-------
>>> arch/powerpc/mm/pgtable_64.c | 14 +++++++-------
>>> arch/powerpc/xmon/xmon.c | 12 +++++++-----
>>> 7 files changed, 41 insertions(+), 45 deletions(-)
>>
>> So turns out this patch *also* breaks my p5020ds :)
>>
>> Even with patch 4 merged, see next.
>>
>> It's the same crash:
>>
>> pcieport 2000:00:00.0: AER enabled with IRQ 480
>> Unable to handle kernel paging request for data at address 0x8000080080080000
>> Faulting instruction address: 0xc0000000000192cc
>> Oops: Kernel access of bad area, sig: 11 [#1]
>> BE SMP NR_CPUS=24 CoreNet Generic
>> Modules linked in:
>> CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.19.0-rc3-gcc7x-g98c847323b3a #1
>> NIP: c0000000000192cc LR: c0000000005d0f9c CTR: 0000000000100000
>> REGS: c0000000f31bb400 TRAP: 0300 Not tainted (4.19.0-rc3-gcc7x-g98c847323b3a)
>> MSR: 0000000080029000 <CE,EE,ME> CR: 24000224 XER: 00000000
>> DEAR: 8000080080080000 ESR: 0000000000800000 IRQMASK: 0
>> GPR00: c0000000005d0f84 c0000000f31bb688 c00000000117dc00 8000080080080000
>> GPR04: 0000000000000000 0000000000400000 00000ffbff241010 c0000000f31b8000
>> GPR08: 0000000000000000 0000000000100000 0000000000000000 c0000000012d4710
>> GPR12: 0000000084000422 c0000000012ff000 c000000000002774 0000000000000000
>> GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
>> GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
>> GPR24: 0000000000000000 0000000000000000 8000080080080000 c0000000ffff89a8
>> GPR28: c0000000f3576400 c0000000f3576410 0000000000400000 c0000000012ecc98
>> NIP [c0000000000192cc] ._memset_io+0x6c/0x9c
>> LR [c0000000005d0f9c] .fsl_qman_probe+0x198/0x928
>> Call Trace:
>> [c0000000f31bb688] [c0000000005d0f84] .fsl_qman_probe+0x180/0x928 (unreliable)
>> [c0000000f31bb728] [c0000000006432ec] .platform_drv_probe+0x60/0xb4
>> [c0000000f31bb7a8] [c00000000064083c] .really_probe+0x294/0x35c
>> [c0000000f31bb848] [c000000000640d2c] .__driver_attach+0x148/0x14c
>> [c0000000f31bb8d8] [c00000000063d7dc] .bus_for_each_dev+0xb0/0x118
>> [c0000000f31bb988] [c00000000063ff28] .driver_attach+0x34/0x4c
>> [c0000000f31bba08] [c00000000063f648] .bus_add_driver+0x174/0x2bc
>> [c0000000f31bbaa8] [c0000000006418bc] .driver_register+0x90/0x180
>> [c0000000f31bbb28] [c000000000643270] .__platform_driver_register+0x60/0x7c
>> [c0000000f31bbba8] [c000000000ee2a70] .fsl_qman_driver_init+0x24/0x38
>> [c0000000f31bbc18] [c0000000000023fc] .do_one_initcall+0x64/0x2b8
>> [c0000000f31bbcf8] [c000000000e9f480] .kernel_init_freeable+0x3a8/0x494
>> [c0000000f31bbda8] [c000000000002798] .kernel_init+0x24/0x148
>> [c0000000f31bbe28] [c0000000000009e8] .ret_from_kernel_thread+0x58/0x70
>> Instruction dump:
>> 4e800020 2ba50003 40dd003c 3925fffc 5488402e 7929f082 7d082378 39290001
>> 550a801e 7d2903a6 7d4a4378 794a0020 <91430000> 38630004 4200fff8 70a50003
>>
>>
>> Comparing a working vs broken kernel, it seems to boil down to the fact
>> that we're filtering out more PTE bits now that we use pte_pgprot() in
>> ioremap_prot().
>>
>> With the old code we get:
>> ioremap_prot: addr 0xff800000 flags 0x241215
>> ioremap_prot: addr 0xff800000 flags 0x241215
>> map_kernel_page: ea 0x8000080080080000 pa 0xff800000 pte 0xff800241215
>>
>>
>> And now we get:
>> ioremap_prot: addr 0xff800000 flags 0x241215 pte 0x241215
>> ioremap_prot: addr 0xff800000 pte 0x241215
>> ioremap_prot: addr 0xff800000 prot 0x241014
>> map_kernel_page: ea 0x8000080080080000 pa 0xff800000 pte 0xff800241014
>>
>> So we're losing 0x201, which for nohash book3e is:
>>
>> #define _PAGE_PRESENT 0x000001 /* software: pte contains a translation */
>> #define _PAGE_PSIZE_4K 0x000200
>>
>>
>> I haven't worked out if it's one or both of those that matter.
>
> At least missing _PAGE_PRESENT is an issue I believe.
>>
>> The question is what's the right way to fix it? Should pte_pgprot() not
>> be filtering those bits out on book3e?
>
> I think we should not use pte_pggrot() for that then. What about the
> below fix ?
>
> Christophe
>
> From: Christophe Leroy <christophe.leroy@xxxxxx>
> Date: Wed, 17 Oct 2018 05:56:25 +0000
> Subject: [PATCH] powerpc/mm: don't use pte_pgprot() in ioremap_prot()
>
> pte_pgprot() filters out some required flags like _PAGE_PRESENT.
>
> This patch replaces pte_pgprot() by __pgprot(pte_val())
> in ioremap_prot()
>
> Fixes: 26973fa5ac0e ("powerpc/mm: use pte helpers in generic code")
> Signed-off-by: Christophe Leroy <christophe.leroy@xxxxxx>
> ---
> arch/powerpc/mm/pgtable_32.c | 3 ++-
> arch/powerpc/mm/pgtable_64.c | 4 ++--
> 2 files changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/arch/powerpc/mm/pgtable_32.c b/arch/powerpc/mm/pgtable_32.c
> index 5877f5aa8f5d..a606e2f4937b 100644
> --- a/arch/powerpc/mm/pgtable_32.c
> +++ b/arch/powerpc/mm/pgtable_32.c
> @@ -122,7 +122,8 @@ ioremap_prot(phys_addr_t addr, unsigned long size,
> unsigned long flags)
> pte = pte_exprotect(pte);
> pte = pte_mkprivileged(pte);
>
> - return __ioremap_caller(addr, size, pte_pgprot(pte),
> __builtin_return_address(0));
> + return __ioremap_caller(addr, size, __pgprot(pte_val(pte)),
> + __builtin_return_address(0));


That means we pass the pfn bits also to __ioremap_caller right? How about