On 2018 Apr 17, Ghannam, Yazen wrote:
> > -----Original Message-----
> > From: linux-edac-ow...@vger.kernel.org <linux-edac-
> > ow...@vger.kernel.org> On Behalf Of Johannes Hirte
> > Sent: Monday, April 16, 2018 7:56 AM
> > To: Ghannam, Yazen <yazen.ghan...@amd.com>
> > Cc: linux-e...@vger.kernel.org; linux-kernel@vger.kernel.org; b...@suse.de;
> > tony.l...@intel.com; x...@kernel.org
> > Subject: Re: [PATCH 3/3] x86/MCE/AMD: Get address from already initialized
> > block
> >
> > On 2018 Apr 14, Johannes Hirte wrote:
> > > On 2018 Feb 01, Yazen Ghannam wrote:
> > > > From: Yazen Ghannam <yazen.ghan...@amd.com>
> > > >
> > > > The block address is saved after the block is initialized when
> > > > threshold_init_device() is called.
> > > >
> > > > Use the saved block address, if available, rather than trying to
> > > > rediscover it.
> > > >
> > > > We can avoid some *on_cpu() calls in the init path that will cause a
> > > > call trace when resuming from suspend.
> > > >
> > > > Cc: <sta...@vger.kernel.org> # 4.14.x
> > > > Signed-off-by: Yazen Ghannam <yazen.ghan...@amd.com>
> > > > ---
> > > > arch/x86/kernel/cpu/mcheck/mce_amd.c | 15 +++++++++++++++
> > > > 1 file changed, 15 insertions(+)
> > > >
> > > > diff --git a/arch/x86/kernel/cpu/mcheck/mce_amd.c
> > b/arch/x86/kernel/cpu/mcheck/mce_amd.c
> > > > index bf53b4549a17..8c4f8f30c779 100644
> > > > --- a/arch/x86/kernel/cpu/mcheck/mce_amd.c
> > > > +++ b/arch/x86/kernel/cpu/mcheck/mce_amd.c
> > > > @@ -436,6 +436,21 @@ static u32 get_block_address(unsigned int cpu,
> > u32 current_addr, u32 low, u32 hi
> > > > {
> > > > u32 addr = 0, offset = 0;
> > > >
> > > > + if ((bank >= mca_cfg.banks) || (block >= NR_BLOCKS))
> > > > + return addr;
> > > > +
> > > > + /* Get address from already initialized block. */
> > > > + if (per_cpu(threshold_banks, cpu)) {
> > > > + struct threshold_bank *bankp = per_cpu(threshold_banks,
> > cpu)[bank];
> > > > +
> > > > + if (bankp && bankp->blocks) {
> > > > + struct threshold_block *blockp = &bankp-
> > >blocks[block];
> > > > +
> > > > + if (blockp)
> > > > + return blockp->address;
> > > > + }
> > > > + }
> > > > +
> > > > if (mce_flags.smca) {
> > > > if (smca_get_bank_type(bank) == SMCA_RESERVED)
> > > > return addr;
> > > > --
> > > > 2.14.1
> > >
> > > I have a KASAN: slab-out-of-bounds, and git bisect points me to this
> > > change:
> > >
> > > Apr 13 00:40:32 probook kernel:
> > ================================================================
> > ==
> > > Apr 13 00:40:32 probook kernel: BUG: KASAN: slab-out-of-bounds in
> > get_block_address.isra.3+0x1e9/0x520
> > > Apr 13 00:40:32 probook kernel: Read of size 4 at addr ffff8803f165ddf4 by
> > task swapper/0/1
> > > Apr 13 00:40:32 probook kernel:
> > > Apr 13 00:40:32 probook kernel: CPU: 1 PID: 1 Comm: swapper/0 Not
> > tainted 4.16.0-10757-g4ca8ba4ccff9 #532
> > > Apr 13 00:40:32 probook kernel: Hardware name: HP HP ProBook 645
> > G2/80FE, BIOS N77 Ver. 01.12 12/19/2017
> > > Apr 13 00:40:32 probook kernel: Call Trace:
> > > Apr 13 00:40:32 probook kernel: dump_stack+0x5b/0x8b
> > > Apr 13 00:40:32 probook kernel: ? get_block_address.isra.3+0x1e9/0x520
> > > Apr 13 00:40:32 probook kernel: print_address_description+0x65/0x270
> > > Apr 13 00:40:32 probook kernel: ? get_block_address.isra.3+0x1e9/0x520
> > > Apr 13 00:40:32 probook kernel: kasan_report+0x232/0x350
> > > Apr 13 00:40:32 probook kernel: get_block_address.isra.3+0x1e9/0x520
> > > Apr 13 00:40:32 probook kernel: ? kobject_init_and_add+0xde/0x130
> > > Apr 13 00:40:32 probook kernel: ? get_name+0x390/0x390
> > > Apr 13 00:40:32 probook kernel: ? kasan_unpoison_shadow+0x30/0x40
> > > Apr 13 00:40:32 probook kernel: ? kasan_kmalloc+0xa0/0xd0
> > > Apr 13 00:40:32 probook kernel: allocate_threshold_blocks+0x12c/0xc60
> > > Apr 13 00:40:32 probook kernel: ? kobject_add_internal+0x800/0x800
> > > Apr 13 00:40:32 probook kernel: ? get_block_address.isra.3+0x520/0x520
> > > Apr 13 00:40:32 probook kernel: ? kasan_kmalloc+0xa0/0xd0
> > > Apr 13 00:40:32 probook kernel:
> > mce_threshold_create_device+0x35b/0x990
> > > Apr 13 00:40:32 probook kernel: ? init_special_inode+0x1d0/0x230
> > > Apr 13 00:40:32 probook kernel: threshold_init_device+0x98/0xa7
> > > Apr 13 00:40:32 probook kernel: ?
> > mcheck_vendor_init_severity+0x43/0x43
> > > Apr 13 00:40:32 probook kernel: do_one_initcall+0x76/0x30c
> > > Apr 13 00:40:32 probook kernel: ?
> > trace_event_raw_event_initcall_finish+0x190/0x190
> > > Apr 13 00:40:32 probook kernel: ? kasan_unpoison_shadow+0xb/0x40
> > > Apr 13 00:40:32 probook kernel: ? kasan_unpoison_shadow+0x30/0x40
> > > Apr 13 00:40:32 probook kernel: kernel_init_freeable+0x3d6/0x471
> > > Apr 13 00:40:32 probook kernel: ? rest_init+0xf0/0xf0
> > > Apr 13 00:40:32 probook kernel: kernel_init+0xa/0x120
> > > Apr 13 00:40:32 probook kernel: ? rest_init+0xf0/0xf0
> > > Apr 13 00:40:32 probook kernel: ret_from_fork+0x22/0x40
> > > Apr 13 00:40:32 probook kernel:
> > > Apr 13 00:40:32 probook kernel: Allocated by task 1:
> > > Apr 13 00:40:32 probook kernel: kasan_kmalloc+0xa0/0xd0
> > > Apr 13 00:40:32 probook kernel: kmem_cache_alloc_trace+0xf3/0x1f0
> > > Apr 13 00:40:32 probook kernel: allocate_threshold_blocks+0x1bc/0xc60
> > > Apr 13 00:40:32 probook kernel:
> > mce_threshold_create_device+0x35b/0x990
> > > Apr 13 00:40:32 probook kernel: threshold_init_device+0x98/0xa7
> > > Apr 13 00:40:32 probook kernel: do_one_initcall+0x76/0x30c
> > > Apr 13 00:40:32 probook kernel: kernel_init_freeable+0x3d6/0x471
> > > Apr 13 00:40:32 probook kernel: kernel_init+0xa/0x120
> > > Apr 13 00:40:32 probook kernel: ret_from_fork+0x22/0x40
> > > Apr 13 00:40:32 probook kernel:
> > > Apr 13 00:40:32 probook kernel: Freed by task 0:
> > > Apr 13 00:40:32 probook kernel: (stack is not available)
> > > Apr 13 00:40:32 probook kernel:
> > > Apr 13 00:40:32 probook kernel: The buggy address belongs to the object at
> > ffff8803f165dd80
> > > which belongs to the cache kmalloc-128 of size 128
> > > Apr 13 00:40:32 probook kernel: The buggy address is located 116 bytes
> > inside of
> > > 128-byte region [ffff8803f165dd80, ffff8803f165de00)
> > > Apr 13 00:40:32 probook kernel: The buggy address belongs to the page:
> > > Apr 13 00:40:32 probook kernel: page:ffffea000fc59740 count:1
> > mapcount:0 mapping:0000000000000000 index:0x0
> > > Apr 13 00:40:32 probook kernel: flags: 0x2000000000000100(slab)
> > > Apr 13 00:40:32 probook kernel: raw: 2000000000000100
> > 0000000000000000 0000000000000000 0000000180150015
> > > Apr 13 00:40:32 probook kernel: raw: dead000000000100
> > dead000000000200 ffff8803f3403340 0000000000000000
> > > Apr 13 00:40:32 probook kernel: page dumped because: kasan: bad access
> > detected
> > > Apr 13 00:40:32 probook kernel:
> > > Apr 13 00:40:32 probook kernel: Memory state around the buggy address:
> > > Apr 13 00:40:32 probook kernel: ffff8803f165dc80: fc fc fc fc fc fc fc
> > > fc 00 00
> > 00 00 00 00 00 00
> > > Apr 13 00:40:32 probook kernel: ffff8803f165dd00: 00 00 00 00 00 00 00 fc
> > fc fc fc fc fc fc fc fc
> > > Apr 13 00:40:32 probook kernel: >ffff8803f165dd80: 00 00 00 00 00 00 00
> > 00 00 00 00 00 00 fc fc fc
> > > Apr 13 00:40:32 probook kernel:
> > > ^
> > > Apr 13 00:40:32 probook kernel: ffff8803f165de00: fc fc fc fc fc fc fc
> > > fc fc fc
> > fc fc fc fc fc fc
> > > Apr 13 00:40:32 probook kernel: ffff8803f165de80: fc fc fc fc fc fc fc
> > > fc fc fc
> > fc fc fc fc fc fc
> > > Apr 13 00:40:32 probook kernel:
> > ================================================================
> > ==
> > >
> >
> > Putting the whole chaching part under the
> >
> > if (mce_flags.smca) {
> >
> > solved the issue on my Carrizo.
> >
>
> Thanks for reporting this. I'm able to reproduce this on my Fam17h system. The
> caching should still be the same on non-SMCA systems. Putting it all under the
> SMCA flags effectively removes it on Carrizo.
>
> Here are when get_block_address() is called:
> 1) Boot time MCE init. Called on each CPU. No caching.
> 2) Init of the MCE device. Called on a single CPU. Values are cached here.
> 3) CPU on/offling which calls MCE init. Should use the cached values.
>
> It seems to me that the KASAN bug is detected during #2 though it's not yet
> clear
> to me what the issue is. I need to read up on KASAN and keep debugging.
The out-of-bound access happens in get_block_address:
if (bankp && bankp->blocks) {
struct threshold_block *blockp blockp = &bankp->blocks[block];
with block=1. This doesn't exists. I don't even find any array here.
There is a linked list, created in allocate_threshold_blocks. On my
system I get 17 lists with one element each.
--
Regards,
Johannes
