About device_links_purge() warning

[Xu Yang]

Hi,

I meet a kernel warning dump in device_del() -> device_links_purge():

    list_for_each_entry_safe_reverse(link, ln, &dev->links.consumers, s_node) {
 ->         WARN_ON(link->status != DL_STATE_DORMANT &&
                    link->status != DL_STATE_NONE);
            __device_link_del(&link->kref);
    }

[  128.196320] ------------[ cut here ]------------
[  128.200932] WARNING: drivers/base/core.c:1641 at device_del+0x2b0/0x398, CPU#0: insmod/704
[  128.209188] Modules linked in: tcpci(+) dwc3_imx8mp(+) overlay snd_soc_fsl_asoc_card phy_fsl_imx8mq_usb snd_soc_imx_audmux mxc_jpeg_encdec tcpm v4l2_jpeg snd_soc_imx_card typec snd_soc_fsl_micfil snd_soc_wm8962 snd_soc_fsl_sai imx_pcm_dma flexcan can_dev snd_soc_fsl_utils pwm_fan fuse
[  128.234387] CPU: 0 UID: 0 PID: 704 Comm: insmod Not tainted 7.1.0-rc1-00016-g9186b1f497a9-dirty #505 PREEMPT
[  128.244281] Hardware name: NXP i.MX95 19X19 board (DT)
[  128.249403] pstate: 20400009 (nzCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[  128.256351] pc : device_del+0x2b0/0x398
[  128.260184] lr : device_del+0x1cc/0x398
[  128.264014] sp : ffff800082983720
[  128.267314] x29: ffff800082983750 x28: ffffbbdae88f80e0 x27: 0000000000000000
[  128.274438] x26: 0000000000000000 x25: ffff800082983cb0 x24: ffff000080a58810
[  128.281562] x23: 0000000000400100 x22: ffff00008bb65cc8 x21: ffff00008bb65cc0
[  128.288686] x20: ffffbbdae8ec4c40 x19: ffff00008bb65c10 x18: 0000000000000000
[  128.295810] x17: 3034303066316334 x16: 3a6d726f6674616c x15: 005708e68a088db3
[  128.302934] x14: 00000000000001e2 x13: 0000000000000000 x12: 0000000000000022
[  128.310058] x11: 00000000000000c0 x10: 0000000000000b20 x9 : 000000000000003c
[  128.317182] x8 : ffff800082983630 x7 : ffff000080001700 x6 : fffffdffc22bd380
[  128.324306] x5 : ffffbbdae6caef6c x4 : ffff00008ae7c000 x3 : ffff00008af4e1c0
[  128.331430] x2 : ffff00008bb65cb8 x1 : 0000000000000003 x0 : ffff000093330400
[  128.338557] Call trace:
[  128.340993]  device_del+0x2b0/0x398 (P)
[  128.344823]  platform_device_del+0x2c/0xa0
[  128.348913]  platform_device_unregister+0x14/0x3c
[  128.353611]  of_platform_device_destroy+0xf0/0x100
[  128.358392]  device_for_each_child_reverse+0x70/0xc8
[  128.363352]  of_platform_depopulate+0x34/0x80
[  128.367703]  dwc3_imx8mp_probe+0x24c/0x3dc [dwc3_imx8mp]
[  128.373004]  platform_probe+0x5c/0xa4
[  128.376664]  really_probe+0x15c/0x384
[  128.380319]  __driver_probe_device+0x1a0/0x1c0
[  128.384759]  driver_probe_device+0x3c/0x11c
[  128.388936]  __driver_attach+0x74/0x1a0
[  128.392767]  bus_for_each_dev+0x7c/0xdc
[  128.396597]  driver_attach+0x24/0x30
[  128.400168]  bus_add_driver+0x138/0x2ac
[  128.403996]  driver_register+0x68/0x130
[  128.407829]  __platform_driver_register+0x20/0x2c
[  128.412527]  dwc3_imx8mp_driver_init+0x20/0x1000 [dwc3_imx8mp]
[  128.418348]  do_one_initcall+0x60/0x1d4
[  128.422181]  do_init_module+0x54/0x244
[  128.425923]  load_module+0x16c4/0x1dc0
[  128.429669]  init_module_from_file+0xd8/0xf0
[  128.433933]  __arm64_sys_finit_module+0x1b8/0x334
[  128.438631]  invoke_syscall.constprop.0+0x40/0x108
[  128.443414]  el0_svc_common.constprop.0+0x38/0xd8
[  128.448112]  do_el0_svc+0x1c/0x28
[  128.451423]  el0_svc+0x38/0x140
[  128.454558]  el0t_64_sync_handler+0xa0/0xe4
[  128.458735]  el0t_64_sync+0x198/0x19c
[  128.462395] ---[ end trace 0000000000000000 ]---

Simply speaking, this happens when a supplier device is removed while a
consumer device is probing, particularly when their modules are loaded
at the same time.

The case as below:

usb@4c010010 {
    compatible = "fsl,imx95-dwc3", "fsl,imx8mp-dwc3";

    usb@4c100000 {
        compatible = "snps,dwc3";
        
        port {
            usb3_data_hs: endpoint {
                remote-endpoint = <&typec_con_hs>; <---+ (supplier)
            };                                         |
        };                                             |
    };                                                 |
};                                                     |
                                                       |
tcpc@50 {                                              |
    compatible = "nxp,ptn5110", "tcpci";               |
                                                       |
    connector {                                        |
        compatible = "usb-c-connector";                |
                                                       |
        port {                                         |
            typec_con_hs: endpoint {                   |
                remote-endpoint = <&usb3_data_hs>; ----+ (consumer)
            };
        };
    };
};

Node usb@4c010010 will be populated as device 4c010010.usb.
Node usb@4c100000 will be populated as device 4c100000.usb.
Node tcpc@50 will be populated as device 6-0050.

Issue Description:

1. The system adds devices 4c010010.usb and 6-0050 when it parses the OF tree.
2. When the imx8mp-dwc3 module is loaded, device 4c010010.usb gets probed,
   then dwc3_imx8mp_probe() populates a child device 4c100000.usb.
3. When 4c100000.usb is added, it creates a device link:
   [  126.021683] i2c 6-0050: Linked as a sync state only consumer to 4c100000.usb
4. During the probing of device 4c010010.usb, tcpci module is loaded, device
   6-0050 gets probed, and device_links_check_suppliers() sets the above device
   linke status as DL_STATE_CONSUMER_PROBE.
5. Probing device 4c010010.usb encounters an error, then dwc3_imx8mp_probe()
   depopulates the child device 4c100000.usb in the cleanup path. However,
   tcpci_probe() hasn't finished yet. The warning occurs.

The system doesn't prevent device 6-0050 from being probed because 6-0050
is a sync only consumer of 4c100000.usb.

Any ideas on how to avoid the warning?

Thanks,
Xu Yang