1:53 p.m.
Sorry -- I see that NR_CPUS is 128 for ppc64 in crash...
But the question remains, do you want to absolutely depend
upon that always being the case in the kernel?
Dave
2:24 p.m.
New subject: poisoned per_cpu data not handled on ppc64
Dave Anderson wrote:
Sorry -- I see that NR_CPUS is 128 for ppc64 in crash...
But the question remains, do you want to absolutely depend
upon that always being the case in the kernel?
Dave
Actually, looking again at the paca code, it doesn't seem that
it is dependent upon the kernel NR_CPUS value.
So perhaps all my bitching is much ado about nothing?
Dave
3:03 p.m.
New subject: poisoned per_cpu data not handled on ppc64
crash-utility-bounces(a)redhat.com wrote on 03/09/2006 12:24:26 PM:
Dave Anderson wrote:
> Sorry -- I see that NR_CPUS is 128 for ppc64 in crash...
>
> But the question remains, do you want to absolutely depend
> upon that always being the case in the kernel?
>
> Dave
Actually, looking again at the paca code, it doesn't seem that
it is dependent upon the kernel NR_CPUS value.
Dave,
Since, we are iterating on nr_paca, though that it may not be an issue
even if the kernel changed NR_CPUS to < 128. The only issue is we are
reading NR_CPUS/8 for cpu_online_map - not using the remaining part.
Probably, it will be more clear if you the cpu_online_map size is based on
nr_paca.
Like moving the following line after calculated nr_paca:
+ readmem(symbol_value("cpu_online_map"), KVADDR,
&cpu_online_map[0],
+ nr_paca/8, "cpu_online_map", FAULT_ON_ERROR);
+
Can I go ahead and change it? - drawback of this approach is if the
NR_CPUS in the kernel is greater than 128.
or do you prefer actual allocating the online_map based on nr_paca? I will
make changes based on your suggestions.
Thanks
Haren
So perhaps all my bitching is much ado about nothing?
Dave
--
Crash-utility mailing list
Crash-utility(a)redhat.com
https://www.redhat.com/mailman/listinfo/crash-utility
4:01 p.m.
New subject: poisoned per_cpu data not handled on ppc64
Haren Myneni wrote:
crash-utility-bounces(a)redhat.com wrote on 03/09/2006 12:24:26 PM:
> Dave Anderson wrote:
>
> > Sorry -- I see that NR_CPUS is 128 for ppc64 in crash...
> >
> > But the question remains, do you want to absolutely depend
> > upon that always being the case in the kernel?
> >
> > Dave
>
> Actually, looking again at the paca code, it doesn't seem that
> it is dependent upon the kernel NR_CPUS value.
Dave,
Since, we are iterating on nr_paca, though that it may not be an issue even if the kernel
changed NR_CPUS to < 128.
The only issue is we are reading NR_CPUS/8 for cpu_online_map - not using the remaining
part.
Probably, it will be more clear if you the cpu_online_map size is based on nr_paca.
Like moving the following line after calculated nr_paca:
+ readmem(symbol_value("cpu_online_map"), KVADDR,
&cpu_online_map[0],
+ nr_paca/8, "cpu_online_map", FAULT_ON_ERROR);
+
Can I go ahead and change it? - drawback of this approach is if the NR_CPUS in the kernel
is greater than 128.
Well, if NR_CPUS goes above 128 in the kernel, then
crash will fail in kernel_init() with:
if (kt->cpus > NR_CPUS) {
error(WARNING,
"calculated number of cpus (%d) greater than compiled-in NR_CPUS
(%d)\n",
kt->cpus, NR_CPUS);
error(FATAL, "recompile crash with larger NR_CPUS\n");
}
so that's not really a problem.
or do you prefer actual allocating the online_map based on nr_paca? I will make changes
based on your suggestions.
I don't care either way -- I just want you guys to be happy. ;-)
So just tell me how you want it, i.e., either keep the patch as is,
or send an updated one.
Dave
3:15 a.m.
New subject: poisoned per_cpu data not handled on ppc64
crash-utility-bounces(a)redhat.com wrote on 03/09/2006 02:01:40 PM:
Haren Myneni wrote:
crash-utility-bounces(a)redhat.com wrote on 03/09/2006 12:24:26 PM:
> Dave Anderson wrote:
>
> > Sorry -- I see that NR_CPUS is 128 for ppc64 in crash...
> >
> > But the question remains, do you want to absolutely depend
> > upon that always being the case in the kernel?
> >
> > Dave
>
> Actually, looking again at the paca code, it doesn't seem that
> it is dependent upon the kernel NR_CPUS value.
Dave,
Since, we are iterating on nr_paca, though that it may not be an
issue even if the kernel changed NR_CPUS to < 128. The only issue is
we are reading NR_CPUS/8 for cpu_online_map - not using the remaining
part.
Probably, it will be more clear if you the cpu_online_map size is
based on nr_paca.
Like moving the following line after calculated nr_paca:
+ readmem(symbol_value("cpu_online_map"), KVADDR,
&cpu_online_map[0],
+ nr_paca/8, "cpu_online_map",
FAULT_ON_ERROR);
+
Can I go ahead and change it? - drawback of this approach is if the
NR_CPUS in the kernel is greater than 128.
Well, if NR_CPUS goes above 128 in the kernel, then
crash will fail in kernel_init() with:
if (kt->cpus > NR_CPUS) {
error(WARNING,
"calculated number of cpus (%d) greater than compiled-in
NR_CPUS (%d)\n",
kt->cpus, NR_CPUS);
error(FATAL, "recompile crash with larger NR_CPUS\n");
}
so that's not really a problem.
Dave,
The above error messages will not be printed anyway since the
ppc64_paca_init will be executed later. At this point, kt->cpus will be 1.
Hence, made minor changes to the previously posted patch.
Thanks
Haren
or do you prefer actual allocating the online_map based on nr_paca?
I will make changes based on your suggestions.
I don't care either way -- I just want you guys to be happy. ;-)
So just tell me how you want it, i.e., either keep the patch as is,
or send an updated one.
Dave
--
Crash-utility mailing list
Crash-utility(a)redhat.com
https://www.redhat.com/mailman/listinfo/crash-utility
3:04 p.m.
New subject: poisoned per_cpu data not handled on ppc64
On Thu, 2006-03-09 at 15:24 -0500, Dave Anderson wrote:
Dave Anderson wrote:
> Sorry -- I see that NR_CPUS is 128 for ppc64 in crash...
>
> But the question remains, do you want to absolutely depend
> upon that always being the case in the kernel?
>
> Dave
Actually, looking again at the paca code, it doesn't seem that
it is dependent upon the kernel NR_CPUS value.
So perhaps all my bitching is much ado about nothing?
Well, I use NR_CPUS as the max value only if I can't find out
"paca" structure length from gdb.
if (!(nr_paca = get_array_length("paca", NULL, 0)))
nr_paca = NR_CPUS;
If we can't figure out "paca" struct size from "gdb", I am sure
there are other things that will break.
Thanks,
Badari
6833
days inactive
6837
days old
devel@lists.crash-utility.osci.io
5 comments
3 participants
participants (3)
-
Badari Pulavarty -
Dave Anderson -
Haren Myneni