Michael Ellerman <mpe@xxxxxxxxxxxxxx> a écrit :
Christophe Leroy <christophe.leroy@xxxxxx> writes:
A few changes to retrieve DAR and DSISR from struct regs
instead of retrieving them directly, as they may have
changed due to a TLB miss.
Also modifies hash_page() and friends to work with virtual
data addresses instead of physical ones.
Signed-off-by: Christophe Leroy <christophe.leroy@xxxxxx>
---
arch/powerpc/kernel/entry_32.S | 4 +++
arch/powerpc/kernel/head_32.S | 19 +++++++++++---
arch/powerpc/kernel/head_32.h | 4 ++-
arch/powerpc/mm/book3s32/hash_low.S | 46
+++++++++++++++++++++-------------
arch/powerpc/mm/book3s32/mmu.c | 9 +++++--
arch/powerpc/platforms/Kconfig.cputype | 2 ++
6 files changed, 61 insertions(+), 23 deletions(-)
This is faulting with qemu mac99 model:
Key type id_resolver registered
Key type id_legacy registered
BUG: Unable to handle kernel data access on read at 0x2f0db684
Faulting instruction address: 0x00004130
Oops: Kernel access of bad area, sig: 11 [#1]
BE PAGE_SIZE=4K MMU=Hash PowerMac
Modules linked in:
CPU: 0 PID: 65 Comm: modprobe Not tainted 5.4.0-rc2-gcc49+ #63
NIP: 00004130 LR: 000008c8 CTR: b7eb86e0
REGS: f106de80 TRAP: 0300 Not tainted (5.4.0-rc2-gcc49+)
MSR: 00003012 <FP,ME,DR,RI> CR: 4106df38 XER: 20000000
DAR: 2f0db684 DSISR: 40000000
GPR00: b7ec5d64 f106df38 00000000 bf988a70 00000000 2f0db540
b7ec3620 bf988d38
GPR08: 10000880 0000d032 72656773 f106df38 b7ed10ec 00000000
b7ed3d38 b7ee8900
GPR16: bf988d10 00000001 00000000 bf988d10 b7ec3620 bf988d50
b7ee76ec b7ee7320
GPR24: 10000878 00000000 b7ee8900 00000000 10029f00 10000879
b7ee7ff4 bf988d30
NIP [00004130] 0x4130
LR [000008c8] 0x8c8
Call Trace:
[f106df38] [c0016224] ret_from_syscall+0x0/0x34 (unreliable)
--- interrupt: c01 at 0xb7ed0f50
LR = 0xb7ec5d64
Instruction dump:
db8300e0 XXXXXXXX XXXXXXXX XXXXXXXX fc00048e XXXXXXXX XXXXXXXX XXXXXXXX
60a52000 XXXXXXXX XXXXXXXX XXXXXXXX 80850144 XXXXXXXX XXXXXXXX XXXXXXXX
---[ end trace 265da51c6d8b86c5 ]---
I think I'll have to drop this series for now.
What about taking the series up to patch 13 ?
I think the problem is related to hash handling as it is working well
on the 8xx (and on the 83xx which is a nohash book3s). I'll look at
book3s hash next week.
Christophe
