Re: [PATCH v3 2/4] hwmon: (lm90) Use hwmon_notify_event()

From: Dmitry Osipenko
Date: Mon Feb 21 2022 - 08:59:33 EST

Next message: Stefano Garzarella: "Re: [PATCH] vhost/vsock: don't check owner in vhost_vsock_stop() while releasing"
Previous message: Michael S. Tsirkin: "Re: [syzbot] kernel BUG in vhost_get_vq_desc"
In reply to: Jon Hunter: "Re: [PATCH v3 2/4] hwmon: (lm90) Use hwmon_notify_event()"
Next in thread: Guenter Roeck: "Re: [PATCH v3 2/4] hwmon: (lm90) Use hwmon_notify_event()"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

21.02.2022 16:50, Jon Hunter пишет:
>
> On 21/02/2022 12:59, Dmitry Osipenko wrote:
>> 21.02.2022 15:56, Jon Hunter пишет:
>>>
>>> On 21/02/2022 12:36, Dmitry Osipenko wrote:
>>>> 21.02.2022 15:01, Jon Hunter пишет:
>>>>> Hi Dmitry,
>>>>>
>>>>> On 18/06/2021 22:54, Dmitry Osipenko wrote:
>>>>>> Use hwmon_notify_event() to notify userspace and thermal core about
>>>>>> temperature changes.
>>>>>>
>>>>>> Suggested-by: Guenter Roeck <linux@xxxxxxxxxxxx>
>>>>>> Signed-off-by: Dmitry Osipenko <digetx@xxxxxxxxx>
>>>>>> ---
>>>>>>     drivers/hwmon/lm90.c | 44
>>>>>> +++++++++++++++++++++++++++++++++-----------
>>>>>>     1 file changed, 33 insertions(+), 11 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/hwmon/lm90.c b/drivers/hwmon/lm90.c
>>>>>> index 2e057fad05b4..e7b678a40b39 100644
>>>>>> --- a/drivers/hwmon/lm90.c
>>>>>> +++ b/drivers/hwmon/lm90.c
>>>>>> @@ -465,6 +465,7 @@ enum lm90_temp11_reg_index {
>>>>>>     struct lm90_data {
>>>>>>         struct i2c_client *client;
>>>>>> +    struct device *hwmon_dev;
>>>>>>         u32 channel_config[4];
>>>>>>         struct hwmon_channel_info temp_info;
>>>>>>         const struct hwmon_channel_info *info[3];
>>>>>> @@ -1731,22 +1732,41 @@ static bool lm90_is_tripped(struct i2c_client
>>>>>> *client, u16 *status)
>>>>>>           if ((st & (LM90_STATUS_LLOW | LM90_STATUS_LHIGH |
>>>>>> LM90_STATUS_LTHRM)) ||
>>>>>>             (st2 & MAX6696_STATUS2_LOT2))
>>>>>> -        dev_warn(&client->dev,
>>>>>> -             "temp%d out of range, please check!\n", 1);
>>>>>> +        dev_dbg(&client->dev,
>>>>>> +            "temp%d out of range, please check!\n", 1);
>>>>>>         if ((st & (LM90_STATUS_RLOW | LM90_STATUS_RHIGH |
>>>>>> LM90_STATUS_RTHRM)) ||
>>>>>>             (st2 & MAX6696_STATUS2_ROT2))
>>>>>> -        dev_warn(&client->dev,
>>>>>> -             "temp%d out of range, please check!\n", 2);
>>>>>> +        dev_dbg(&client->dev,
>>>>>> +            "temp%d out of range, please check!\n", 2);
>>>>>>         if (st & LM90_STATUS_ROPEN)
>>>>>> -        dev_warn(&client->dev,
>>>>>> -             "temp%d diode open, please check!\n", 2);
>>>>>> +        dev_dbg(&client->dev,
>>>>>> +            "temp%d diode open, please check!\n", 2);
>>>>>>         if (st2 & (MAX6696_STATUS2_R2LOW | MAX6696_STATUS2_R2HIGH |
>>>>>>                MAX6696_STATUS2_R2THRM | MAX6696_STATUS2_R2OT2))
>>>>>> -        dev_warn(&client->dev,
>>>>>> -             "temp%d out of range, please check!\n", 3);
>>>>>> +        dev_dbg(&client->dev,
>>>>>> +            "temp%d out of range, please check!\n", 3);
>>>>>>         if (st2 & MAX6696_STATUS2_R2OPEN)
>>>>>> -        dev_warn(&client->dev,
>>>>>> -             "temp%d diode open, please check!\n", 3);
>>>>>> +        dev_dbg(&client->dev,
>>>>>> +            "temp%d diode open, please check!\n", 3);
>>>>>> +
>>>>>> +    if (st & LM90_STATUS_LLOW)
>>>>>> +        hwmon_notify_event(data->hwmon_dev, hwmon_temp,
>>>>>> +                   hwmon_temp_min, 0);
>>>>>> +    if (st & LM90_STATUS_RLOW)
>>>>>> +        hwmon_notify_event(data->hwmon_dev, hwmon_temp,
>>>>>> +                   hwmon_temp_min, 1);
>>>>>> +    if (st2 & MAX6696_STATUS2_R2LOW)
>>>>>> +        hwmon_notify_event(data->hwmon_dev, hwmon_temp,
>>>>>> +                   hwmon_temp_min, 2);
>>>>>> +    if (st & LM90_STATUS_LHIGH)
>>>>>> +        hwmon_notify_event(data->hwmon_dev, hwmon_temp,
>>>>>> +                   hwmon_temp_max, 0);
>>>>>> +    if (st & LM90_STATUS_RHIGH)
>>>>>> +        hwmon_notify_event(data->hwmon_dev, hwmon_temp,
>>>>>> +                   hwmon_temp_max, 1);
>>>>>> +    if (st2 & MAX6696_STATUS2_R2HIGH)
>>>>>> +        hwmon_notify_event(data->hwmon_dev, hwmon_temp,
>>>>>> +                   hwmon_temp_max, 2);
>>>>>
>>>>>
>>>>> We observed a random null pointer deference crash somewhere in the
>>>>> thermal core (crash log below is not very helpful) when calling
>>>>> mutex_lock(). It looks like we get an interrupt when this crash
>>>>> happens.
>>>>>
>>>>> Looking at the lm90 driver, per the above, I now see we are calling
>>>>> hwmon_notify_event() from the lm90 interrupt handler. Looking at
>>>>> hwmon_notify_event() I see that ...
>>>>>
>>>>> hwmon_notify_event()
>>>>>     --> hwmon_thermal_notify()
>>>>>       --> thermal_zone_device_update()
>>>>>         --> update_temperature()
>>>>>           --> mutex_lock()
>>>>>
>>>>> So although I don't completely understand the crash, it does seem
>>>>> that we should not be calling hwmon_notify_event() from the
>>>>> interrupt handler.
>>>>>
>>>>> BTW I have not reproduced this myself yet, so I have just been
>>>>> reviewing the code to try and understand this.
>>>>
>>>> Matt Merhar was experiencing a similar issue on T30 Ouya, but I never
>>>> managed to reproduce it on Nexus 7 and Acer A500 tablets, and couldn't
>>>> spot any problem in the code. IIRC, it was a NULL dereference of
>>>> another
>>>> pointer within that code.
>>>
>>>
>>> OK. From looking at the above I don't think we can call
>>> hwmon_notify_event() from an interrupt handler because this is going to
>>> try and request a mutex. So we need to fix that.
>>
>> The interrupt is threaded, so it can take a mutex.
>
>
> Ah yes, I clearly overlooked that detail.
>
> Good news is that I have been able to reproduce this on Jetson Xavier by
> ...
>
> $ echo 40000 | sudo tee
> /sys/devices/platform/bpmp/bpmp\:i2c/i2c-0/0-004c/hwmon/hwmon0/temp1_max
> 40000
> [ 105.890995] Unable to handle kernel NULL pointer dereference at
> virtual address 00000000000003cd
> [ 105.900105] Mem abort info:
> [ 105.903328]   ESR = 0x96000021
> [ 105.906673]   EC = 0x25: DABT (current EL), IL = 32 bits
> [ 105.912407]   SET = 0, FnV = 0
> [ 105.915751]   EA = 0, S1PTW = 0
> [ 105.919230]   FSC = 0x21: alignment fault
> [ 105.923698] Data abort info:
> [ 105.926853]   ISV = 0, ISS = 0x00000021
> [ 105.931139]   CM = 0, WnR = 0
> [ 105.934420] user pgtable: 64k pages, 48-bit VAs, pgdp=0000000101f6b600
> [ 105.941230] [00000000000003cd] pgd=0000000000000000,
> p4d=0000000000000000, pud=0000000000000000
> [ 105.950864] Internal error: Oops: 96000021 [#1] PREEMPT SMP
> [ 105.956608] Modules linked in: btrfs blake2b_generic libcrc32c xor
> xor_neon raid6_pq zstd_compress rfkill snd_soc_tegra210_mixer
> snd_soc_tegra210_adx snd_soc_tegra210_dmic snd_soc_tegra210_mvc
> snd_soc_tegra210_amx snd_soc_tegra210_sfc snd_soc_tegra210_admaif
> snd_soc_tegra_pcm snd_soc_tegra210_i2s snd_soc_tegra210_ahub uvcvideo
> videobuf2_vmalloc tegra210_adma videobuf2_memops videobuf2_v4l2 cec
> videobuf2_common drm_kms_
> helper videodev mc snd_soc_rt5659 snd_soc_rl6231 pwm_tegra
> tegra_aconnect snd_hda_codec_hdmi lm90 tegra_bpmp_thermal snd_hda_tegra
> snd_soc_tegra_audio_graph_card snd_hda_codec snd_hda_core
> phy_tegra194_p2u snd_soc_
> audio_graph_card at24 snd_soc_simple_card_utils pwm_fan pcie_tegra194
> crct10dif_ce drm ip_tables x_tables ipv6
> [ 106.032497] CPU: 3 PID: 296 Comm: irq/126-lm90 Tainted: G
> O      5.16.0-tegra-291805-gf905a41db850 #3
> [ 106.042869] Hardware name: Unknown NVIDIA Jetson AGX Xavier Developer
> Kit/NVIDIA Jetson AGX Xavier Developer Kit, BIOS v1.1.2-901d3c52ed23
> 02/14/2022
> [ 106.056392] pstate: 60400009 (nZCv daif +PAN -UAO -TCO -DIT -SSBS
> BTYPE=--)
> [ 106.063903] pc : mutex_lock+0x18/0x60
> [ 106.067750] lr : thermal_zone_device_update+0x40/0x2e0
> [ 106.073161] sp : ffff80001494fc60
> [ 106.076553] x29: ffff80001494fc60 x28: ffff59bb27801c00 x27:
> ffffa4295d826790
> [ 106.084052] x26: ffff59bb27801c00 x25: 0000000000000000 x24:
> ffff59bb27801c00
> [ 106.091541] x23: ffffa4295d826870 x22: ffff59bb27801c00 x21:
> 00000000000003cd
> [ 106.098905] x20: ffff59bb28078f08 x19: ffffffffffffffed x18:
> 0000000000000000
> [ 106.106387] x17: ffffa4295d82689c x16: ffffa4292d400b7c x15:
> 0000000000000040
> [ 106.113766] x14: ffffa4295f8889a0 x13: 0000000000000228 x12:
> 0000000000000000
> [ 106.121294] x11: 0000000000000000 x10: 0000000000000000 x9 :
> 0000000000000000
> [ 106.128793] x8 : 0000000000c2f000 x7 : 0000000000000001 x6 :
> 0000000000000000
> [ 106.136133] x5 : 006800047a8e0f07 x4 : 0000000000000000 x3 :
> 00000000000003cd
> [ 106.143473] x2 : ffff59bb27801c00 x1 : 0000000000000000 x0 :
> 00000000000003cd
> [ 106.150813] Call trace:
> [ 106.153333] mutex_lock+0x18/0x60
> [ 106.156804] hwmon_notify_event+0xfc/0x110
> [ 106.161164] 0xffffa4292d400a74
> [ 106.164417] 0xffffa4292d400b7c
> [ 106.167659] irq_thread_fn+0x2c/0xa0
> [ 106.171359] irq_thread+0x134/0x240
> [ 106.174971] kthread+0x178/0x190
> [ 106.178469] ret_from_fork+0x10/0x20
> [ 106.182187] Code: d503201f d503201f d2800001 aa0103e4 (c8e47c02)
> [ 106.188550] ---[ end trace 62bf0e0b37a16815 ]---
> [ 106.193261] Kernel panic - not syncing: Oops: Fatal exception
> [ 106.199106] SMP: stopping secondary CPUs
> [ 106.203401] Kernel Offset: 0x24294d740000 from 0xffff800010000000
> [ 106.209584] PHYS_OFFSET: 0xffffa645e0000000
> [ 106.214011] CPU features: 0x0,40000843,06400846
> [ 106.218651] Memory Limit: none
> [ 106.221822] ---[ end Kernel panic - not syncing: Oops: Fatal
> exception ]---
>
>
> I am wondering if this is some sort of race condition in the thermal
> shutdown path. I would be interested to know if you see the same.

Indeed, it feels like a race condition. I tried to reproduce again on
A500 using next-20220217 and nothing bad happens.

# sensors
nct1008-i2c-2-4c
Adapter: 7000d000.i2c
temp1: +31.0°C (low = -64.0°C, high = +30.0°C) ALARM (HIGH)
(crit = +120.0°C, hyst = +110.0°C)
temp2: +37.8°C (low = -64.0°C, high = +30.0°C) ALARM (HIGH)
(crit = +115.0°C, hyst = +105.0°C)

Next message: Stefano Garzarella: "Re: [PATCH] vhost/vsock: don't check owner in vhost_vsock_stop() while releasing"
Previous message: Michael S. Tsirkin: "Re: [syzbot] kernel BUG in vhost_get_vq_desc"
In reply to: Jon Hunter: "Re: [PATCH v3 2/4] hwmon: (lm90) Use hwmon_notify_event()"
Next in thread: Guenter Roeck: "Re: [PATCH v3 2/4] hwmon: (lm90) Use hwmon_notify_event()"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]