2023-03-21 07:30:18

by HaoTien Hsu

[permalink] [raw]
Subject: [PATCH v4] usb: xhci: tegra: fix sleep in atomic call

From: Wayne Chang <[email protected]>

When we set the OTG port to Host mode, we observed the following splat:
[ 167.057718] BUG: sleeping function called from invalid context at
include/linux/sched/mm.h:229
[ 167.057872] Workqueue: events tegra_xusb_usb_phy_work
[ 167.057954] Call trace:
[ 167.057962] dump_backtrace+0x0/0x210
[ 167.057996] show_stack+0x30/0x50
[ 167.058020] dump_stack_lvl+0x64/0x84
[ 167.058065] dump_stack+0x14/0x34
[ 167.058100] __might_resched+0x144/0x180
[ 167.058140] __might_sleep+0x64/0xd0
[ 167.058171] slab_pre_alloc_hook.constprop.0+0xa8/0x110
[ 167.058202] __kmalloc_track_caller+0x74/0x2b0
[ 167.058233] kvasprintf+0xa4/0x190
[ 167.058261] kasprintf+0x58/0x90
[ 167.058285] tegra_xusb_find_port_node.isra.0+0x58/0xd0
[ 167.058334] tegra_xusb_find_port+0x38/0xa0
[ 167.058380] tegra_xusb_padctl_get_usb3_companion+0x38/0xd0
[ 167.058430] tegra_xhci_id_notify+0x8c/0x1e0
[ 167.058473] notifier_call_chain+0x88/0x100
[ 167.058506] atomic_notifier_call_chain+0x44/0x70
[ 167.058537] tegra_xusb_usb_phy_work+0x60/0xd0
[ 167.058581] process_one_work+0x1dc/0x4c0
[ 167.058618] worker_thread+0x54/0x410
[ 167.058650] kthread+0x188/0x1b0
[ 167.058672] ret_from_fork+0x10/0x20

The function tegra_xusb_padctl_get_usb3_companion eventually calls
tegra_xusb_find_port and this in turn calls kasprintf which might sleep
and so cannot be called from an atomic context.

Fix this by moving the call to tegra_xusb_padctl_get_usb3_companion to
the tegra_xhci_id_work function where it is really needed.

Fixes: f836e7843036 ("usb: xhci-tegra: Add OTG support")
Cc: [email protected]
Signed-off-by: Wayne Chang <[email protected]>
Signed-off-by: Haotien Hsu <[email protected]>
---
V3 -> V4: Remove copyright change from this patch
V2 -> V3: Add version information
V1 -> V2: Add "Fixes" and "Cc:" lines and update copyright years
---
drivers/usb/host/xhci-tegra.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/usb/host/xhci-tegra.c b/drivers/usb/host/xhci-tegra.c
index 1ff22f675930..b40e897ec092 100644
--- a/drivers/usb/host/xhci-tegra.c
+++ b/drivers/usb/host/xhci-tegra.c
@@ -1360,6 +1360,10 @@ static void tegra_xhci_id_work(struct work_struct *work)

mutex_unlock(&tegra->lock);

+ tegra->otg_usb3_port = tegra_xusb_padctl_get_usb3_companion(
+ tegra->padctl,
+ tegra->otg_usb2_port);
+
if (tegra->host_mode) {
/* switch to host mode */
if (tegra->otg_usb3_port >= 0) {
@@ -1474,9 +1478,6 @@ static int tegra_xhci_id_notify(struct notifier_block *nb,
}

tegra->otg_usb2_port = tegra_xusb_get_usb2_port(tegra, usbphy);
- tegra->otg_usb3_port = tegra_xusb_padctl_get_usb3_companion(
- tegra->padctl,
- tegra->otg_usb2_port);

tegra->host_mode = (usbphy->last_event == USB_EVENT_ID) ? true : false;

--
2.25.1



2023-03-23 16:33:17

by Greg Kroah-Hartman

[permalink] [raw]
Subject: Re: [PATCH v4] usb: xhci: tegra: fix sleep in atomic call

On Tue, Mar 21, 2023 at 03:29:46PM +0800, Haotien Hsu wrote:
> From: Wayne Chang <[email protected]>
>
> When we set the OTG port to Host mode, we observed the following splat:

What "OTG" port? That is not a USB thing anymore at all. How is this
part of a xhci controller?

> [ 167.057718] BUG: sleeping function called from invalid context at
> include/linux/sched/mm.h:229
> [ 167.057872] Workqueue: events tegra_xusb_usb_phy_work
> [ 167.057954] Call trace:
> [ 167.057962] dump_backtrace+0x0/0x210
> [ 167.057996] show_stack+0x30/0x50
> [ 167.058020] dump_stack_lvl+0x64/0x84
> [ 167.058065] dump_stack+0x14/0x34
> [ 167.058100] __might_resched+0x144/0x180
> [ 167.058140] __might_sleep+0x64/0xd0
> [ 167.058171] slab_pre_alloc_hook.constprop.0+0xa8/0x110
> [ 167.058202] __kmalloc_track_caller+0x74/0x2b0
> [ 167.058233] kvasprintf+0xa4/0x190
> [ 167.058261] kasprintf+0x58/0x90
> [ 167.058285] tegra_xusb_find_port_node.isra.0+0x58/0xd0
> [ 167.058334] tegra_xusb_find_port+0x38/0xa0
> [ 167.058380] tegra_xusb_padctl_get_usb3_companion+0x38/0xd0
> [ 167.058430] tegra_xhci_id_notify+0x8c/0x1e0
> [ 167.058473] notifier_call_chain+0x88/0x100
> [ 167.058506] atomic_notifier_call_chain+0x44/0x70
> [ 167.058537] tegra_xusb_usb_phy_work+0x60/0xd0
> [ 167.058581] process_one_work+0x1dc/0x4c0
> [ 167.058618] worker_thread+0x54/0x410
> [ 167.058650] kthread+0x188/0x1b0
> [ 167.058672] ret_from_fork+0x10/0x20
>
> The function tegra_xusb_padctl_get_usb3_companion eventually calls
> tegra_xusb_find_port and this in turn calls kasprintf which might sleep
> and so cannot be called from an atomic context.
>
> Fix this by moving the call to tegra_xusb_padctl_get_usb3_companion to
> the tegra_xhci_id_work function where it is really needed.
>
> Fixes: f836e7843036 ("usb: xhci-tegra: Add OTG support")
> Cc: [email protected]
> Signed-off-by: Wayne Chang <[email protected]>
> Signed-off-by: Haotien Hsu <[email protected]>
> ---
> V3 -> V4: Remove copyright change from this patch
> V2 -> V3: Add version information
> V1 -> V2: Add "Fixes" and "Cc:" lines and update copyright years
> ---
> drivers/usb/host/xhci-tegra.c | 7 ++++---
> 1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/usb/host/xhci-tegra.c b/drivers/usb/host/xhci-tegra.c
> index 1ff22f675930..b40e897ec092 100644
> --- a/drivers/usb/host/xhci-tegra.c
> +++ b/drivers/usb/host/xhci-tegra.c
> @@ -1360,6 +1360,10 @@ static void tegra_xhci_id_work(struct work_struct *work)
>
> mutex_unlock(&tegra->lock);
>
> + tegra->otg_usb3_port = tegra_xusb_padctl_get_usb3_companion(
> + tegra->padctl,
> + tegra->otg_usb2_port);
> +

You have 100 columns, please use them, ending a line with "(" is not
generally a good idea.

thanks,

greg k-h

2023-03-24 06:52:21

by HaoTien Hsu

[permalink] [raw]
Subject: Re: [PATCH v4] usb: xhci: tegra: fix sleep in atomic call

On Thu, 2023-03-23 at 17:22 +0100, Greg Kroah-Hartman wrote:
> External email: Use caution opening links or attachments
>
>
> On Tue, Mar 21, 2023 at 03:29:46PM +0800, Haotien Hsu wrote:
> > From: Wayne Chang <[email protected]>
> >
> > When we set the OTG port to Host mode, we observed the following
> > splat:
>
> What "OTG" port? That is not a USB thing anymore at all. How is
> this
> part of a xhci controller?

I will use 'dual-role port' instead.

When the role of a super-speed port is switched, we need to clear or
set port power.
Our host mode controller and device mode controller share the same
super-speed port hardware.
When a SS port is going to be switched to device mode role, we have to
clear PP at host mode controller side to release SS port hardware to
device mode controller.
When a SS port is going to be switched to host mode role, we have to
set PP at host mode controller side to acquire SS port hardware.

>
> > [ 167.057718] BUG: sleeping function called from invalid context
> > at
> > include/linux/sched/mm.h:229
> > [ 167.057872] Workqueue: events tegra_xusb_usb_phy_work
> > [ 167.057954] Call trace:
> > [ 167.057962] dump_backtrace+0x0/0x210
> > [ 167.057996] show_stack+0x30/0x50
> > [ 167.058020] dump_stack_lvl+0x64/0x84
> > [ 167.058065] dump_stack+0x14/0x34
> > [ 167.058100] __might_resched+0x144/0x180
> > [ 167.058140] __might_sleep+0x64/0xd0
> > [ 167.058171] slab_pre_alloc_hook.constprop.0+0xa8/0x110
> > [ 167.058202] __kmalloc_track_caller+0x74/0x2b0
> > [ 167.058233] kvasprintf+0xa4/0x190
> > [ 167.058261] kasprintf+0x58/0x90
> > [ 167.058285] tegra_xusb_find_port_node.isra.0+0x58/0xd0
> > [ 167.058334] tegra_xusb_find_port+0x38/0xa0
> > [ 167.058380] tegra_xusb_padctl_get_usb3_companion+0x38/0xd0
> > [ 167.058430] tegra_xhci_id_notify+0x8c/0x1e0
> > [ 167.058473] notifier_call_chain+0x88/0x100
> > [ 167.058506] atomic_notifier_call_chain+0x44/0x70
> > [ 167.058537] tegra_xusb_usb_phy_work+0x60/0xd0
> > [ 167.058581] process_one_work+0x1dc/0x4c0
> > [ 167.058618] worker_thread+0x54/0x410
> > [ 167.058650] kthread+0x188/0x1b0
> > [ 167.058672] ret_from_fork+0x10/0x20
> >
> > The function tegra_xusb_padctl_get_usb3_companion eventually calls
> > tegra_xusb_find_port and this in turn calls kasprintf which might
> > sleep
> > and so cannot be called from an atomic context.
> >
> > Fix this by moving the call to tegra_xusb_padctl_get_usb3_companion
> > to
> > the tegra_xhci_id_work function where it is really needed.
> >
> > Fixes: f836e7843036 ("usb: xhci-tegra: Add OTG support")
> > Cc: [email protected]
> > Signed-off-by: Wayne Chang <[email protected]>
> > Signed-off-by: Haotien Hsu <[email protected]>
> > ---
> > V3 -> V4: Remove copyright change from this patch
> > V2 -> V3: Add version information
> > V1 -> V2: Add "Fixes" and "Cc:" lines and update copyright years
> > ---
> > drivers/usb/host/xhci-tegra.c | 7 ++++---
> > 1 file changed, 4 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/usb/host/xhci-tegra.c b/drivers/usb/host/xhci-
> > tegra.c
> > index 1ff22f675930..b40e897ec092 100644
> > --- a/drivers/usb/host/xhci-tegra.c
> > +++ b/drivers/usb/host/xhci-tegra.c
> > @@ -1360,6 +1360,10 @@ static void tegra_xhci_id_work(struct
> > work_struct *work)
> >
> > mutex_unlock(&tegra->lock);
> >
> > + tegra->otg_usb3_port = tegra_xusb_padctl_get_usb3_companion(
> > + tegra-
> > >padctl,
> > + tegra-
> > >otg_usb2_port);
> > +
>
> You have 100 columns, please use them, ending a line with "(" is not
> generally a good idea.

I will resubmit a new patch.

>
> thanks,
>
> greg k-h

2023-03-24 09:36:26

by HaoTien Hsu

[permalink] [raw]
Subject: Re: [PATCH v4] usb: xhci: tegra: fix sleep in atomic call

On Fri, 2023-03-24 at 06:46 +0000, Haotien Hsu wrote:
> On Thu, 2023-03-23 at 17:22 +0100, Greg Kroah-Hartman wrote:
> > External email: Use caution opening links or attachments
> >
> >
> > On Tue, Mar 21, 2023 at 03:29:46PM +0800, Haotien Hsu wrote:
> > > From: Wayne Chang <[email protected]>
> > >
> > > When we set the OTG port to Host mode, we observed the following
> > > splat:
> >
> > What "OTG" port? That is not a USB thing anymore at all. How is
> > this
> > part of a xhci controller?
>
> I will use 'dual-role port' instead.
>
> When the role of a super-speed port is switched, we need to clear or
> set XHCI PORTSC.PP.
> Our host mode controller and device mode controller share the same
> super-speed port hardware.
> When a super-speed port is going to be switched to device mode role,
> we have to clear XHCI PORTSC.PP at host mode controller side to
> release super-speed port hardware to device mode controller.
> When a super-speed port is going to be switched to host mode role, we
> have to set XHCI PORTSC.PP at host mode controller side to
> acquire super-speed port hardware.
>
I replaced SS with super-speed and PP with 'XHCI PORTSC.PP' in the
above description to make it clear.

> > > [ 167.057718] BUG: sleeping function called from invalid context
> > > at
> > > include/linux/sched/mm.h:229
> > > [ 167.057872] Workqueue: events tegra_xusb_usb_phy_work
> > > [ 167.057954] Call trace:
> > > [ 167.057962] dump_backtrace+0x0/0x210
> > > [ 167.057996] show_stack+0x30/0x50
> > > [ 167.058020] dump_stack_lvl+0x64/0x84
> > > [ 167.058065] dump_stack+0x14/0x34
> > > [ 167.058100] __might_resched+0x144/0x180
> > > [ 167.058140] __might_sleep+0x64/0xd0
> > > [ 167.058171] slab_pre_alloc_hook.constprop.0+0xa8/0x110
> > > [ 167.058202] __kmalloc_track_caller+0x74/0x2b0
> > > [ 167.058233] kvasprintf+0xa4/0x190
> > > [ 167.058261] kasprintf+0x58/0x90
> > > [ 167.058285] tegra_xusb_find_port_node.isra.0+0x58/0xd0
> > > [ 167.058334] tegra_xusb_find_port+0x38/0xa0
> > > [ 167.058380] tegra_xusb_padctl_get_usb3_companion+0x38/0xd0
> > > [ 167.058430] tegra_xhci_id_notify+0x8c/0x1e0
> > > [ 167.058473] notifier_call_chain+0x88/0x100
> > > [ 167.058506] atomic_notifier_call_chain+0x44/0x70
> > > [ 167.058537] tegra_xusb_usb_phy_work+0x60/0xd0
> > > [ 167.058581] process_one_work+0x1dc/0x4c0
> > > [ 167.058618] worker_thread+0x54/0x410
> > > [ 167.058650] kthread+0x188/0x1b0
> > > [ 167.058672] ret_from_fork+0x10/0x20
> > >
> > > The function tegra_xusb_padctl_get_usb3_companion eventually
> > > calls
> > > tegra_xusb_find_port and this in turn calls kasprintf which might
> > > sleep
> > > and so cannot be called from an atomic context.
> > >
> > > Fix this by moving the call to
> > > tegra_xusb_padctl_get_usb3_companion
> > > to
> > > the tegra_xhci_id_work function where it is really needed.
> > >
> > > Fixes: f836e7843036 ("usb: xhci-tegra: Add OTG support")
> > > Cc: [email protected]
> > > Signed-off-by: Wayne Chang <[email protected]>
> > > Signed-off-by: Haotien Hsu <[email protected]>
> > > ---
> > > V3 -> V4: Remove copyright change from this patch
> > > V2 -> V3: Add version information
> > > V1 -> V2: Add "Fixes" and "Cc:" lines and update copyright years
> > > ---
> > > drivers/usb/host/xhci-tegra.c | 7 ++++---
> > > 1 file changed, 4 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/usb/host/xhci-tegra.c
> > > b/drivers/usb/host/xhci-
> > > tegra.c
> > > index 1ff22f675930..b40e897ec092 100644
> > > --- a/drivers/usb/host/xhci-tegra.c
> > > +++ b/drivers/usb/host/xhci-tegra.c
> > > @@ -1360,6 +1360,10 @@ static void tegra_xhci_id_work(struct
> > > work_struct *work)
> > >
> > > mutex_unlock(&tegra->lock);
> > >
> > > + tegra->otg_usb3_port =
> > > tegra_xusb_padctl_get_usb3_companion(
> > > + tegra-
> > > > padctl,
> > > + tegra-
> > > > otg_usb2_port);
> > > +
> >
> > You have 100 columns, please use them, ending a line with "(" is
> > not
> > generally a good idea.
>
> I will resubmit a new patch.
>
> > thanks,
> >
> > greg k-h