[PATCH] usb: dwc3: qcom: Add support to skip phy management by USB core

[PATCH] usb: dwc3: qcom: Add support to skip phy management by USB core

[Krishna Kurapati]

DWC3 driver does manage phys itself sufficiently for Qualcomm platforms.
If xhci-skip-phy-init is not set, the HCD core does a phy_init and phy_
exit is done only when roothub is being destroyed. Due to this, during
system suspend in host mode, although phy_exit is done by DWC3 core, the
init_count on phy is never down to zero since HCD core also did an init.
consequently causing phy's exit routine to not be called. Hence, add
support to skip phy management by USB core.

Some Exynos platforms still do rely on USB core for phy_calibrate calls,
hence disable USB core management for Qualcomm platforms only.

Signed-off-by: Krishna Kurapati <krishna.kurapati@oss.qualcomm.com>
---
Link to PHY calibrate handling for Exynos platforms:
https://lore.kernel.org/all/20190829053028.32438-1-m.szyprowski@samsung.com/
---
 drivers/usb/dwc3/dwc3-qcom.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/drivers/usb/dwc3/dwc3-qcom.c b/drivers/usb/dwc3/dwc3-qcom.c
index ac68b4218b56..ba26e61df915 100644
--- a/drivers/usb/dwc3/dwc3-qcom.c
+++ b/drivers/usb/dwc3/dwc3-qcom.c
@@ -19,6 +19,7 @@
 #include <linux/iopoll.h>
 #include <linux/usb/hcd.h>
 #include <linux/usb.h>
+#include <linux/property.h>
 #include "core.h"
 #include "glue.h"
 
@@ -419,6 +420,16 @@ static irqreturn_t qcom_dwc3_resume_irq(int irq, void *data)
 	return IRQ_HANDLED;
 }
 
+static int dwc3_qcom_set_swnode(struct device *dev)
+{
+	const struct property_entry props[] = {
+		PROPERTY_ENTRY_BOOL("xhci-skip-phy-init-quirk"),
+		{}
+	};
+
+	return device_create_managed_software_node(dev, props, NULL);
+}
+
 static void dwc3_qcom_select_utmi_clk(struct dwc3_qcom *qcom)
 {
 	/* Configure dwc3 to use UTMI clock as PIPE clock not present */
@@ -683,6 +694,10 @@ static int dwc3_qcom_probe(struct platform_device *pdev)
 	if (ignore_pipe_clk)
 		dwc3_qcom_select_utmi_clk(qcom);
 
+	ret = dwc3_qcom_set_swnode(dev);
+	if (ret)
+		goto clk_disable;
+
 	qcom->mode = usb_get_dr_mode(dev);
 
 	if (qcom->mode == USB_DR_MODE_HOST) {

---
base-commit: 80574c40598aedbc1751c528e414d7e224bc6313
change-id: 20260720-dwc3-skip-init-quirk-c110eae068ac

Best regards,
--  
Krishna Kurapati <krishna.kurapati@oss.qualcomm.com>

Re: [PATCH] usb: dwc3: qcom: Add support to skip phy management by USB core

[Thinh Nguyen]

On Thu, Jul 23, 2026, Krishna Kurapati wrote:
> DWC3 driver does manage phys itself sufficiently for Qualcomm platforms.
> If xhci-skip-phy-init is not set, the HCD core does a phy_init and phy_
> exit is done only when roothub is being destroyed. Due to this, during
> system suspend in host mode, although phy_exit is done by DWC3 core, the
> init_count on phy is never down to zero since HCD core also did an init.
> consequently causing phy's exit routine to not be called. Hence, add
> support to skip phy management by USB core.
> 
> Some Exynos platforms still do rely on USB core for phy_calibrate calls,
> hence disable USB core management for Qualcomm platforms only.

So this is a fix applicable to all qcom platforms right? Do we need to
add Fixes tag and backport this change?

Regardless,

Acked-by: Thinh Nguyen <Thinh.Nguyen@synopsys.com>

Thanks,
Thinh

> 
> Signed-off-by: Krishna Kurapati <krishna.kurapati@oss.qualcomm.com>
> ---
> Link to PHY calibrate handling for Exynos platforms:
> https://urldefense.com/v3/__https://lore.kernel.org/all/20190829053028.32438-1-m.szyprowski@samsung.com/__;!!A4F2R9G_pg!aP037GeN4WssfY2NPRwwFsPNYLtng9TCmzBFu8aoGmYWzLzD7vxM-Mt1sOwfv6KUBe7GsvyRbuktzot9dM8jEcYTWSdnmhMkjHJ3rQ$ 
> ---
>  drivers/usb/dwc3/dwc3-qcom.c | 15 +++++++++++++++
>  1 file changed, 15 insertions(+)
> 
> diff --git a/drivers/usb/dwc3/dwc3-qcom.c b/drivers/usb/dwc3/dwc3-qcom.c
> index ac68b4218b56..ba26e61df915 100644
> --- a/drivers/usb/dwc3/dwc3-qcom.c
> +++ b/drivers/usb/dwc3/dwc3-qcom.c
> @@ -19,6 +19,7 @@
>  #include <linux/iopoll.h>
>  #include <linux/usb/hcd.h>
>  #include <linux/usb.h>
> +#include <linux/property.h>
>  #include "core.h"
>  #include "glue.h"
>  
> @@ -419,6 +420,16 @@ static irqreturn_t qcom_dwc3_resume_irq(int irq, void *data)
>  	return IRQ_HANDLED;
>  }
>  
> +static int dwc3_qcom_set_swnode(struct device *dev)
> +{
> +	const struct property_entry props[] = {
> +		PROPERTY_ENTRY_BOOL("xhci-skip-phy-init-quirk"),
> +		{}
> +	};
> +
> +	return device_create_managed_software_node(dev, props, NULL);
> +}
> +
>  static void dwc3_qcom_select_utmi_clk(struct dwc3_qcom *qcom)
>  {
>  	/* Configure dwc3 to use UTMI clock as PIPE clock not present */
> @@ -683,6 +694,10 @@ static int dwc3_qcom_probe(struct platform_device *pdev)
>  	if (ignore_pipe_clk)
>  		dwc3_qcom_select_utmi_clk(qcom);
>  
> +	ret = dwc3_qcom_set_swnode(dev);
> +	if (ret)
> +		goto clk_disable;
> +
>  	qcom->mode = usb_get_dr_mode(dev);
>  
>  	if (qcom->mode == USB_DR_MODE_HOST) {
> 
> ---
> base-commit: 80574c40598aedbc1751c528e414d7e224bc6313
> change-id: 20260720-dwc3-skip-init-quirk-c110eae068ac
> 
> Best regards,
> --  
> Krishna Kurapati <krishna.kurapati@oss.qualcomm.com>
> 

Re: [PATCH] usb: dwc3: qcom: Add support to skip phy management by USB core

[Val Packett]

On 7/23/26 7:54 AM, Krishna Kurapati wrote:
> DWC3 driver does manage phys itself sufficiently for Qualcomm platforms.
> If xhci-skip-phy-init is not set, the HCD core does a phy_init and phy_
> exit is done only when roothub is being destroyed. Due to this, during
> system suspend in host mode, although phy_exit is done by DWC3 core, the
> init_count on phy is never down to zero since HCD core also did an init.
> consequently causing phy's exit routine to not be called. Hence, add
> support to skip phy management by USB core.
>
> Some Exynos platforms still do rely on USB core for phy_calibrate calls,
> hence disable USB core management for Qualcomm platforms only.
>
> Signed-off-by: Krishna Kurapati <krishna.kurapati@oss.qualcomm.com>
> ---
> Link to PHY calibrate handling for Exynos platforms:
> https://lore.kernel.org/all/20190829053028.32438-1-m.szyprowski@samsung.com/

Hi,

this was merged into -next a few days ago, and I'm seeing warnings on 
boot now:

[    0.613188] sysfs: cannot create duplicate filename 
'/devices/platform/soc@0/a800000.usb/software_node'
[    0.613195] CPU: 10 UID: 0 PID: 93 Comm: kworker/u49:0 Tainted: G    
     W           7.3.0-rc2-next-20260907-uwu+ #143 PREEMPT(full)
[    0.613197] Tainted: [W]=WARN
[    0.613198] Hardware name: Dell Inc. Latitude 7455/0FK7MX, BIOS 
2.13.0 09/19/2025
[    0.613199] Workqueue: events_unbound deferred_probe_work_func
[    0.613206] Call trace:
[..]
[    0.613226]  device_create_managed_software_node+0xe8/0x108
[    0.613227]  dwc3_qcom_probe+0x5a4/0x7b0

There is indeed already a software node being created in the shared dwc3 
core code.

Could there be any other consequences than a duplicate sysfs path?

Maybe the property should instead be added to all qcom device trees?

Thanks,
~val

Re: [PATCH] usb: dwc3: qcom: Add support to skip phy management by USB core

[Val Packett]

On 9/8/26 7:33 PM, Val Packett wrote:
>
> On 7/23/26 7:54 AM, Krishna Kurapati wrote:
>> DWC3 driver does manage phys itself sufficiently for Qualcomm platforms.
>> If xhci-skip-phy-init is not set, the HCD core does a phy_init and phy_
>> exit is done only when roothub is being destroyed. Due to this, during
>> system suspend in host mode, although phy_exit is done by DWC3 core, the
>> init_count on phy is never down to zero since HCD core also did an init.
>> consequently causing phy's exit routine to not be called. Hence, add
>> support to skip phy management by USB core.
>>
>> Some Exynos platforms still do rely on USB core for phy_calibrate calls,
>> hence disable USB core management for Qualcomm platforms only.
>>
>> Signed-off-by: Krishna Kurapati <krishna.kurapati@oss.qualcomm.com>
>> ---
>> Link to PHY calibrate handling for Exynos platforms:
>> https://lore.kernel.org/all/20190829053028.32438-1-m.szyprowski@samsung.com/ 
>>
>
> Hi,
>
> this was merged into -next a few days ago, and I'm seeing warnings on 
> boot now:
>
> [    0.613188] sysfs: cannot create duplicate filename 
> '/devices/platform/soc@0/a800000.usb/software_node'
> [    0.613195] CPU: 10 UID: 0 PID: 93 Comm: kworker/u49:0 Tainted: G  
>       W           7.3.0-rc2-next-20260907-uwu+ #143 PREEMPT(full)
> [    0.613197] Tainted: [W]=WARN
> [    0.613198] Hardware name: Dell Inc. Latitude 7455/0FK7MX, BIOS 
> 2.13.0 09/19/2025
> [    0.613199] Workqueue: events_unbound deferred_probe_work_func
> [    0.613206] Call trace:
> [..]
> [    0.613226]  device_create_managed_software_node+0xe8/0x108
> [    0.613227]  dwc3_qcom_probe+0x5a4/0x7b0
>
> There is indeed already a software node being created in the shared 
> dwc3 core code.
>
> Could there be any other consequences than a duplicate sysfs path?
>
Answering my own question: even worse messages when the controller 
switches to gadget mode (which happens even with chargers when DRD is 
enabled)

[    5.620951] refcount_t: underflow; use-after-free.
[    5.620957] WARNING: lib/refcount.c:28 at 
refcount_warn_saturate+0xd8/0x140, CPU#2: kworker/2:1/98
[    5.621075] CPU: 2 UID: 0 PID: 98 Comm: kworker/2:1 Tainted: G       
W           7.3.0-rc2-next-20260907-uwu+ #143 PREEMPT(full)
[    5.621077] Tainted: [W]=WARN
[    5.621078] Hardware name: Dell Inc. Latitude 7455/0FK7MX, BIOS 
2.13.0 09/19/2025
[    5.621079] Workqueue: events_freezable __dwc3_set_mode
[..]
[    5.621114] Call trace:
[    5.621115]  refcount_warn_saturate+0xd8/0x140 (P)
[    5.621117]  kobject_put+0x11c/0x230
[    5.621121]  software_node_notify_remove+0xdc/0xf8
[    5.621124]  device_del+0x1dc/0x328
[    5.621126]  usb_disconnect+0x1d8/0x348
[    5.621129]  usb_remove_hcd+0x100/0x2a8
[    5.621131]  xhci_plat_remove+0x8c/0x170
[    5.621134]  platform_remove+0x28/0x40
[    5.621135]  device_release_driver_internal+0x174/0x290
[    5.621138]  device_release_driver+0x20/0x38
[    5.621140]  bus_remove_device+0x19c/0x1f8
[    5.621142]  device_del+0x1c4/0x328
[    5.621143]  platform_device_unregister+0x34/0xc0
[    5.621145]  dwc3_host_exit+0x50/0x70
[    5.621146]  __dwc3_set_mode+0xa4/0x378


[..]


[    5.644535] Unable to handle kernel paging request at virtual address 
006563697665642f
[    5.644537] Mem abort info:
[    5.644538]   ESR = 0x0000000096000004
[    5.644539]   EC = 0x25: DABT (current EL), IL = 32 bits
[    5.644540]   SET = 0, FnV = 0
[    5.644541]   EA = 0, S1PTW = 0
[    5.644542]   FSC = 0x04: level 0 translation fault
[    5.644543] Data abort info:
[    5.644543]   ISV = 0, ISS = 0x00000004, ISS2 = 0x00000000
[    5.644545]   CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[    5.644546]   GCS = 0, Overlay = 0, DirtyBit = 0
[    5.644546] [006563697665642f] address between user and kernel 
address ranges
[    5.644548] Internal error: Oops: 0000000096000004 [#1]  SMP
[    5.644653] CPU: 2 UID: 0 PID: 98 Comm: kworker/2:1 Tainted: G       
W           7.3.0-rc2-next-20260907-uwu+ #143 PREEMPT(full)
[    5.644655] Tainted: [W]=WARN
[    5.644656] Hardware name: Dell Inc. Latitude 7455/0FK7MX, BIOS 
2.13.0 09/19/2025
[    5.644657] Workqueue: events_freezable __dwc3_set_mode
[    5.644664] pc : __pi_strcmp+0x9c/0x140
[    5.644668] lr : software_node_property_present+0x60/0x98
[    5.644697] Call trace:
[    5.644698]  __pi_strcmp+0x9c/0x140 (P)
[    5.644700]  device_property_present+0x9c/0xc0
[    5.644703]  dwc3_gadget_init+0x230/0x7b0
[    5.644704]  __dwc3_set_mode+0x314/0x378
[    5.644707]  process_scheduled_works+0x1b8/0x538
[    5.644710]  worker_thread+0x1fc/0x2f8
[    5.644711]  kthread+0x114/0x148
[    5.644713]  ret_from_fork+0x10/0x20


> Thanks,
> ~val
>

Re: [PATCH] usb: dwc3: qcom: Add support to skip phy management by USB core

[Thinh Nguyen]

On Tue, Sep 08, 2026, Val Packett wrote:
> 
> On 9/8/26 7:33 PM, Val Packett wrote:
> > 
> > On 7/23/26 7:54 AM, Krishna Kurapati wrote:
> > > DWC3 driver does manage phys itself sufficiently for Qualcomm platforms.
> > > If xhci-skip-phy-init is not set, the HCD core does a phy_init and phy_
> > > exit is done only when roothub is being destroyed. Due to this, during
> > > system suspend in host mode, although phy_exit is done by DWC3 core, the
> > > init_count on phy is never down to zero since HCD core also did an init.
> > > consequently causing phy's exit routine to not be called. Hence, add
> > > support to skip phy management by USB core.
> > > 
> > > Some Exynos platforms still do rely on USB core for phy_calibrate calls,
> > > hence disable USB core management for Qualcomm platforms only.
> > > 
> > > Signed-off-by: Krishna Kurapati <krishna.kurapati@oss.qualcomm.com>
> > > ---
> > > Link to PHY calibrate handling for Exynos platforms:
> > > https://urldefense.com/v3/__https://lore.kernel.org/all/20190829053028.32438-1-m.szyprowski@samsung.com/__;!!A4F2R9G_pg!aoY0pB27Tlxo3u1RzNzwPHpGsjuAgJWtPeKMvDrL___6gEQb28ew8Rg-O1877xBHB92Zv2fY3_Zbb2aoEQY$
> > > 
> > 
> > Hi,
> > 
> > this was merged into -next a few days ago, and I'm seeing warnings on
> > boot now:
> > 
> > [    0.613188] sysfs: cannot create duplicate filename
> > '/devices/platform/soc@0/a800000.usb/software_node'
> > [    0.613195] CPU: 10 UID: 0 PID: 93 Comm: kworker/u49:0 Tainted: G   
> >     W           7.3.0-rc2-next-20260907-uwu+ #143 PREEMPT(full)
> > [    0.613197] Tainted: [W]=WARN
> > [    0.613198] Hardware name: Dell Inc. Latitude 7455/0FK7MX, BIOS
> > 2.13.0 09/19/2025
> > [    0.613199] Workqueue: events_unbound deferred_probe_work_func
> > [    0.613206] Call trace:
> > [..]
> > [    0.613226]  device_create_managed_software_node+0xe8/0x108
> > [    0.613227]  dwc3_qcom_probe+0x5a4/0x7b0
> > 
> > There is indeed already a software node being created in the shared dwc3
> > core code.
> > 
> > Could there be any other consequences than a duplicate sysfs path?
> > 
> Answering my own question: even worse messages when the controller switches
> to gadget mode (which happens even with chargers when DRD is enabled)
> 


Thanks for the report Val.

Hi Krishna,

Can you help send a revert patch until a proper change is in available?

Thanks,
Thinh

Re: [PATCH] usb: dwc3: qcom: Add support to skip phy management by USB core

[Krishna Kurapati]

On 9/9/2026 7:16 AM, Thinh Nguyen wrote:
> On Tue, Sep 08, 2026, Val Packett wrote:
>>
>> On 9/8/26 7:33 PM, Val Packett wrote:
>>>
>>> On 7/23/26 7:54 AM, Krishna Kurapati wrote:
>>>> DWC3 driver does manage phys itself sufficiently for Qualcomm platforms.
>>>> If xhci-skip-phy-init is not set, the HCD core does a phy_init and phy_
>>>> exit is done only when roothub is being destroyed. Due to this, during
>>>> system suspend in host mode, although phy_exit is done by DWC3 core, the
>>>> init_count on phy is never down to zero since HCD core also did an init.
>>>> consequently causing phy's exit routine to not be called. Hence, add
>>>> support to skip phy management by USB core.
>>>>
>>>> Some Exynos platforms still do rely on USB core for phy_calibrate calls,
>>>> hence disable USB core management for Qualcomm platforms only.
>>>>
>>>> Signed-off-by: Krishna Kurapati <krishna.kurapati@oss.qualcomm.com>
>>>> ---
>>>> Link to PHY calibrate handling for Exynos platforms:
>>>> https://urldefense.com/v3/__https://lore.kernel.org/all/20190829053028.32438-1-m.szyprowski@samsung.com/__;!!A4F2R9G_pg!aoY0pB27Tlxo3u1RzNzwPHpGsjuAgJWtPeKMvDrL___6gEQb28ew8Rg-O1877xBHB92Zv2fY3_Zbb2aoEQY$
>>>>
>>>
>>> Hi,
>>>
>>> this was merged into -next a few days ago, and I'm seeing warnings on
>>> boot now:
>>>
>>> [    0.613188] sysfs: cannot create duplicate filename
>>> '/devices/platform/soc@0/a800000.usb/software_node'
>>> [    0.613195] CPU: 10 UID: 0 PID: 93 Comm: kworker/u49:0 Tainted: G
>>>      W           7.3.0-rc2-next-20260907-uwu+ #143 PREEMPT(full)
>>> [    0.613197] Tainted: [W]=WARN
>>> [    0.613198] Hardware name: Dell Inc. Latitude 7455/0FK7MX, BIOS
>>> 2.13.0 09/19/2025
>>> [    0.613199] Workqueue: events_unbound deferred_probe_work_func
>>> [    0.613206] Call trace:
>>> [..]
>>> [    0.613226]  device_create_managed_software_node+0xe8/0x108
>>> [    0.613227]  dwc3_qcom_probe+0x5a4/0x7b0
>>>
>>> There is indeed already a software node being created in the shared dwc3
>>> core code.
>>>
>>> Could there be any other consequences than a duplicate sysfs path?
>>>
>> Answering my own question: even worse messages when the controller switches
>> to gadget mode (which happens even with chargers when DRD is enabled)
>>
> 
> 
> Thanks for the report Val.
> 
> Hi Krishna,
> 
> Can you help send a revert patch until a proper change is in available?
> 
Sure, will send a revert patch.

Regards,
Krishna,