Re: [PATCH] pmdomain: core: connect consumer with supplier via device link

From: Ulf Hansson

Date: Sat Nov 14 2026 - 06:34:44 EST


On Thu, Nov 12, 2026 at 8:45 AM Yixun Lan <dlan@xxxxxxxxxx> wrote:
>
> Connect downstream Power Domain consumer with the supplier through
> the device link API, So when the pmdomain driver is unloaded or unbinded,
> the consumer can be automatically notified and purged gracefully.
>
> Signed-off-by: Yixun Lan <dlan@xxxxxxxxxx>

Can you check if the below commit that sits on my fixes branch solves
this problem for you?

e279833a5362 pmdomain: core: Wait for device link removals before
dropping genpd->dev

Kind regards
Uffe

> ---
> The power domain drivers under pmdomain subystem suffer fom
> the problem that unable to notify the downstream consumer with
> module unload or unlink event, left with stale sysfs interface.
> Curently, the solution is a) use builtin_platform_driver() API
> to disable the module unloading b) adding suppress_bind_attrs flag
> to disable the unlink interface, which prevent it from being
> triggered from user space.
>
> Take spacemit K3 platform [1] for example, without this patch,
> Removing spacemit_pm_domains module will trigger a failure to
> remove power domain, access pm_genpd_summary will result an
> kernel oops, See detail log below
>
> domain status children performance
> /device runtime status managed by
> ------------------------------------------------------------------------------
> lcd1 on 0
> cac88000.dp unsupported 0 SW
> lcd0 off-0 0
> audio on 0
> gpu on 0
> cac00000.gpu active 0 SW
> vpu off-0 0
>
> gpu lcd1 pm_genpd_summary
>
> (gpu and lcd1 isn't clean up)
>
> [ 181.353671] spacemit_pm_domains spacemit_ccu.k3-apmu-pmdomain.3: failed to remove domain gpu: -16 - state may be inconsistent
> [ 181.353889] PM: genpd_remove: unable to remove lcd1
> [ 181.353896] spacemit_pm_domains spacemit_ccu.k3-apmu-pmdomain.3: failed to remove domain lcd1: -16 - state may be inconsistent
>
> [ 206.228045] CPU: 5 UID: 0 PID: 2560 Comm: cat Not tainted 7.3.0-rc12 #81 PREEMPTLAZY
> [ 206.238002] Hardware name: SpacemiT K3 Pico-ITX (DT)
> [ 206.243027] epc : summary_show+0xaa/0x308
> [ 206.247081] ra : summary_show+0x264/0x308
> [ 206.251224] epc : ffffffff809f50fa ra : ffffffff809f52b4 sp : ffffffc601273a80
> [ 206.258542] gp : ffffffff82333b98 tp : ffffffd601d5ec80 t0 : ffffffff80026b38
> [ 206.265859] t1 : ffffffd6092d3cb0 t2 : ffffffff81601810 s0 : ffffffc601273b10
> [ 206.273177] s1 : ffffffd601b92470 a0 : ffffffd601b92588 a1 : 000000000000000a
> [ 206.280494] a2 : ffffffd609ae8000 a3 : 0000000000000000 a4 : ffffffd601d5ec80
> [ 206.287811] a5 : 0000000000000000 a6 : ffffffffffffe000 a7 : 0000000000000000
> [ 206.295128] s2 : ffffffd6092d3cb0 s3 : ffffffd601b92470 s4 : ffffffd601b94040
> [ 206.302447] s5 : 0000000000000000 s6 : 0000000000000003 s7 : ffffffff8221e3a8
> [ 206.309764] s8 : 0000000000000001 s9 : ffffffff8172b8c8 s10: 0000000000000004
> [ 206.317082] s11: 0000000000000000 t3 : 000000000000017a t4 : 0000000000000002
> [ 206.324399] t5 : 0000000000000000 t6 : ffffffd609ae7178 ssp : 0000000000000000
> [ 206.331804] status: 0000000200000120 badaddr: 0000000000000010 cause: 000000000000000d
> [ 206.339829] [<ffffffff809f50fa>] summary_show+0xaa/0x308
> [ 206.345206] [<ffffffff804b67f8>] seq_read_iter+0xe0/0x390
> [ 206.350671] [<ffffffff804b6b86>] seq_read+0xde/0x158
> [ 206.355697] [<ffffffff8063dee4>] full_proxy_read+0x54/0x88
> [ 206.361250] [<ffffffff8047a03e>] vfs_read+0xae/0x380
> [ 206.366275] [<ffffffff8047adc2>] ksys_read+0x6a/0xf0
> [ 206.371301] [<ffffffff8047ae68>] __riscv_sys_read+0x20/0x30
> [ 206.376944] [<ffffffff8101fa6a>] do_trap_ecall_u+0x19a/0x518
> [ 206.382674] [<ffffffff81033acc>] handle_exception+0x16c/0x178
> [ 206.388503] Code: 0a13 c00a 4c05 4b0d 3783 540a 3823 f804 3423 f804 (6b9c) 8552
> [ 206.396050] ---[ end trace 0000000000000000 ]---
> Segmentation fault cat /sys/kernel/debug/pm_genpd/pm_genpd_summary
>
> ====================================================================================
>
> With this patch applied, either unload the module or trigger an
> unbind operation, the downstream gpu, lcd1 will exit gracefully
>
> or
>
> domain status children performance
> /device runtime status managed by
> ------------------------------------------------------------------------------
>
> Link: https://patch.msgid.link/20261108-04-k3-pm-support-v3-0-b9dc0857039b@xxxxxxxxxx [1]
> ---
> drivers/pmdomain/core.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/pmdomain/core.c b/drivers/pmdomain/core.c
> index 964e44c2f78f..870fc8083bca 100644
> --- a/drivers/pmdomain/core.c
> +++ b/drivers/pmdomain/core.c
> @@ -3543,6 +3543,8 @@ static int __genpd_dev_pm_attach(struct device *dev, struct device *base_dev,
> return -EPROBE_DEFER;
> }
>
> + device_link_add(dev, &pd->dev, DL_FLAG_AUTOREMOVE_CONSUMER);
> +
> return 1;
>
> err:
>
> ---
> base-commit: 6628e7d4c6b25c8d18f1345e3a20d21829aae7e4
> change-id: 20261112-01-pm-device-link-9fdd0fe59a49
>
> Best regards,
> --
> Yixun Lan <dlan@xxxxxxxxxx>
>