On 5/16/23 16:36, Greg Kroah-Hartman wrote:
On Fri, May 12, 2023 at 11:33:31PM +0200, Mirsad Goran Todorovac wrote:
Hi,
On 5/9/23 04:59, Greg Kroah-Hartman wrote:
On Tue, May 09, 2023 at 01:51:35AM +0200, Mirsad Goran Todorovac wrote:
On 08. 05. 2023. 16:01, Greg Kroah-Hartman wrote:
On Mon, May 08, 2023 at 08:51:55AM +0200, Greg Kroah-Hartman wrote:
On Mon, May 08, 2023 at 08:30:07AM +0200, Mirsad Goran Todorovac wrote:
Hi,
There seems to be a kernel memory leak in the USB keyboard driver.
The leaked memory allocs are 96 and 512 bytes.
The platform is Ubuntu 22.04 LTS on a assembled AMD Ryzen 9 with X670E PG
Lightning mobo,
and Genius SlimStar i220 GK-080012 keyboard.
(Logitech M100 HID mouse is not affected by the bug.)
BIOS is:
*-firmware
description: BIOS
vendor: American Megatrends International, LLC.
physical id: 0
version: 1.21
date: 04/26/2023
size: 64KiB
The kernel is 6.3.0-torvalds-<id>-13466-gfc4354c6e5c2.
The keyboard is recognised as Chicony:
*-usb
description: Keyboard
product: CHICONY USB Keyboard
vendor: CHICONY
physical id: 2
bus info: usb@5:2
logical name: input35
logical name: /dev/input/event4
logical name: input35::capslock
logical name: input35::numlock
logical name: input35::scrolllock
logical name: input36
logical name: /dev/input/event5
logical name: input37
logical name: /dev/input/event6
logical name: input38
logical name: /dev/input/event8
version: 2.30
capabilities: usb-2.00 usb
configuration: driver=usbhid maxpower=100mA
speed=1Mbit/s
The bug is easily reproduced by unplugging the USB keyboard, waiting about a
couple of seconds,
and then reconnect and scan for memory leaks twice.
The kmemleak log is as follows [edited privacy info]:
root@hostname:/home/username# cat /sys/kernel/debug/kmemleak
unreferenced object 0xffff8dd020037c00 (size 96):
comm "systemd-udevd", pid 435, jiffies 4294892550 (age 8909.356s)
hex dump (first 32 bytes):
5d 8e 4e b9 ff ff ff ff 00 00 00 00 00 00 00 00 ].N.............
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
backtrace:
[<ffffffffb81a74be>] __kmem_cache_alloc_node+0x22e/0x2b0
[<ffffffffb8127b6e>] kmalloc_trace+0x2e/0xa0
[<ffffffffb87543d9>] class_create+0x29/0x80
[<ffffffffb8880d24>] usb_register_dev+0x1d4/0x2e0
As the call to class_create() in this path is now gone in 6.4-rc1, can
you retry that release to see if this is still there or not?
No, wait, it's still there, I was looking at a development branch of
mine that isn't sent upstream yet. And syzbot just reported the same
thing:
https://lore.kernel.org/r/00000000000058d15f05fb264013@xxxxxxxxxx
So something's wrong here, let me dig into it tomorrow when I get a
chance...
If this could help, here is the bisect of the bug (I could not discern what
could possibly be wrong):
user@host:~/linux/kernel/linux_torvalds$ git bisect log
git bisect start
# bad: [ac9a78681b921877518763ba0e89202254349d1b] Linux 6.4-rc1
git bisect bad ac9a78681b921877518763ba0e89202254349d1b
# good: [c9c3395d5e3dcc6daee66c6908354d47bf98cb0c] Linux 6.2
git bisect good c9c3395d5e3dcc6daee66c6908354d47bf98cb0c
# good: [85496c9b3bf8dbe15e2433d3a0197954d323cadc] Merge branch
'net-remove-some-rcu_bh-cruft'
git bisect good 85496c9b3bf8dbe15e2433d3a0197954d323cadc
# good: [b68ee1c6131c540a62ecd443be89c406401df091] Merge tag 'scsi-misc' of
git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
git bisect good b68ee1c6131c540a62ecd443be89c406401df091
# bad: [888d3c9f7f3ae44101a3fd76528d3dd6f96e9fd0] Merge tag 'sysctl-6.4-rc1'
of git://git.kernel.org/pub/scm/linux/kernel/git/mcgrof/linux
git bisect bad 888d3c9f7f3ae44101a3fd76528d3dd6f96e9fd0
# good: [34b62f186db9614e55d021f8c58d22fc44c57911] Merge tag
'pci-v6.4-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/pci/pci
git bisect good 34b62f186db9614e55d021f8c58d22fc44c57911
# good: [34da76dca4673ab1819830b4924bb5b436325b26] Merge tag
'for-linus-2023042601' of
git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid
git bisect good 34da76dca4673ab1819830b4924bb5b436325b26
# good: [97b2ff294381d05e59294a931c4db55276470cb5] Merge tag
'staging-6.4-rc1' of
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging
git bisect good 97b2ff294381d05e59294a931c4db55276470cb5
# good: [2025b2ca8004c04861903d076c67a73a0ec6dfca] mcb-lpc: Reallocate
memory region to avoid memory overlapping
git bisect good 2025b2ca8004c04861903d076c67a73a0ec6dfca
# bad: [d06f5a3f7140921ada47d49574ae6fa4de5e2a89] cdx: fix build failure due
to sysfs 'bus_type' argument needing to be const
git bisect bad d06f5a3f7140921ada47d49574ae6fa4de5e2a89
# good: [dcfbb67e48a2becfce7990386e985b9c45098ee5] driver core: class: use
lock_class_key already present in struct subsys_private
git bisect good dcfbb67e48a2becfce7990386e985b9c45098ee5
# bad: [6f14c02220c791d5c46b0f965b9340c58f3d503d] driver core: create
class_is_registered()
git bisect bad 6f14c02220c791d5c46b0f965b9340c58f3d503d
# good: [2f9e87f5a2941b259336c7ea6c5a1499ede4554a] driver core: Add a
comment to set_primary_fwnode() on nullifying
git bisect good 2f9e87f5a2941b259336c7ea6c5a1499ede4554a
# bad: [02fe26f25325b547b7a31a65deb0326c04bb5174] firmware_loader: Add debug
message with checksum for FW file
git bisect bad 02fe26f25325b547b7a31a65deb0326c04bb5174
# good: [884f8ce42ccec9d0bf11d8bf9f111e5961ca1c82] driver core: class:
implement class_get/put without the private pointer.
git bisect good 884f8ce42ccec9d0bf11d8bf9f111e5961ca1c82
# bad: [3f84aa5ec052dba960baca4ab8a352d43d47028e] base: soc: populate
machine name in soc_device_register if empty
git bisect bad 3f84aa5ec052dba960baca4ab8a352d43d47028e
# bad: [7b884b7f24b42fa25e92ed724ad82f137610afaf] driver core: class.c:
convert to only use class_to_subsys
git bisect bad 7b884b7f24b42fa25e92ed724ad82f137610afaf
# first bad commit: [7b884b7f24b42fa25e92ed724ad82f137610afaf] driver core:
class.c: convert to only use class_to_subsys
user@host:~/linux/kernel/linux_torvalds$
This helps a lot, thanks. I got the reference counting wrong somewhere
in here, I thought I tested this better, odd it shows up now...
I'll try to work on it this week.
I have figured out that the leak occurs on keyboard unplugging only, one
or two leaks (maybe a race condition?).
Please NOTE that the number of leaks is now odd:
root@defiant:/home/marvin# cat /sys/kernel/debug/kmemleak | grep comm
comm "systemd-udevd", pid 330, jiffies 4294892588 (age 715.772s)
comm "systemd-udevd", pid 330, jiffies 4294892588 (age 715.772s)
comm "kworker/6:0", pid 54, jiffies 4294907989 (age 654.224s)
comm "kworker/6:0", pid 54, jiffies 4294907989 (age 654.272s)
comm "kworker/6:3", pid 3046, jiffies 4294935362 (age 544.780s)
comm "kworker/6:0", pid 54, jiffies 4294964122 (age 429.740s)
comm "kworker/6:0", pid 54, jiffies 4294964122 (age 429.784s)
root@defiant:/home/marvin#
At one time unplugging keyboard generated only one leak, but only at one
time. As it requires manually unplugging keyboard, I didn't seem to find a
way to automate it, but it doesn't seem to require root access.
BTW, I've seen in syzbot output that kmemleak output has debug source file
names and line numbers. I couldn't make that work with the dbg .deb.
I will do some more homework, but this was a rough week.
I made up a patch based on code inspection alone, as I couldn't
reproduce this locally at all:
https://lore.kernel.org/r/2023051628-thumb-boaster-5680@gregkh
and it seemed to pass syzbot's tests.
I've included it here below, can you test it as well?
Hm, I only tested with a USB mouse unplug/plug cycle, maybe the issue is
a keyboard?
thanks,
greg k-h
-------------
diff --git a/drivers/base/class.c b/drivers/base/class.c
index ac1808d1a2e8..9b44edc8416f 100644
--- a/drivers/base/class.c
+++ b/drivers/base/class.c
@@ -320,6 +322,7 @@ void class_dev_iter_init(struct class_dev_iter *iter, const struct class *class,
start_knode = &start->p->knode_class;
klist_iter_init_node(&sp->klist_devices, &iter->ki, start_knode);
iter->type = type;
+ iter->sp = sp;
}
EXPORT_SYMBOL_GPL(class_dev_iter_init);
@@ -361,6 +364,7 @@ EXPORT_SYMBOL_GPL(class_dev_iter_next);
void class_dev_iter_exit(struct class_dev_iter *iter)
{
klist_iter_exit(&iter->ki);
+ subsys_put(iter->sp);
}
EXPORT_SYMBOL_GPL(class_dev_iter_exit);
diff --git a/include/linux/device/class.h b/include/linux/device/class.h
index 9deeaeb457bb..abf3d3bfb6fe 100644
--- a/include/linux/device/class.h
+++ b/include/linux/device/class.h
@@ -74,6 +74,7 @@ struct class {
struct class_dev_iter {
struct klist_iter ki;
const struct device_type *type;
+ struct subsys_private *sp;
};
int __must_check class_register(const struct class *class);
The build with the latest 6.4-rc2 and without this patch still leaked,
the build with the same commit and this patch applied was successful:
root@defiant:/home/marvin# cat /sys/kernel/debug/kmemleak
root@defiant:/home/marvin#
Tried three times, and it is a OK.
Congratulations! This had fixed the leak.
I wonder why it didn't show in the other contexts, hardware and archs?
Best regards,
Mirsad
