Re: threshold_init_device/kobject_uevent_env oops

[Yinghai Lu]

On Jan 25, 2008 11:24 PM, Greg KH <gregkh@xxxxxxx> wrote:
>
> On Fri, Jan 25, 2008 at 11:08:53PM -0800, Yinghai Lu wrote:
> > On Jan 25, 2008 10:14 PM, Greg KH <gregkh@xxxxxxx> wrote:
> > >
> > > On Fri, Jan 25, 2008 at 10:04:19PM -0800, Yinghai Lu wrote:
> > > > On Jan 25, 2008 2:50 PM, Greg KH <gregkh@xxxxxxx> wrote:
> > > > > On Fri, Jan 25, 2008 at 02:47:11PM -0800, Greg KH wrote:
> > > > > > On Fri, Jan 25, 2008 at 11:35:56PM +0100, Ingo Molnar wrote:
> > > > > > >
> > > > > > > * Greg KH <gregkh@xxxxxxx> wrote:
> > > > > > >
> > > > > > > > On Fri, Jan 25, 2008 at 01:05:40PM -0800, Yinghai Lu wrote:
> > > > > > > > > current linus tree + x86.git
> > > > > > > > >
> > > > > > > > > got
> > > > > > > > >
> > > > > > > > > Calling initcall 0xffffffff80b93d98: threshold_init_device+0x0/0x3f()
> > > > > > > > > BUG: unable to handle kernel NULL pointer dereference at 0000000000000040
> > > > > > > > > IP: [<ffffffff80458e20>] kobject_uevent_env+0x2a/0x3d9
> > > > > > > >
> > > > > > > > Does this happen on just Linus's tree?
> > > > > > > >
> > > > > > > > Can you send me a .config file for this?
> > > > > > > >
> > > > > > > > What is threshold_init()?  Is it something new in the x86.git tree?
> > > > > > >
> > > > > > > no. A quick grep shows that it is in a file that _your_ changes in
> > > > > > > Linus' latest have touched:
> > > > > > >
> > > > > > >   arch/x86/kernel/cpu/mcheck/mce_amd_64.c
> > > > > >
> > > > > > Ok, those are pretty much just search/and/replace type changes, but I
> > > > > > have been running x86-64 boxes with these changes in place.
> > > > >
> > > > > Oh wait, I do see a change.  We are now (finally) emitting a kobject
> > > > > uevent for these devices, which somehow the code can't handle properly.
> > > > >
> > > > > Let me go poke this some more, unfortunatly I don't have any AMD 64
> > > > > boxes here anymore, only Intel based processors, so I can't run this
> > > > > module...
> > > >
> > > > it only happens with AMD Quad Core CPU or Fam 10h.
> > > >
> > > > works well with AMD opteron Rev E, and Rev F.
> > >
> > > So this only dies on a multi-core system?  Or does 2 processor boxes
> > > work, but not 4?
> >
> > 2 sockets x quad core will fail (fam 10h)
> > 2 sockets x dual core works....( rev E, and rev F opteron)
> >
> > there are some changs between opteron and fam10h.  fam10h may have
> > more local vectors for MCE...
> > or more banks and blocks...
> >
> > will look at AMD64 Bios and kernel porting guide for Fam 10h again..
> >
> > wonder if your code uncover some bugs ...
>
> No, the logic in this function is just crazy.  It's recursive, but we
> can circumvent the creation for the kobject and whole creation of the
> threshold_block if some conditions are met.  That's why we see the
> allocate_threshold_blocks so many times in the callstack, yet only a few
> kobjects created.

i produced one patch that remove the recursive. will test it and your
patch Monday.

YH
diff --git a/arch/x86/kernel/cpu/mcheck/mce_amd_64.c b/arch/x86/kernel/cpu/mcheck/mce_amd_64.c
index 65621fd..5c4cb21 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_amd_64.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_amd_64.c
@@ -553,8 +552,9 @@ static __cpuinit int threshold_create_device(unsigned int cpu)
 	unsigned int bank;
 	int err = 0;
 
+	printk(KERN_DEBUG "threshold_create_device: cpu %d, bank_map=%02x\n", cpu, per_cpu(bank_map,cpu));
 	for (bank = 0; bank < NR_BANKS; ++bank) {
-		if (!(per_cpu(bank_map, cpu) & 1 << bank))
+		if (!(per_cpu(bank_map, cpu) & (1 << bank)))
 			continue;
 		err = threshold_create_bank(cpu, bank);
 		if (err)
@@ -637,7 +637,7 @@ static void threshold_remove_device(unsigned int cpu)
 	unsigned int bank;
 
 	for (bank = 0; bank < NR_BANKS; ++bank) {
-		if (!(per_cpu(bank_map, cpu) & 1 << bank))
+		if (!(per_cpu(bank_map, cpu) & (1 << bank)))
 			continue;
 		threshold_remove_bank(cpu, bank);
 	}