Re: [PATCH v2 4/7] thunderbolt: add platform NHI driver

From: Chunfeng Yun (云春峰)

Date: Tue Nov 17 2026 - 07:14:54 EST


On 二, 2026-11-10 at 15:37 +0100, Mika Westerberg wrote:
> On Tue, Nov 10, 2026 at 07:48:52PM +0800, Chunfeng Yun wrote:
> >  
> > -void nhi_enable_int_throttling(struct tb_nhi *nhi);
> >  void nhi_disable_interrupts(struct tb_nhi *nhi);
> >  irqreturn_t tb_nhi_isr(int irq, void *data);
> >  irqreturn_t tb_ring_msix(int irq, void *data);
> > diff --git a/drivers/thunderbolt/pci.c b/drivers/thunderbolt/pci.c
> > index 9379c598a221..698f47ec63da 100644
> > --- a/drivers/thunderbolt/pci.c
> > +++ b/drivers/thunderbolt/pci.c
> > @@ -736,4 +736,5 @@ static struct pci_driver nhi_driver = {
> >   .shutdown = nhi_pci_shutdown,
> >   .driver.pm = &tb_nhi_pm_ops,
> >  };
> > +
>
> I guess this is unintented change?
>
will drop it

> >  module_pci_driver(nhi_driver);
> > diff --git a/drivers/thunderbolt/platform.c
> > b/drivers/thunderbolt/platform.c
> > new file mode 100644
> > index 000000000000..3562b6c7ea7d
> > --- /dev/null
> > +++ b/drivers/thunderbolt/platform.c
> > @@ -0,0 +1,172 @@
> > +// SPDX-License-Identifier: GPL-2.0-only
> > +/*
> > + * Thunderbolt driver - NHI driver (Platform bus specific
> > implementation)
> > + *
> > + * Author: Chunfeng Yun <chunfeng.yun@xxxxxxxxxxxx>
> > + */
> > +
> > +#include <linux/acpi.h>
> > +#include <linux/interrupt.h>
> > +#include <linux/errno.h>
> > +#include <linux/of.h>
> > +#include <linux/platform_device.h>
> > +#include <linux/pm_runtime.h>
> > +#include <linux/pm_wakeirq.h>
> > +#include <linux/slab.h>
> > +
> > +#include "nhi.h"
> > +#include "tb.h"
> > +
> > +static int nhi_platform_init(struct tb_nhi *nhi)
> > +{
> > + /* avoid device link fail for pcie tunnel */
>
> Do you have the ACPI _DSD there in place? 
Yes, we use the same uefi as WoA which pass Windows HLK and usb-if
certitfication

> Then I don't think you need
> these. And "pcie" -> "PCIe" everywhere (except in code).
ACK

>
> > + pm_runtime_set_active(nhi->dev);
> > + pm_runtime_get_noresume(nhi->dev);
> > + pm_runtime_enable(nhi->dev);
> > +
> > + device_set_wakeup_capable(nhi->dev, true);
> > + device_enable_async_suspend(nhi->dev);
>
> I also suggest not to enable all the fancy power management until you
> have
> the basic stuff like tunneling fully functional.
I'm not sure understand what you mean.

if you mean the function test:
I already test usb4net, usb3 tunnel, PCIe tunnel and dp tunnel data
transfers, and also runtime suspend/resume and system suspend/resume.

if you mean when to enable pm:
do you mean enable pm at the enable of tb_nhi_probe()?

+ pm_runtime_set_active(nhi->dev);
+ pm_runtime_get_noresume(nhi->dev);
+ pm_runtime_enable(nhi->dev);

these three line shall be put before nhi_select_cm(), otherwise, it
fail to link PCIe devices, and I will check it again.

+ device_set_wakeup_capable(nhi->dev, true);
+ device_enable_async_suspend(nhi->dev);

these two line can be put at the end of probe.

if you have any suggestions, please describe more details, I will try
it.

>
> > +
> > + return 0;
> > +}
> > +
> > +static void nhi_platform_clear_wake_irq(void *dev)
> > +{
> > + dev_pm_clear_wake_irq(dev);
> > +}
> > +
> > +static int nhi_platform_init_irq(struct tb_nhi *nhi)
> > +{
> > + struct device *dev = nhi->dev;
> > + int ret;
> > +
> > + ret = devm_request_irq(dev, nhi->irq, tb_nhi_isr,
> > IRQF_NO_SUSPEND,
> > +        dev_name(dev), nhi);
> > + if (ret)
> > + return dev_err_probe(dev, ret, "request irq
> > failed\n");
>
> failed to request interrupt
>
> Please be consistent with the rest of the driver.
ACK
>
> > +
> > + if (nhi->wake_irq) {
> > + ret = dev_pm_set_dedicated_wake_irq(dev, nhi-
> > >wake_irq);
> > + if (ret) {
> > + return dev_err_probe(dev, ret, "set wake
> > irq %d failed\n", nhi->wake_irq);
>
> Split this into 2 lines.
ACK
>
> > + }
> > +
> > + dev_info(dev, "wake irq %d\n", nhi->wake_irq);
>
> Remove this. The driver should be silent when it works. You can add
> debug
> but I don't think this is useful.
ACK
>
> > +
> > + ret = devm_add_action_or_reset(dev,
> > nhi_platform_clear_wake_irq, dev);
>
> Can you pass dev_pm_clear_wake_irq directly here?

the parameter type do not match which cause build warning

kernel/mediatek/drivers/thunderbolt/platform.c:8:
kernel/mediatek/drivers/thunderbolt/platform.c: In function
'nhi_platform_init_irq':
kernel/mediatek/drivers/thunderbolt/platform.c:53:53: error: passing
argument 2 of '__devm_add_action_or_reset' from incompatible pointer
type [-Werror=incompatible-pointer-types]
53 | ret = devm_add_action_or_reset(dev,
dev_pm_clear_wake_irq, dev);
|
^~~~~~~~~~~~~~~~~~~~~
| |
| void
(*)(struct device *)
upstream/k7/kernel/mediatek/include/linux/device/devres.h:283:41: note:
in definition of macro 'devm_add_action_or_reset'
283 | __devm_add_action_or_reset(dev, action, data, #action)
| ^~~~~~
upstream/k7/kernel/mediatek/include/linux/device/devres.h:271:73: note:
expected 'void (*)(void *)' but argument is of type 'void (*)(struct
device *)'
271 | static inline int __devm_add_action_or_reset(struct device
*dev, void (*action)(void *),

>
> > + }
> > +
> > + return ret;
> > +}
> > +
> > +static void nhi_platform_exit(struct tb_nhi *nhi)
> > +{
> > + devm_free_irq(nhi->dev, nhi->irq, nhi);
>
> Isn't it released automatically?
refer to nhi_pci_exit() which give the reason that avoid queuing the
interrupt_work again, I will add a common in next version.

>
> > + flush_work(&nhi->interrupt_work);
> > +
> > + pm_runtime_put_noidle(nhi->dev);
> > + pm_runtime_disable(nhi->dev);
> > + pm_runtime_set_suspended(nhi->dev);
> > +}
> > +
> > +static const struct tb_nhi_ops nhi_platform_default_ops = {
> > + .init = nhi_platform_init,
> > + .shutdown = nhi_platform_exit,
> > + .init_interrupts = nhi_platform_init_irq,
> > +};
> > +
> > +static int nhi_platform_probe(struct platform_device *pdev)
> > +{
> > + struct device *dev = &pdev->dev;
> > + struct tb_nhi *nhi;
> > +
> > + nhi = devm_kzalloc(dev, sizeof(*nhi), GFP_KERNEL);
>
> devm_kzalloc_obj()
ACK
>
> > + if (!nhi)
> > + return -ENOMEM;
> > +
> > + nhi->dev = dev;
> > + nhi->ops = &nhi_platform_default_ops;
> > +
> > + nhi->iobase = devm_platform_ioremap_resource(pdev, 0);
> > + if (IS_ERR(nhi->iobase))
> > + return PTR_ERR(nhi->iobase);
> > +
> > + nhi->irq = platform_get_irq(pdev, 0);
> > + if (nhi->irq < 0)
> > + return nhi->irq;
> > +
> > + /* ignore error */
>
> This should explain why it's okay to ignore the error.
ACK, it is implementation specific, so make it optional

>
> > + nhi->wake_irq = platform_get_irq_optional(pdev, 1);
> > + if (nhi->wake_irq == -EPROBE_DEFER)
> > + return nhi->wake_irq;
> > + else if (nhi->wake_irq < 0)
> > + nhi->wake_irq = 0;
> > +
> > + return tb_nhi_probe(nhi);
> > +}
> > +
> > +static void nhi_platform_do_remove(struct platform_device *pdev,
> > bool reset)
> > +{
> > + struct device *dev = &pdev->dev;
> > + struct tb *tb = platform_get_drvdata(pdev);
> > + struct tb_nhi *nhi = tb->nhi;
>
> Reverse-christmas tree if possible:
ACK
>
> struct tb *tb = platform_get_drvdata(pdev);
> struct device *dev = &pdev->dev;
> struct tb_nhi *nhi = tb->nhi;
>
> > +
> > + /*
> > + * On system shutdown/reboot force a host router reset so
> > the
> > + * CM asserts DPR on connected tbt3 devices before the
> > router tree
>
> Thunderbolt 3
ACK
>
> > + * is removed (see tb_stop()).
> > + */
> > + if (reset)
> > + nhi->host_reset = true;
> > +
> > + pm_runtime_get_sync(dev);
> > + pm_runtime_dont_use_autosuspend(dev);
> > +
> > + tb_domain_remove(tb);
> > + wait_for_completion(&nhi->domain_released);
> > + tb_nhi_shutdown(nhi);
> > +}
> > +
> > +static void nhi_platform_remove(struct platform_device *pdev)
> > +{
> > + nhi_platform_do_remove(pdev, false);
> > +}
> > +
> > +static void nhi_platform_shutdown(struct platform_device *pdev)
> > +{
> > + nhi_platform_do_remove(pdev, true);
> > +}
> > +
> > +
> > +#ifdef CONFIG_OF
> > +static const struct of_device_id nhi_platform_of_match[] = {
> > + { .compatible = "usb4-nhi-platform" },
> > + { }
> > +};
>
> I think this was agreed to be dropped for now.

will drop it

Thanks
>
> > +MODULE_DEVICE_TABLE(of, nhi_platform_of_match);
> > +#endif
> > +
> > +#ifdef CONFIG_ACPI
> > +static const struct acpi_device_id nhi_platform_acpi_match[] = {
> > + { "ACPI0015", 0 },
> > + { }
> > +};
> > +MODULE_DEVICE_TABLE(acpi, nhi_platform_acpi_match);
> > +#endif
> > +
> > +static struct platform_driver nhi_platform_driver = {
> > + .probe = nhi_platform_probe,
> > + .remove = nhi_platform_remove,
> > + .shutdown = nhi_platform_shutdown,
> > + .driver = {
> > + .name = "thunderbolt",
> > + .of_match_table =
> > of_match_ptr(nhi_platform_of_match),
> > + .acpi_match_table =
> > ACPI_PTR(nhi_platform_acpi_match),
> > + .pm = &tb_nhi_pm_ops,
> > + },
> > +};
> > +
> > +module_platform_driver(nhi_platform_driver);
> > +MODULE_ALIAS("platform:thunderbolt");
> > +MODULE_DESCRIPTION("Thunderbolt/USB4 platform host interface
> > driver");
> > +MODULE_LICENSE("GPL v2");
> > diff --git a/include/linux/thunderbolt.h
> > b/include/linux/thunderbolt.h
> > index 57502da29080..72497acac254 100644
> > --- a/include/linux/thunderbolt.h
> > +++ b/include/linux/thunderbolt.h
> > @@ -511,6 +511,8 @@ void tb_service_properties_changed(struct
> > tb_service *svc);
> >   * @going_away: The host controller device is about to disappear
> > so when
> >   * this flag is set, avoid touching the hardware
> > anymore.
> >   * @iommu_dma_protection: An IOMMU will isolate external-facing
> > ports.
> > + * @irq: interrupt number of NHI.
> > + * @wake_irq: optional dedicated wakeup interrupt.
> >   * @interrupt_work: Work scheduled to handle ring interrupt when
> > no
> >   *     MSI-X is used.
> >   * @hop_count: Number of rings (end point hops) supported by NHI.
> > @@ -532,6 +534,8 @@ struct tb_nhi {
> >   u32 *interrupt_mask;
> >   bool going_away;
> >   bool iommu_dma_protection;
> > + int irq;
> > + int wake_irq;
> >   struct work_struct interrupt_work;
> >   u32 hop_count;
> >   unsigned long quirks;
> > --
> > 2.45.2