Re: edac KASAN warning in experimental arm64 allmodconfig boot

From: John Garry
Date: Mon Oct 14 2019 - 12:56:15 EST

On 14/10/2019 17:15, James Morse wrote:
Hi John,

Hi James,

On 14/10/2019 16:18, John Garry wrote:
I'm experimenting by trying to boot an allmodconfig arm64 kernel, as mentioned here:


One thing that I noticed - it's hard to miss actually - is the amount of complaining from
KASAN about the EDAC/ghes code. Maybe this is something I should not care about/red
herring, or maybe something genuine. Let me know what you think.

Hmmm, I thought I tested this recently...

Log snippet (I cut off after the first KASAN warning):

[ 70.471011][ T1] random: get_random_u32 called from new_slab+0x360/0x698 with

[ 70.478671][ T1] [Firmware Bug]: APEI: Invalid bit width + offset in GAR

(this one's for you right?)

Yeah, I'll report it. It might be already fixed.

[ 70.700412][ T1] ------------[ cut here ]------------

[ 70.802080][ T1] Call trace:
[ 70.802093][ T1] debug_print_object+0xec/0x130
[ 70.802106][ T1] __debug_check_no_obj_freed+0x114/0x290
[ 70.802119][ T1] debug_check_no_obj_freed+0x18/0x28
[ 70.802130][ T1] slab_free_freelist_hook+0x18c/0x228
[ 70.802140][ T1] kfree+0x264/0x420
[ 70.802157][ T1] _edac_mc_free+0x6c/0x210
[ 70.814163][ T1] edac_mc_free+0x68/0x88
[ 70.814177][ T1] ghes_edac_unregister+0x44/0x70
[ 70.814193][ T1] ghes_remove+0x274/0x2a0

Ugh. This must be the test driver remove thing.

Yeah, the probe, remove, probe again flow from CONFIG_DEBUG_TEST_DRIVER_REMOVE.

I've reproduced this, but had to remove the parent GHES twice. It looks like it tries to
use the first ghes_edac global variables when freeing the second. ghes_init prevents it
from re-allocating over the top.

The below diff fixes it for me.

And for me by the looks of it. That's with CONFIG_DEBUG_KOBJECT_RELEASE now unset, but I expect the same with it set.

(I'll post it as a proper patch once I've done the

diff --git a/drivers/edac/ghes_edac.c b/drivers/edac/ghes_edac.c
index d413a0bdc9ad..955b59b6aade 100644
--- a/drivers/edac/ghes_edac.c
+++ b/drivers/edac/ghes_edac.c
@@ -554,6 +554,7 @@ void ghes_edac_unregister(struct ghes *ghes)

mci = ghes_pvt->mci;
+ ghes_pvt = NULL;



BTW, I am not sure if my response to Boris was rejected due to attachments, as but it is here: