Hi Geert,
On 15.11.2022 14:06, Geert Uytterhoeven wrote:
> On Wed, Oct 26, 2022 at 9:03 PM Pierre Gondois <pierre.gondois@xxxxxxx> wrote:
>> Refcounts to DT nodes are only incremented in the function
>> and never decremented. Decrease the refcounts when necessary.
>>
>> Signed-off-by: Pierre Gondois <pierre.gondois@xxxxxxx>
> Thanks for your patch, which is now commit 3da72e18371c41a6
> ("cacheinfo: Decrement refcount in cache_setup_of_node()") in
> driver-core-next.
>
> I have bisected a refcount underflow during s2ram to this commit:
Similar issue can be reproduced with qemu/arm64 'virt' machine during boot:
cacheinfo: Unable to detect cache hierarchy for CPU 0
------------[ cut here ]------------
refcount_t: addition on 0; use-after-free.
WARNING: CPU: 0 PID: 1 at lib/refcount.c:25
refcount_warn_saturate+0xa0/0x144
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 6.1.0-rc1+ #13076
Hardware name: linux,dummy-virt (DT)
pstate: 600000c5 (nZCv daIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : refcount_warn_saturate+0xa0/0x144
lr : refcount_warn_saturate+0xa0/0x144
...
Call trace:
refcount_warn_saturate+0xa0/0x144
kobject_get+0xbc/0xc0
of_node_get+0x20/0x34
of_fwnode_get+0x34/0x54
fwnode_handle_get+0x30/0x4c
fwnode_get_nth_parent+0xec/0x104
fwnode_full_name_string+0x48/0xc0
device_node_string+0x278/0x530
pointer+0x3a0/0x4bc
vsnprintf+0x23c/0x750
vprintk_store+0xf8/0x4a0
vprintk_emit+0x110/0x360
vprintk_default+0x38/0x44
vprintk+0xc0/0xe4
_printk+0x5c/0x84
of_node_release+0x140/0x150
kobject_put+0xa4/0x120
of_node_put+0x18/0x2c
free_cache_attributes+0x13c/0x1e0
detect_cache_attributes+0x4f4/0x580
update_siblings_masks+0x28/0x300
store_cpu_topology+0x58/0x80
smp_prepare_cpus+0x38/0x120
kernel_init_freeable+0x1b8/0x350
kernel_init+0x24/0x130
ret_from_fork+0x10/0x20
irq event stamp: 74
hardirqs last enabled at (73): [<ffffb81addf20ac0>]
__up_console_sem+0x6c/0x70
hardirqs last disabled at (74): [<ffffb81addf22658>]
vprintk_store+0x298/0x4a0
softirqs last enabled at (10): [<ffffb81adde1046c>] _stext+0x46c/0x5d8
softirqs last disabled at (3): [<ffffb81adde166d0>] ____do_softirq+0x10/0x20
---[ end trace 0000000000000000 ]---
I've also observed it during the manual CPU hot plug test (the log comes
from OdroidC4 board):
# for i in /sys/devices/system/cpu/cpu[1-8]; do echo 0 >$i/online;
psci: CPU1 killed (polled 4 ms)
------------[ cut here ]------------
refcount_t: addition on 0; use-after-free.
WARNING: CPU: 2 PID: 23 at lib/refcount.c:25
refcount_warn_saturate+0xa0/0x144
Modules linked in: ipv6 meson_gxl dw_hdmi_i2s_audio dwmac_generic
rc_odroid axg_audio sclk_div clk_phase mdio_mux_meson_g12a meson_ir
snd_soc_meson_axg_tdmout snd
son_g12a_tohdmitx snd_soc_meson_codec_glue crct10dif_ce
snd_soc_meson_axg_fifo meson_gxbb_wdt reset_meson_audio_arb pwm_meson
meson_dw_hdmi meson_rng rtc_meson_vr
_platform snd_soc_meson_axg_sound_card drm_display_helper rng_core
snd_soc_meson_card_utils stmmac panfrost meson_drm drm_shmem_helper
snd_soc_meson_axg_tdm_inter
anvas nvmem_meson_efuse snd_soc_meson_axg_tdm_formatter display_connector
CPU: 2 PID: 23 Comm: cpuhp/2 Not tainted 6.1.0-rc5-next-20221115+ #5935
Hardware name: Hardkernel ODROID-C4 (DT)
pstate: 604000c9 (nZCv daIF +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : refcount_warn_saturate+0xa0/0x144
lr : refcount_warn_saturate+0xa0/0x144
...
Call trace:
refcount_warn_saturate+0xa0/0x144
kobject_get+0xbc/0xc0
of_node_get+0x20/0x34
of_fwnode_get+0x34/0x54
fwnode_handle_get+0x30/0x4c
fwnode_get_nth_parent+0xec/0x104
fwnode_full_name_string+0x48/0xc0
device_node_string+0x278/0x530
pointer+0x3a0/0x4bc
vsnprintf+0x23c/0x750
vprintk_store+0xf8/0x4a0
vprintk_emit+0x10c/0x35c
vprintk_default+0x38/0x44
vprintk+0xc0/0xe4
_printk+0x5c/0x84
of_node_release+0x140/0x150
kobject_put+0xa4/0x120
of_node_put+0x18/0x2c
free_cache_attributes+0x13c/0x1e0
cacheinfo_cpu_pre_down+0x80/0x90
cpuhp_invoke_callback+0x16c/0x2b0
cpuhp_thread_fun+0x18c/0x250
smpboot_thread_fn+0x1f4/0x2a0
kthread+0x118/0x11c
ret_from_fork+0x10/0x20
irq event stamp: 204
hardirqs last enabled at (203): [<ffff800008330260>]
put_cpu_partial+0x1bc/0x1fc
hardirqs last disabled at (204): [<ffff8000081247ac>]
vprintk_store+0x298/0x4a0
softirqs last enabled at (16): [<ffff80000801046c>] _stext+0x46c/0x5d8
softirqs last disabled at (3): [<ffff800008016700>] ____do_softirq+0x10/0x20
---[ end trace 0000000000000000 ]---
Best regards
--
Marek Szyprowski, PhD
Samsung R&D Institute Poland
