On 25. 11. 23, 19:10, Francesco Dolcini wrote:
> On Fri, Nov 24, 2023 at 09:33:09PM +0100, Francesco Dolcini wrote:
>> Hello all,
>> while doing some test with current [1] Linux mainline I randomly hit a
>> warning. It is not systematic and I cannot really tell when
>> it was introduced, posting here to collect some ideas.
>>
>> Amitkumar, Neeraj: to me the issue is around the bluetooth/btnxpuart
>> driver, however I could also be plain wrong.
>>
>> The issue was reproduced on a Toradex Verdin AM62 [2] that is based on a TI
>> AM625 SOC (arm64) running with a arm64 defconfig and built with GCC 9 [3].
>>
>> [ 9.599027] Loaded X.509 cert 'sforshee: 00b28ddf47aef9cea7'
>> [ 9.962266] Bluetooth: hci0: Frame reassembly failed (-84)
>> [ 9.972939] ------------[ cut here ]------------
>> [ 9.977922] serial serial0: receive_buf returns -84 (count = 6)
>> [ 9.994857] WARNING: CPU: 0 PID: 37 at drivers/tty/serdev/serdev-ttyport.c:37 ttyport_receive_buf+0xd8/0xf8
>> [ 10.004840] Modules linked in: mwifiex_sdio(+) mwifiex snd_soc_simple_card crct10dif_ce cfg80211 snd_soc_simple_card_utils k3_j72xx_bandgap rti_wdt rtc_ti_k3 btnxpuart bluetooth sa2ul ecdh_generic ecc sha256_generic tidss rfkill libsha256 drm_dma_helper snd_soc_davinci_mcasp authenc omap_mailbox snd_soc_ti_udma snd_soc_ti_edma snd_soc_ti_sdma atmel_mxt_ts ina2xx snd_soc_nau8822 ti_sn65dsi83 tc358768 ti_ads1015 tps65219_pwrbutton at24 m_can_platform industrialio_triggered_buffer drm_kms_helper m_can kfifo_buf rtc_ds1307 lm75 pwm_tiehrpwm can_dev spi_omap2_mcspi panel_lvds pwm_bl libcomposite fuse drm backlight ipv6
>> [ 10.059984] CPU: 0 PID: 37 Comm: kworker/u4:2 Not tainted 6.7.0-rc2-00147-gf1a09972a45a #1
>> [ 10.071793] Hardware name: Toradex Verdin AM62 WB on Verdin Development Board (DT)
>> [ 10.082898] Workqueue: events_unbound flush_to_ldisc
>> [ 10.091345] pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
>> [ 10.101820] pc : ttyport_receive_buf+0xd8/0xf8
>> [ 10.109712] lr : ttyport_receive_buf+0xd8/0xf8
>> [ 10.117581] sp : ffff800082b9bd20
>> [ 10.124202] x29: ffff800082b9bd20 x28: ffff00000000ee05 x27: ffff0000002f21c0
>> [ 10.134735] x26: ffff000002931820 x25: 61c8864680b583eb x24: ffff0000002f21b8
>> [ 10.145209] x23: ffff00000026e740 x22: ffff0000002f21e0 x21: ffffffffffffffac
>> [ 10.155686] x20: ffff000000da5c00 x19: 0000000000000006 x18: 0000000000000000
>> [ 10.166178] x17: ffff7fffbe0e7000 x16: ffff800080000000 x15: 000039966db1c650
>> [ 10.176564] x14: 000000000000022c x13: 000000000000022c x12: 0000000000000000
>> [ 10.186979] x11: 000000000000000a x10: 0000000000000a60 x9 : ffff800082b9bb80
>> [ 10.197352] x8 : ffff00000026f200 x7 : ffff00003fd90080 x6 : 00000000000022e5
>> [ 10.207680] x5 : 00000000410fd030 x4 : 0000000000c0000e x3 : ffff7fffbe0e7000
>> [ 10.218051] x2 : 0000000000000002 x1 : 0000000000000000 x0 : 0000000000000000
>> [ 10.228393] Call trace:
>> [ 10.233989] ttyport_receive_buf+0xd8/0xf8
>> [ 10.241224] flush_to_ldisc+0xbc/0x1a4
>> [ 10.248117] process_scheduled_works+0x16c/0x28c
>> [ 10.255851] worker_thread+0x16c/0x2e0
>> [ 10.262673] kthread+0x11c/0x128
>> [ 10.268953] ret_from_fork+0x10/0x20
>> [ 10.275460] ---[ end trace 0000000000000000 ]---
>> [ 10.294674] Bluetooth: hci0: Frame reassembly failed (-84)
>> [ 10.461657] Bluetooth: hci0: Frame reassembly failed (-84)
>> [ 10.472025] Bluetooth: hci0: Frame reassembly failed (-84)
>
> I think that what is happening is the following:
>
> -> serdev-ttyport.c:ttyport_receive_buf()
> -> btnxpuart.c:btnxpuart_receive_buf()
> -> h4_recv_buf() errors out
> -> return -84
> -> warn because ret is <0
>
> Is this the desired behavior? If I understand correct recv_buf() is supposed to
> return how many bytes it has consumed, e.g. something from 0 to count.
You understand it correctly. serdev should adopt:
commit 0468a8071d7cfb0f5bc02b0888cec4525551299f
Author: Jiri Slaby (SUSE) <[email protected]>
Date: Thu Aug 10 11:14:44 2023 +0200
tty: make counts in tty_port_client_operations hooks size_t
and propagate it further to its users. The commit was meant to exactly
avoid these mistakes (but one layer upper).
btmtkuart_receive_buf is broken the same way.
Do you want to fix it yourself?
thanks,
--
js
suse labs