Hi Tim,
On 29.02.24 01:46, Tim Harvey wrote:
> Greetings,
>
> Commit 481c2d14627d ("tpm,tpm_tis: Disable interrupts after 1000
> unhandled IRQs") introduced a kernel warning for boards that do not
> have a TPM loaded but have one defined in the device-tree (ie a
> subloaded board assembly):
>
> [ 2.434431] WARNING: CPU: 3 PID: 55 at kernel/workqueue.c:3397
> __flush_work.isra.0+0x2ac/0x2d8
> [ 2.443069] Modules linked in:
> [ 2.446133] CPU: 3 PID: 55 Comm: kworker/u8:3 Not tainted
> 6.6.8-00021-g232153790aa6-dirty #434
> [ 2.454758] Hardware name: Gateworks Venice GW73xx-2x i.MX8MP
> Development Kit (DT)
> [ 2.462337] Workqueue: events_unbound async_run_entry_fn
> [ 2.467669] pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> [ 2.474647] pc : __flush_work.isra.0+0x2ac/0x2d8
> [ 2.479278] lr : flush_work+0x10/0x1c
> [ 2.482950] sp : ffff80008270ba40
> [ 2.486271] x29: ffff80008270ba40 x28: 00000000ffffffff x27: 00000000ffffffff
> [ 2.493425] x26: ffff000000293000 x25: ffff80008218aff8 x24: ffff800080a1c6a8
> [ 2.500578] x23: 00000000ffffffff x22: ffff80008213ca18 x21: 0000000000000001
> [ 2.507731] x20: ffff000000e142c0 x19: ffff000000e14280 x18: 0000000000000000
> [ 2.514884] x17: 0000000000000000 x16: 000000000000001c x15: 0000000000000000
> [ 2.522038] x14: ffff00003fd94100 x13: 0000000000000002 x12: 0000000000000000
> [ 2.529191] x11: 0000000000000400 x10: 0000000000000910 x9 : 00000000ffffffff
> [ 2.536346] x8 : 0000000000000cc1 x7 : ffff8000804c76d8 x6 : 00000000090ee8ad
> [ 2.543500] x5 : ffff00000026c380 x4 : 0000000000000000 x3 : 0000000000000000
> [ 2.550655] x2 : 0000000000000000 x1 : ffff00000026c380 x0 : 0000000000000000
> [ 2.557806] Call trace:
> [ 2.560261] __flush_work.isra.0+0x2ac/0x2d8
> [ 2.564544] flush_work+0x10/0x1c
> [ 2.567869] tpm_tis_remove+0x8c/0xc4
> [ 2.571545] tpm_tis_core_init+0x194/0x7b8
> [ 2.575656] tpm_tis_spi_probe+0xa4/0xd4
> [ 2.579593] tpm_tis_spi_driver_probe+0x34/0x64
> [ 2.584136] spi_probe+0x84/0xe4
> [ 2.587377] really_probe+0x148/0x2c0
> [ 2.591052] __driver_probe_device+0x78/0x12c
> [ 2.595417] driver_probe_device+0xd8/0x15c
> [ 2.599609] __device_attach_driver+0xb8/0x134
> [ 2.604064] bus_for_each_drv+0x84/0xe0
> [ 2.607913] __device_attach_async_helper+0xac/0xd0
> [ 2.612808] async_run_entry_fn+0x34/0xe0
> [ 2.616832] process_one_work+0x138/0x260
> [ 2.620851] worker_thread+0x32c/0x438
> [ 2.624609] kthread+0x118/0x11c
> [ 2.627844] ret_from_fork+0x10/0x20
> [ 2.631428] ---[ end trace 0000000000000000 ]---
>
> The issue is caused by tpm_tis_remove() calling flush_work() on the
> work queue that has not yet been initialized. I can move the INIT_WORK
> call added in this commit into the probe to fix this:
I posted a fix for this issue a few weeks ago:
https://lore.kernel.org/all/20240201113646.31734-1-l.sanfilippo@xxxxxxxxxx/
It is already included in Jarkos tree.
BR,
Lino
> diff --git a/drivers/char/tpm/tpm_tis_core.c b/drivers/char/tpm/tpm_tis_core.c
> index d5ea22196b5b..e62294b3a437 100644
> --- a/drivers/char/tpm/tpm_tis_core.c
> +++ b/drivers/char/tpm/tpm_tis_core.c
> @@ -1133,6 +1133,8 @@ int tpm_tis_core_init(struct device *dev, struct
> tpm_tis_data *priv, int irq,
>
> dev_set_drvdata(&chip->dev, priv);
>
> + INIT_WORK(&priv->free_irq_work, tpm_tis_free_irq_func);
> +
> rc = tpm_tis_read32(priv, TPM_DID_VID(0), &vendor);
> if (rc < 0)
> return rc;
> @@ -1159,8 +1161,6 @@ int tpm_tis_core_init(struct device *dev, struct
> tpm_tis_data *priv, int irq,
> }
> }
>
> - INIT_WORK(&priv->free_irq_work, tpm_tis_free_irq_func);
> -
> if (chip->ops->clk_enable != NULL)
> chip->ops->clk_enable(chip, true);
>
> However I wonder if we should simply have error checking for an
> invalid vendor of 0xffffffff. Is there any reason to not return an
> -ENODEV if vendor == 0xffffffff?
>
> diff --git a/drivers/char/tpm/tpm_tis_core.c b/drivers/char/tpm/tpm_tis_core.c
> index d5ea22196b5b..1bb1b3e1b2d0 100644
> --- a/drivers/char/tpm/tpm_tis_core.c
> +++ b/drivers/char/tpm/tpm_tis_core.c
> @@ -1136,6 +1136,8 @@ int tpm_tis_core_init(struct device *dev, struct
> tpm_tis_data *priv, int irq,
> rc = tpm_tis_read32(priv, TPM_DID_VID(0), &vendor);
> if (rc < 0)
> return rc;
> + if (rc == 0xffffffff)
> + return -ENODEV;
>
> priv->manufacturer_id = vendor;
>
> I'm not sure if a hard coded 0xffffffff check is best here.
>
> Best regards,
>
> Tim
>
