Hi,
In the formal KDB patch, the user can create both local and global
hardware breakpoints. But the code within kdb_bp_install_global() doesn't
install the global hardware breakpoint into each CPU. Because
kdb_bp_install_global() is only executed by the initial CPU, while other CPUs
only install its local breakpoints in kdb_bp_install_local(). That means the
only way to create a global hardware break point is to create one local
hardware breakpoint in one CPU, then switch to another CPU and create a new
local hardware breakpoint at the same address. In KDB v4.0 patch, there is no
way to create 2 breakpoints for one address. So smphdr patch is helpful
especially for KDB v4.0.
What I did in smphdr patch is to move the installation of both local
and global hardware breakpoints into function kdb_bp_install_local(), while
keep the instruction ones in the function kdb_bp_install_global(). I also
renamed the two functions according to their logic. In this way, command "bpha"
takes effect.
In addition, some kernel patches support cpu affinity. A process can be
confined to a specific subset of CPUs, even a specific CPU. With the support of
different hardware breakpoints on different CPUs, you can create at most 4*2
hardware breakpoints when debug bugs caused by running 2 processes
concurrently, 4 for each process. This feature is not commonly used, but it
also do no harm.
Do you have more advices?
Thanks.
Sonic Zhang
-----Original Message-----
From: Vamsi Krishna S . [mailto:vamsi@xxxxxxxxxx]
Sent: 2003?4?1? 16:27
To: Zhang, Sonic
Cc: Keith Owens; Hua Qin; kdb@xxxxxxxxxxx
Subject: Re: [PATCH] kdb on SMP (breakpoint hits multiple cpus simultaneously)
smphdr patch does two things AFAICS:
- s/kdb_bp_*_local/kdb_bp_*_dbreg and
s/kdb_bp_*_global/kdb_bp_*_dbinst
- keep the state of debug registers per-cpu, so that you could implement
per-cpu breakpoints, as probably intended in bph and bpha commands.
I think these patches are okay, but unrelated to my patch. Regarding
the smphdr patches, in my opinion, they are not really needed.. I don't
see how cpu-local breakpoints are useful.. how can one tell on which
particular CPU alone a given piece of code executes. Sonic Zhang,
can you please explain why this is needed and correct me if I am
wrong about the smphdr patch.
BTW: my patch is useful to detect and handle conditions in which
multiple cpus hit breakpoints at about the same time, one of them
wins the race to enter into kdb, and at the kdb prompt, the user
removes all breakpoints.. so we won't be able to find KDB breakpoints
on other cpus when they eventually enter KDB thereby causing
unhandled trap3/trap1 in kernel -> oops. My patch is needed for
both instruction-replacement breakpoints and debug-reg assisted
breakpoints. I will post an updated patch.
Thanks,
Vamsi.
On Tue, Apr 01, 2003 at 09:47:02AM +0800, Zhang, Sonic wrote:
> Hi,
>
> The "smphdr" patch only deal with the hardware breakpoints. It doesn't
> affect the logic of "int3" instruction breakpoint in the formal KDB patch. So
> if this problem occurs in the formal KDB patch, it is very possible that it
> still exists after applying the "smphdr" patch.
>
> Thanks.
>
> Sonic Zhang
>
>
> -----Original Message-----
> From: Keith Owens [mailto:kaos@xxxxxxx]
> Sent: 2003?3?31? 19:02
> To: vamsi@xxxxxxxxxx
> Cc: Hua Qin; kdb@xxxxxxxxxxx
> Subject: Re: [PATCH] kdb on SMP (breakpoint hits multiple cpus
> simultaneously)
>
>
> On Mon, 31 Mar 2003 15:40:16 +0530,
> "Vamsi Krishna S ." <vamsi@xxxxxxxxxx> wrote:
> >Here is a simple way to handle this situation. Before deciding that
> >this trap is not due to one of KDB's breakpoints, check if this
> >is a breakpoint that was recently removed.
>
> Sonic Zhang has cleaned up the breakpoint handling. I am waiting for
> feedback on these patches before including them in kdb.
> ftp://oss.sgi.com/projects/kdb/download/v4.0/kdb-smphdr-v4.0-2.4.20*
>
--
Vamsi Krishna S.
Linux Technology Center,
IBM Software Lab, Bangalore.
Ph: +91 80 5044959
Internet: vamsi@xxxxxxxxxx
|