On Mon, May 04, 2026 at 06:05:25PM +0200, Stefan Wahren wrote:
> Am 04.05.26 um 15:42 schrieb Jarkko Sakkinen:
> > Hi Stefan,
> > 
> > On Mon, Apr 20, 2026 at 12:25:21PM +0200, Stefan Wahren wrote:
> > > Hi,
> > > we use a custom i.MX93 board, which based on Phytec Phycore i.MX93 with a
> > > TPM connected via SPI. If I enable CONFIG_DEBUG_ATOMIC_SLEEP=y in our 
> > > kernel
> > > config with mainline kernel 6.18.23 and reboot our board, I will get the
> > > following warning:
> > Please provide results with the latest mainline kernel.
> > 
> Here you are
> 
> [   69.085410] do not call blocking ops when !TASK_RUNNING; state=1 set at
> [<0000000090f4020f>] prepare_to_wait_event+0x54/0x14c
> [   69.097076] WARNING: kernel/sched/core.c:9086 at __might_sleep+0x74/0x7c,
> CPU#0: systemd-shutdow/1
> [   69.106026] Modules linked in: flexcan can_dev rtc_rv3028 mse102x
> phy_can_transceiver fuse autofs4
> [   69.114991] CPU: 0 UID: 0 PID: 1 Comm: systemd-shutdow Not tainted
> 7.1.0-rc2-00005-g6d35786de281-dirty #45 PREEMPT
> [   69.125405] Hardware name: chargebyte Charge SOM Evaluation Kit (DT)
> [   69.131741] pstate: 60400009 (nZCv daif +PAN -UAO -TCO -DIT -SSBS
> BTYPE=--)
> [   69.138690] pc : __might_sleep+0x74/0x7c
> [   69.142608] lr : __might_sleep+0x74/0x7c
> [   69.146525] sp : ffff800081b7b710
> [   69.149827] x29: ffff800081b7b710 x28: 0000000000000080 x27:
> ffff000002cafe80
> [   69.156951] x26: ffff800081965000 x25: 0000000000000018 x24:
> 0000000000000000
> [   69.164075] x23: ffff800081b7b8f7 x22: 0000000000000000 x21:
> 0000000000000080
> [   69.171199] x20: 000000000000013c x19: ffff8000813e2880 x18:
> 0000000000000000
> [   69.178323] x17: 003fffffffffffff x16: 000000000017d600 x15:
> ffff00003fda1a00
> [   69.185447] x14: ffff000000132c00 x13: 00000000000002ce x12:
> 0000000000000000
> [   69.192571] x11: 00000000000000c0 x10: 00000000000009e0 x9 :
> ffff800081b7b580
> [   69.199695] x8 : ffff0000001335c0 x7 : ffff00003fda1280 x6 :
> 0000000000000000
> [   69.206819] x5 : 000000001fffffff x4 : 0000000000000001 x3 :
> 0000000000000010
> [   69.213943] x2 : 0000000000000000 x1 : 0000000000000000 x0 :
> ffff000000132b80
> [   69.221068] Call trace:
> [   69.223503]  __might_sleep+0x74/0x7c (P)
> [   69.227420]  mutex_lock+0x24/0x80
> [   69.230731]  spi_bus_lock+0x20/0x50
> [   69.234214]  tpm_tis_spi_transfer_full+0x70/0x2c4
> [   69.238912]  tpm_tis_spi_read_bytes+0x3c/0x48
> [   69.243263]  tpm_tis_status+0x58/0xf8
> [   69.246920]  wait_for_tpm_stat_cond+0x30/0x90
> [   69.251271]  wait_for_tpm_stat+0x1cc/0x2e0
> [   69.255361]  tpm_tis_send_data+0xdc/0x35c
> [   69.259365]  tpm_tis_send_main+0x74/0x18c
> [   69.263369]  tpm_tis_send+0xd4/0x13c
> [   69.266940]  tpm_transmit+0xc4/0x3c4
> [   69.270511]  tpm_transmit_cmd+0x38/0xd4
> [   69.274341]  tpm2_shutdown+0x6c/0xa4
> [   69.277912]  tpm_class_shutdown+0x60/0x88
> [   69.281916]  device_shutdown+0x130/0x258
> [   69.285833]  kernel_restart+0x44/0xa4
> [   69.289491]  __do_sys_reboot+0x114/0x244
> [   69.293408]  __arm64_sys_reboot+0x24/0x30
> [   69.297412]  invoke_syscall+0x54/0x10c
> [   69.301156]  el0_svc_common.constprop.0+0x40/0xe0
> [   69.305853]  do_el0_svc+0x1c/0x28
> [   69.309164]  el0_svc+0x38/0x138
> [   69.312301]  el0t_64_sync_handler+0xa0/0xe4
> [   69.316478]  el0t_64_sync+0x198/0x19c

Thank you. This bug has been active since IRQs were enabled in tpm_tis_spi.
It's quite surprising that getting a report on it has taken so many years.
Anyway, my conclusion is that the bug report is totally legit.

"state=1" in the dump above refers to TASK_INTERRUPTIBLE, which unrolled
the root cause (I believe).

Here's an experimental fix that uses wake_woken(). It ensures that we
read status only while task is running.

BR, Jarkko
>From bf458dd675988a4fd78859a3e6a348071ecd2695 Mon Sep 17 00:00:00 2001
From: Jarkko Sakkinen <[email protected]>
Date: Sat, 9 May 2026 19:55:30 +0300
Subject: [PATCH] tpm: tpm_tis_spi: Use wait_woken() in wait_for_tmp_stat()

wait_event_interruptible_timeout() evaluates its condition after setting
the current task state to TASK_INTERRUPTIBLE.

With CONFIG_DEBUG_ATOMIC_SLEEP this triggers a warning when the IRQ wait
path is used:

    tpm_tis_status()
      tpm_tis_spi_read_bytes()
        tpm_tis_spi_transfer_full()
          spi_bus_lock()
            mutex_lock()

Address this with the following measures:

1. Call wait_tpm_stat_cond() only while tasking is running.
2. Use wait_woken() to wait for changes.

Cc: [email protected] # v4.19+
Cc: Linus Walleij <[email protected]>
Reported-by: Stefan Wahren <[email protected]>
Closes: https://lore.kernel.org/linux-integrity/[email protected]/
Fixes: 1a339b658d9d ("tpm_tis_spi: Pass the SPI IRQ down to the driver")
Signed-off-by: Jarkko Sakkinen <[email protected]>
---
 drivers/char/tpm/tpm_tis_core.c | 40 +++++++++++++++++++--------------
 1 file changed, 23 insertions(+), 17 deletions(-)

diff --git a/drivers/char/tpm/tpm_tis_core.c b/drivers/char/tpm/tpm_tis_core.c
index 21d79ad3b164..097ad0762895 100644
--- a/drivers/char/tpm/tpm_tis_core.c
+++ b/drivers/char/tpm/tpm_tis_core.c
@@ -66,8 +66,8 @@ static int wait_for_tpm_stat(struct tpm_chip *chip, u8 mask,
 		bool check_cancel)
 {
 	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
+	DEFINE_WAIT_FUNC(wait, woken_wake_function);
 	unsigned long stop;
-	long rc;
 	u8 status;
 	bool canceled = false;
 	u8 sts_mask;
@@ -87,23 +87,29 @@ static int wait_for_tpm_stat(struct tpm_chip *chip, u8 mask,
 	/* process status changes with irq support */
 	if (sts_mask) {
 		ret = -ETIME;
-again:
-		timeout = stop - jiffies;
-		if ((long)timeout <= 0)
-			return -ETIME;
-		rc = wait_event_interruptible_timeout(*queue,
-			wait_for_tpm_stat_cond(chip, sts_mask, check_cancel,
-					       &canceled),
-			timeout);
-		if (rc > 0) {
-			if (canceled)
-				return -ECANCELED;
-			ret = 0;
-		}
-		if (rc == -ERESTARTSYS && freezing(current)) {
-			clear_thread_flag(TIF_SIGPENDING);
-			goto again;
+		add_wait_queue(queue, &wait);
+		for (;;) {
+			if (wait_for_tpm_stat_cond(chip, sts_mask, check_cancel,
+						   &canceled)) {
+				ret = canceled ? -ECANCELED : 0;
+				break;
+			}
+
+			timeout = stop - jiffies;
+			if ((long)timeout <= 0)
+				break;
+
+			if (signal_pending(current)) {
+				if (freezing(current)) {
+					clear_thread_flag(TIF_SIGPENDING);
+					continue;
+				}
+				break;
+			}
+
+			wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
 		}
+		remove_wait_queue(queue, &wait);
 	}

 	if (ret)
--
2.47.3

Reply via email to