Skip to contentSkip to conversation

perf/x86/uncore: remove event_list for snb client uncore IMC

PATCH In review [email protected] 3 participants 4 messages Nov 15, 2016 – Nov 16, 2016

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf/x86/uncore: remove event_list for snb client uncore IMC

From: Kan Liang <[email protected]>

A BUG was found by perf_fuzzer after enabled KASAN.

[  205.748005] BUG: KASAN: slab-out-of-bounds in
snb_uncore_imc_event_del+0x6c/0xa0 at addr ffff8800caa43768
[  205.758324] Read of size 8 by task perf_fuzzer/6618
[  205.763589] CPU: 0 PID: 6618 Comm: perf_fuzzer Not tainted 4.9.0-rc5
 #4
[  205.770721] Hardware name: LENOVO 10AM000AUS/SHARKBAY, BIOS
FBKT72AUS 01/26/2014
[  205.778689]  ffff8800c3c479b8 ffffffff816bb796 ffff88011ec00600
ffff8800caa43580
[  205.786759]  ffff8800c3c479e0 ffffffff812fb961 ffff8800c3c47a78
ffff8800caa43580
[  205.794850]  ffff8800caa43580 ffff8800c3c47a68 ffffffff812fbbd8
ffff8800c3c47a28
[  205.802911] Call Trace:
[  205.805559]  [<ffffffff816bb796>] dump_stack+0x63/0x8d
[  205.811135]  [<ffffffff812fb961>] kasan_object_err+0x21/0x70
[  205.817267]  [<ffffffff812fbbd8>] kasan_report_error+0x1d8/0x4c0
[  205.823752]  [<ffffffff81133275>] ? __lock_is_held+0x75/0xc0
[  205.829868]  [<ffffffff81025b12>] ?
snb_uncore_imc_read_counter+0x42/0x50
[  205.837198]  [<ffffffff810222e2>] ?
uncore_perf_event_update+0xe2/0x160
[  205.844337]  [<ffffffff812fc319>] kasan_report+0x39/0x40
[  205.850085]  [<ffffffff81025e3c>] ?

snb_uncore_imc_event_del+0x6c/0xa0

It's caused by accessing box->event_list.

For client IMC, there is no generic counters. It defines its own fixed free running counters. So event_list and n_events are unused. They can be removed safely.

---
 arch/x86/events/intel/uncore_snb.c | 12 ------------
 1 file changed, 12 deletions(-)
arch/x86/events/intel/uncore_snb.c +0 -12
index 81195cc..a3dcc12 100644
@@ -490,24 +490,12 @@ static int snb_uncore_imc_event_add(struct perf_event *event, int flags)
 
 	snb_uncore_imc_event_start(event, 0);
 
-	box->n_events++;
-
 	return 0;
 }
 
 static void snb_uncore_imc_event_del(struct perf_event *event, int flags)
 {
-	struct intel_uncore_box *box = uncore_event_to_box(event);
-	int i;
-
 	snb_uncore_imc_event_stop(event, PERF_EF_UPDATE);
-
-	for (i = 0; i < box->n_events; i++) {
-		if (event == box->event_list[i]) {
-			--box->n_events;
-			break;
-		}
-	}
 }
 
 int snb_pci2phy_map_init(int devid)
Signature · 1 line
2.5.5
#1 lore raw
Quoted 29 lines · * [email protected] <[email protected]> wrote:

From: Kan Liang <[email protected]>

A BUG was found by perf_fuzzer after enabled KASAN.

[  205.748005] BUG: KASAN: slab-out-of-bounds in

snb_uncore_imc_event_del+0x6c/0xa0 at addr ffff8800caa43768

---
 arch/x86/events/intel/uncore_snb.c | 12 ------------
 1 file changed, 12 deletions(-)
arch/x86/events/intel/uncore_snb.c +0 -3
index 81195cc..a3dcc12 100644
@@ -490,24 +490,12 @@ static int snb_uncore_imc_event_add(struct perf_event *event, int flags)
 
 	snb_uncore_imc_event_start(event, 0);
 
-	box->n_events++;
-
 	return 0;
 }
 
 static void snb_uncore_imc_event_del(struct perf_event *event, int flags)
 {
-	struct intel_uncore_box *box = uncore_event_to_box(event);

arch/x86/events/intel/uncore_snb.c static int snb_uncore_imc_event_add(struct perf_event *event, int flags) · line 510

Quoted 10 lines
arch/x86/events/intel/uncore_snb.c +0 -9
-	int i;
-
 	snb_uncore_imc_event_stop(event, PERF_EF_UPDATE);
-
-	for (i = 0; i < box->n_events; i++) {
-		if (event == box->event_list[i]) {
-			--box->n_events;
-			break;
-		}
-	}

I'll apply this fix - but could we please also make sure box->event_list[] _always_ get initialized to a sane state?

If it had a proper zero initial value in box->n_events the bug would not have triggered. So struct intel_uncore_box initialization appears to be sloppy, and that should be looked at as well...

Thanks,

Ingo

author
#2 lore raw
Quoted 34 lines · * [email protected] <[email protected]> wrote:
Quoted 29 lines

From: Kan Liang <[email protected]>

A BUG was found by perf_fuzzer after enabled KASAN.

[  205.748005] BUG: KASAN: slab-out-of-bounds in

snb_uncore_imc_event_del+0x6c/0xa0 at addr ffff8800caa43768

---
 arch/x86/events/intel/uncore_snb.c | 12 ------------
 1 file changed, 12 deletions(-)
arch/x86/events/intel/uncore_snb.c +0 -2
index 81195cc..a3dcc12 100644
@@ -490,24 +490,12 @@ static int snb_uncore_imc_event_add(struct
 perf_event *event, int flags)
 
 	snb_uncore_imc_event_start(event, 0);
 
-	box->n_events++;
-
 	return 0;
 }
 
Quoted 17 lines

arch/x86/events/intel/uncore_snb.c static int snb_uncore_imc_event_add(struct perf_event *event, int flags) · line 510

Quoted 13 lines
arch/x86/events/intel/uncore_snb.c +0 -10
 static void snb_uncore_imc_event_del(struct perf_event *event, int
 flags)  {
-	struct intel_uncore_box *box = uncore_event_to_box(event);
-	int i;
-
 	snb_uncore_imc_event_stop(event, PERF_EF_UPDATE);
-
-	for (i = 0; i < box->n_events; i++) {
-		if (event == box->event_list[i]) {
-			--box->n_events;
-			break;
-		}
-	}

I'll apply this fix - but could we please also make sure box->event_list[] _always_ get initialized to a sane state?

box is allocated by kzalloc_node. It should be always initialized to a sane state.

But the previous code only update n_events, and forget to update event_list in event add. That triggers the bug in event del.

Thanks,
Kan

Quoted 7 lines

If it had a proper zero initial value in box->n_events the bug would not have triggered. So struct intel_uncore_box initialization appears to be sloppy, and that should be looked at as well...

Thanks,

Ingo

#3 lore raw

[tip:perf/urgent] perf/x86/uncore: Fix crash by removing bogus event_list[] handling for SNB client uncore IMC

In reply to [email protected]
perf/x86/uncore: Fix crash by removing bogus event_list[] handling for SNB client uncore IMC
Commit-ID:  c499336cea8bbe15554c6fcea2138658c5395bfe
Gitweb:     http://git.kernel.org/tip/c499336cea8bbe15554c6fcea2138658c5395bfe
Author:     Kan Liang <[email protected]>
AuthorDate: Tue, 15 Nov 2016 13:40:10 -0500
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 16 Nov 2016 09:46:35 +0100

perf/x86/uncore: Fix crash by removing bogus event_list[] handling for SNB client uncore IMC

Vince Weaver reported the following bug when KASAN is enabled:

 [  205.748005] BUG: KASAN: slab-out-of-bounds in snb_uncore_imc_event_del+0x6c/0xa0 at addr ffff8800caa43768
 [  205.758324] Read of size 8 by task perf_fuzzer/6618

It's caused by accessing box->event_list.

For client IMC, there are no generic counters. It defines its own fixed free running counters. So event_list and n_events are unused.

They can be removed safely, which fixes the bug.

( There's still the separate question of how uninitialized state snuck into
this data structure - but that's a separate fix. )

Cc Peter Zijlstra <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
Cc Vince Weaver <[email protected]>
Cc Alexander Shishkin <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Alexander Shishkin <[email protected]>
Cc Linus Torvalds <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
---
 arch/x86/events/intel/uncore_snb.c | 12 ------------
 1 file changed, 12 deletions(-)
arch/x86/events/intel/uncore_snb.c +0 -12
index 81195cc..a3dcc12 100644
@@ -490,24 +490,12 @@ static int snb_uncore_imc_event_add(struct perf_event *event, int flags)
 
 	snb_uncore_imc_event_start(event, 0);
 
-	box->n_events++;
-
 	return 0;
 }
 
 static void snb_uncore_imc_event_del(struct perf_event *event, int flags)
 {
-	struct intel_uncore_box *box = uncore_event_to_box(event);
-	int i;
-
 	snb_uncore_imc_event_stop(event, PERF_EF_UPDATE);
-
-	for (i = 0; i < box->n_events; i++) {
-		if (event == box->event_list[i]) {
-			--box->n_events;
-			break;
-		}
-	}
 }
 
 int snb_pci2phy_map_init(int devid)