Re: [PATCH v4 0/5] Powerpc/Watchpoint: Few important fixes

[Christophe Leroy]

Le 10/10/2019 à 08:25, Ravi Bangoria a écrit :


On 10/10/19 10:14 AM, Ravi Bangoria wrote:

@Christophe, Is patch5 works for you on 8xx?


Getting the following :

root@vgoip:~# ./ptrace-hwbreak
test: ptrace-hwbreak
tags: git_version:v5.4-rc2-710-gf0082e173fe4-dirty
PTRACE_SET_DEBUGREG, WO, len: 1: Ok
PTRACE_SET_DEBUGREG, WO, len: 2: Ok
PTRACE_SET_DEBUGREG, WO, len: 4: Ok
PTRACE_SET_DEBUGREG, WO, len: 8: Ok
PTRACE_SET_DEBUGREG, RO, len: 1: Ok
PTRACE_SET_DEBUGREG, RO, len: 2: Ok
PTRACE_SET_DEBUGREG, RO, len: 4: Ok
PTRACE_SET_DEBUGREG, RO, len: 8: Ok
PTRACE_SET_DEBUGREG, RW, len: 1: Ok
PTRACE_SET_DEBUGREG, RW, len: 2: Ok
PTRACE_SET_DEBUGREG, RW, len: 4: Ok
PTRACE_SET_DEBUGREG, RW, len: 8: Ok
PPC_PTRACE_SETHWDEBUG, MODE_EXACT, WO, len: 1: Ok
PPC_PTRACE_SETHWDEBUG, MODE_EXACT, RO, len: 1: Ok
PPC_PTRACE_SETHWDEBUG, MODE_EXACT, RW, len: 1: Ok
PPC_PTRACE_SETHWDEBUG, MODE_RANGE, DW ALIGNED, WO, len: 6: Ok
PPC_PTRACE_SETHWDEBUG, MODE_RANGE, DW ALIGNED, RO, len: 6: Fail
failure: ptrace-hwbreak


Thanks Christophe. I don't have any 8xx box. I checked qemu and it seems
qemu emulation for 8xx is not yet supported. So I can't debug this. Can
you please check why it's failing?

PPC_PTRACE_SETHWDEBUG internally uses DAWR register and probably 8xx does
not emulate DAWR logic, it only uses DABR to emulate double-word 
watchpoint.
In that case, all testcases that uses PPC_PTRACE_SETHWDEBUG should be
disabled for 8xx. I'll change [PATCH 5] accordingly and resend.

I think the MODE_EXACT ones are OK with the 8xx at the time being.


Also, do you think I should fix hw_breakpoint_validate_len() from [PARCH 1]
for 8xx? I re-checked you recent patch* to allow any address range size for
8xx. With that patch, hw_breakpoint_validate_len() won't get called at all
for 8xx.

At the time being, the 8xx emulates DABR so it has the same limitations 
as BOOK3S.
My patch needs to be rebased on top of your series and I think it needs 
some modifications, as it seems it doesn't properly handle size 1 and 
size 2 breakpoints at least.
So I think that you should leave your Patch1 as is, and I'll modify the 
validate_len() logic while rebasing my patch.

Christophe