In the Linux kernel, the following vulnerability has been resolved:

pmdomain: imx: gpcv2: Adjust delay after power up handshake

The udelay(5) is not enough, sometimes below kernel panic
still be triggered:

[    4.012973] Kernel panic - not syncing: Asynchronous SError Interrupt
[    4.012976] CPU: 2 UID: 0 PID: 186 Comm: (udev-worker) Not tainted 6.12.0-rc2-0.0.0-devel-00004-g8b1b79e88956 #1
[    4.012982] Hardware name: Toradex Verdin iMX8M Plus WB on Dahlia Board (DT)
[    4.012985] Call trace:
[...]
[    4.013029]  arm64_serror_panic+0x64/0x70
[    4.013034]  do_serror+0x3c/0x70
[    4.013039]  el1h_64_error_handler+0x30/0x54
[    4.013046]  el1h_64_error+0x64/0x68
[    4.013050]  clk_imx8mp_audiomix_runtime_resume+0x38/0x48
[    4.013059]  __genpd_runtime_resume+0x30/0x80
[    4.013066]  genpd_runtime_resume+0x114/0x29c
[    4.013073]  __rpm_callback+0x48/0x1e0
[    4.013079]  rpm_callback+0x68/0x80
[    4.013084]  rpm_resume+0x3bc/0x6a0
[    4.013089]  __pm_runtime_resume+0x50/0x9c
[    4.013095]  pm_runtime_get_suppliers+0x60/0x8c
[    4.013101]  __driver_probe_device+0x4c/0x14c
[    4.013108]  driver_probe_device+0x3c/0x120
[    4.013114]  __driver_attach+0xc4/0x200
[    4.013119]  bus_for_each_dev+0x7c/0xe0
[    4.013125]  driver_attach+0x24/0x30
[    4.013130]  bus_add_driver+0x110/0x240
[    4.013135]  driver_register+0x68/0x124
[    4.013142]  __platform_driver_register+0x24/0x30
[    4.013149]  sdma_driver_init+0x20/0x1000 [imx_sdma]
[    4.013163]  do_one_initcall+0x60/0x1e0
[    4.013168]  do_init_module+0x5c/0x21c
[    4.013175]  load_module+0x1a98/0x205c
[    4.013181]  init_module_from_file+0x88/0xd4
[    4.013187]  __arm64_sys_finit_module+0x258/0x350
[    4.013194]  invoke_syscall.constprop.0+0x50/0xe0
[    4.013202]  do_el0_svc+0xa8/0xe0
[    4.013208]  el0_svc+0x3c/0x140
[    4.013215]  el0t_64_sync_handler+0x120/0x12c
[    4.013222]  el0t_64_sync+0x190/0x194
[    4.013228] SMP: stopping secondary CPUs

The correct way is to wait handshake, but it needs BUS clock of
BLK-CTL be enabled, which is in separate driver. So delay is the
only option here. The udelay(10) is a data got by experiment.