[PATCH] firmware/psci: add .begin implementation for psci_suspend_ops

[金塔尖]

xiongxin <xiongxin at kylinos.cn> 于2023年2月10日周五 16:58写道：
>
> nvme_suspend() in the nvme pci driver will call
> pm_suspend_via_firmware() to determine whether to use the suspend to
> idle mode or other suspend, the code show as below:
>
> static int nvme_suspend(struct device *dev)
> {
>         ...
>         if (pm_suspend_via_firmware() || !ctrl->npss ||
>             !pcie_aspm_enabled(pdev) ||
>             (ndev->ctrl.quirks & NVME_QUIRK_SIMPLE_SUSPEND))
>                 return nvme_disable_prepare_reset(ndev, true);
>
>         ... /* suspend to idle mode */
> }
>
> The current ARM64 suspend process does not call the
> pm_set_suspend_via_firmware() function, so when PCIe ASPM is enabled,
> the nvme suspend process under the ARM64 platform will enter the suspend
> to idle mode (the irq free is not called), which will cause the nvme
> driver to report the following irq warning during the resume process:
>
> [  189.233188] ------------[ cut here ]------------
> [  189.237791] Unbalanced enable for IRQ 52
> [  189.241705] WARNING: CPU: 0 PID: 2086 at kernel/irq/manage.c:781 __enable_irq+0x54/0x88
> [  189.249700] Modules linked in: nls_iso8859_1 dm_multipath scsi_dh_rdac scsi_dh_emc scsi_dh_alua snd_hda_codec_hdmi snd_hda_intel snd_intel_dspcfg snd_hda_codec snd_hda_core snd_hwdep snd_pcm snd_seq_midi snd_seq_midi_event snd_rawmidi snd_seq snd_seq_device snd_timer snd joydev input_leds efi_pstore soundcore sch_fq_codel ppdev lp parport ip_tables x_tables autofs4 btrfs blake2b_generic raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor xor_neon raid6_pq libcrc32c raid1 raid0 multipath linear amdgpu gpu_sched drm_buddy radeon video drm_ttm_helper ttm i2c_algo_bit drm_display_helper cec drm_kms_helper syscopyarea sysfillrect sysimgblt hid_generic drm nvme crct10dif_ce ghash_ce usbhid sha2_ce sha256_arm64 hid nvme_core sha1_ce ahci libahci aes_neon_bs aes_neon_blk aes_ce_blk aes_ce_cipher
> [  189.321740] CPU: 0 PID: 2086 Comm: systemd-sleep Not tainted 6.2.0-rc7+ #9
> [  189.328601] Hardware name: Greatwall GW-XXXXXX-XXX/GW-XXXXXX-XXX, BIOS KunLun BIOS V4.0 01/19/2021
> [  189.337543] pstate: 604000c5 (nZCv daIF +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> [  189.344490] pc : __enable_irq+0x54/0x88
> [  189.348313] lr : __enable_irq+0x54/0x88
> [  189.352136] sp : ffff80000d58bb00
> [  189.355436] x29: ffff80000d58bb00 x28: ffff80000a355a54 x27: ffff001f809910dc
> [  189.362559] x26: 0000000000000000 x25: 0000000000000000 x24: 0000000000000000
> [  189.369681] x23: 0000000000000001 x22: 0000000000000000 x21: ffff80000a0d2860
> [  189.376803] x20: 0000000000000034 x19: ffff001f80991000 x18: 0000000000000010
> [  189.383924] x17: 0000000000000000 x16: 0000000000000400 x15: ffff001f87612578
> [  189.391047] x14: 0000000000000000 x13: 3235205152492072 x12: 6f6620656c62616e
> [  189.398168] x11: 00000000ffffdfff x10: 00000000ffffdfff x9 : ffff80000823788c
> [  189.405291] x8 : 000000000002ffe8 x7 : c0000000ffffdfff x6 : 0000000000000001
> [  189.412412] x5 : ffff0020f6f0cb08 x4 : 0000000000000000 x3 : 0000000000000027
> [  189.419534] x2 : ffff0020f6f0cb10 x1 : 0000000000000000 x0 : 0000000000000000
> [  189.426656] Call trace:
> [  189.429089]  __enable_irq+0x54/0x88
> [  189.432566]  resume_irqs+0x118/0x210
> [  189.436129]  resume_device_irqs+0x1c/0x48
> [  189.440125]  dpm_resume_noirq+0x1c/0x30
> [  189.443951]  suspend_devices_and_enter+0x218/0x8c8
> [  189.448731]  pm_suspend+0x31c/0x420
> [  189.452207]  state_store+0x94/0x118
> [  189.455684]  kobj_attr_store+0x18/0x30
> [  189.459421]  sysfs_kf_write+0x4c/0x60
> [  189.463073]  kernfs_fop_write_iter+0x130/0x1c0
> [  189.467504]  vfs_write+0x284/0x358
> [  189.470893]  ksys_write+0x74/0x108
> [  189.474282]  __arm64_sys_write+0x24/0x30
> [  189.478191]  invoke_syscall+0x4c/0x110
> [  189.481930]  el0_svc_common.constprop.0+0x58/0x190
> [  189.486708]  do_el0_svc+0x40/0xb8
> [  189.490011]  el0_svc+0x2c/0xb8
> [  189.493053]  el0t_64_sync_handler+0xb8/0xc0
> [  189.497223]  el0t_64_sync+0x1a4/0x1a8
> [  189.500872] ---[ end trace 0000000000000000 ]---
> ...
>
> Implement pm_set_suspend_via_firmware() in PSCI psci_suspend_ops.begin()
> for ARM64, so that when doing the suspend process, it will directly
> enter the nvme_disable_prepare_reset() process.
>
> Signed-off-by: xiongxin <xiongxin at kylinos.cn>
> Signed-off-by: Tian Yaxiong <tianyaxiong at kylinos.cn>
> Signed-off-by: Wei Yufeng <weiyufeng at kylinos.cn>
>
> diff --git a/drivers/firmware/psci/psci.c b/drivers/firmware/psci/psci.c
> index 447ee4ea5c90..90d6102d45d7 100644
> --- a/drivers/firmware/psci/psci.c
> +++ b/drivers/firmware/psci/psci.c
> @@ -484,6 +484,13 @@ static int psci_system_suspend(unsigned long unused)
>                               pa_cpu_resume, 0, 0);
>  }
>
> +static int psci_system_suspend_begin(suspend_state_t state)
> +{
> +       pm_set_suspend_via_firmware();
> +
> +       return 0;
> +}
> +
>  static int psci_system_suspend_enter(suspend_state_t state)
>  {
>         return cpu_suspend(0, psci_system_suspend);
> @@ -491,6 +498,7 @@ static int psci_system_suspend_enter(suspend_state_t state)
>
>  static const struct platform_suspend_ops psci_suspend_ops = {
>         .valid          = suspend_valid_only_mem,
> +       .begin          = psci_system_suspend_begin,
>         .enter          = psci_system_suspend_enter,
>  };
>
> --
> 2.34.1
>

This is a bug of nvme under the arm64 architecture, please help to
confirm whether the patch is reasonable;