New: System reboot on S3 sleep/wakeup test

https://bugzilla.kernel.org/show_bug.cgi?id=219383

            Bug ID: 219383
           Summary: System reboot on S3 sleep/wakeup test
           Product: Platform Specific/Hardware
           Version: 2.5
          Hardware: All
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P3
         Component: x86-64
          Assignee: [email protected]
          Reporter: [email protected]
        Regression: No

I'm working for LG laptops, and I have run serveral LG PC with ubuntu OS. You may know, most LG laptops has intel soc.
I found out a critical issue, system reboot on S3 sleep/wake up.

Enviornments:
- PC BIOS : Phoenix Technologies
- Intel Jasperlake or Intel Lunarlake
- OS Ubuntu 22.04(Jasperlake), 24.04.1(Lunarlake)
- linux kernel version 6.x.0(Jasperlake) or up-to-date 6.11(Lunarlake)

Symptom:

Running the aging scripts like below, system reboots.
-------------------------
#!/bin/bash
<snip>
for (( i=1; i<=10000 ; i++ ))
sudo rtcwake -m mem -s 10 >> ${LOG} 2>&1
<snip>
-------------------------
The scripts works like below,
1. waits 10 secs
2. echo mem > /sys/power/state
3. waits 10 secs again and wake up system like press power button.

My analysis:

I had reproduced several times to find that BIOS side triggered the system reboots.

| pm_suspend() | syscore_suspend() | acpi_suspend_enter() | ... |  < BIOS > | 
...| acpi_suspend_enter() |  syscore_resume() | ...|

Debugging on BIOS, TPM2 can generate cold reset when it detects something wrong after TPM resuming.
In the BIOS code, if there are active PCR banks that are not supported by the Platform mask, it supposes to be update the TPM allocations and reboot the machine.

It means that something in linux kernel side can effect operations of tpm when going to sleep.
So, I have debuggered and traced the functions related to tpm, such as tpm_chip_start whenever the symptoms represented.

In normal case, tpm_chip_start() called once like below,
tpm_pm_suspend()-> tpm_chip_start().
but issued case, additionally called below
hwrng_fillfn ->
rng_get_data ->

    tpm_hwrng_read ->
      tpm_get_random ->
        tpm_find_get_ops ->
           tpm_try_get_ops ->
             tpm_chip_start ->

I found out that when running hwrng_fillfn(), related to Hardware random number generator, called during system_sleep, it can cause system reboots. To Verify it, I have tested with custom kernel which includes below patch.

-----------------------
From 373e92bb6d471c5fb42bacb97a4caf5375df5522 Mon Sep 17 00:00:00 2001
From: mike Seo <[email protected]>
Date: Thu, 10 Oct 2024 14:04:57 +0900
Subject: [PATCH] test_patch

test_patch for reboot while sleep/wakeup

Signed-off-by mike Seo <[email protected]>

---
 drivers/char/hw_random/core.c | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

drivers/char/hw_random/core.c +17 -0

index 57c51efa5..d3f0059a4 100644
@@ -25,6 +25,7 @@
 #include <linux/slab.h>
 #include <linux/string.h>
 #include <linux/uaccess.h>
+#include <linux/suspend.h>
 
 #define RNG_MODULE_NAME                "hw_random"
 

@@ -469,6 +470,22 @@ static struct attribute *rng_dev_attrs[] = {
 
 ATTRIBUTE_GROUPS(rng_dev);
 
+
+static int hwrng_pm_notification(struct notifier_block *nb, unsigned long
 action, void *data)
+{
+
+       switch (action) {
+       case PM_SUSPEND_PREPARE:
+               is_suspend_prepare = 1;
+               break;
+       default:
+               is_suspend_prepare = 0;
+               break;
+       }
+       return 0;
+}
+
+static struct notifier_block pm_notifier = { .notifier_call =
 hwrng_pm_notification };
 static int hwrng_fillfn(void *unused)

 {
        size_t entropy, entropy_credit = 0; /* in 1/1024 of a bit */

drivers/char/hw_random/core.c +4 -0

@@ -478,6 +495,9 @@ static int hwrng_fillfn(void *unused)
                unsigned short quality;
                struct hwrng *rng;
 
+               while (is_suspend_prepare)
+                       msleep(500);
+
                rng = get_current_rng();
                if (IS_ERR(rng) || !rng)
                        break;

@@ -549,6 +569,7 @@ int hwrng_register(struct hwrng *rng)
                        goto out_unlock;
        }
        mutex_unlock(&rng_mutex);
+       WARN_ON(register_pm_notifier(&pm_notifier));
        return 0;
 out_unlock:
        mutex_unlock(&rng_mutex);

Signature · 15 lines

2.43.0
------------------------

And I had passed over 10000 times of s3 wake/sleep aging test.

Can you make some patches for this issue and merges?

Thank you,
Mike

-- 
You may reply to this email to add a comment.

You are receiving this mail because:
You are watching the assignee of the bug.

Looks like TPM. CCing the proper people.