[PATCH] initialize struct otg_fsm earlier

[PATCH] initialize struct otg_fsm earlier

[Oliver Neukum]

The earlier fix bf88fef0b6f1 ("usb: otg-fsm: Fix hrtimer list
corruption") in effect hid an issue with intialization.
In effect it replaces the racy continous reinitialization
of fsm->hnp_polling_work with a delayed one-time
initialization.

This just makes no sense. As a single initialization
is sufficient, the clean solution is just to do it once
and do it early enough.

Fixes: bf88fef0b6f1 ("usb: otg-fsm: Fix hrtimer list corruption")
Signed-off-by: Oliver Neukum <oneukum@suse.com>
---
 drivers/usb/common/usb-otg-fsm.c | 7 +------
 drivers/usb/phy/phy-fsl-usb.c    | 1 +
 include/linux/usb/otg-fsm.h      | 1 -
 3 files changed, 2 insertions(+), 7 deletions(-)

diff --git a/drivers/usb/common/usb-otg-fsm.c b/drivers/usb/common/usb-otg-fsm.c
index 0697fde51d00..0aa2eb7396ce 100644
--- a/drivers/usb/common/usb-otg-fsm.c
+++ b/drivers/usb/common/usb-otg-fsm.c
@@ -117,7 +117,7 @@ static void otg_leave_state(struct otg_fsm *fsm, enum usb_otg_state old_state)
 	}
 }
 
-static void otg_hnp_polling_work(struct work_struct *work)
+void otg_hnp_polling_work(struct work_struct *work)
 {
 	struct otg_fsm *fsm = container_of(to_delayed_work(work),
 				struct otg_fsm, hnp_polling_work);
@@ -193,11 +193,6 @@ static void otg_start_hnp_polling(struct otg_fsm *fsm)
 	if (!fsm->host_req_flag)
 		return;
 
-	if (!fsm->hnp_work_inited) {
-		INIT_DELAYED_WORK(&fsm->hnp_polling_work, otg_hnp_polling_work);
-		fsm->hnp_work_inited = true;
-	}
-
 	schedule_delayed_work(&fsm->hnp_polling_work,
 					msecs_to_jiffies(T_HOST_REQ_POLL));
 }
diff --git a/drivers/usb/phy/phy-fsl-usb.c b/drivers/usb/phy/phy-fsl-usb.c
index 79617bb0a70e..4553aaab3761 100644
--- a/drivers/usb/phy/phy-fsl-usb.c
+++ b/drivers/usb/phy/phy-fsl-usb.c
@@ -844,6 +844,7 @@ int usb_otg_start(struct platform_device *pdev)
 
 	/* Initialize the state machine structure with default values */
 	SET_OTG_STATE(otg_trans, OTG_STATE_UNDEFINED);
+	INIT_DELAYED_WORK(&fsm->hnp_polling_work, otg_hnp_polling_work);
 	fsm->otg = p_otg->phy.otg;
 
 	/* We don't require predefined MEM/IRQ resource index */
diff --git a/include/linux/usb/otg-fsm.h b/include/linux/usb/otg-fsm.h
index 6135d076c53d..cc0bc4edf227 100644
--- a/include/linux/usb/otg-fsm.h
+++ b/include/linux/usb/otg-fsm.h
@@ -183,7 +183,6 @@ struct otg_fsm {
 	struct mutex lock;
 	u8 *host_req_flag;
 	struct delayed_work hnp_polling_work;
-	bool hnp_work_inited;
 	bool state_changed;
 };
 
-- 
2.42.0

Re: [PATCH] initialize struct otg_fsm earlier

[Peter Chen]

On 23-11-13 18:31:20, Oliver Neukum wrote:
> The earlier fix bf88fef0b6f1 ("usb: otg-fsm: Fix hrtimer list
> corruption") in effect hid an issue with intialization.

Typo, hit an issue with initialization.

> In effect it replaces the racy continous reinitialization
> of fsm->hnp_polling_work with a delayed one-time
> initialization.
> 
> This just makes no sense. As a single initialization
> is sufficient, the clean solution is just to do it once
> and do it early enough.

The phy-fsl-usb is not the only user, there are other users, eg,
chipidea.

Peter
> 
> Fixes: bf88fef0b6f1 ("usb: otg-fsm: Fix hrtimer list corruption")
> Signed-off-by: Oliver Neukum <oneukum@suse.com>
> ---
>  drivers/usb/common/usb-otg-fsm.c | 7 +------
>  drivers/usb/phy/phy-fsl-usb.c    | 1 +
>  include/linux/usb/otg-fsm.h      | 1 -
>  3 files changed, 2 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/usb/common/usb-otg-fsm.c b/drivers/usb/common/usb-otg-fsm.c
> index 0697fde51d00..0aa2eb7396ce 100644
> --- a/drivers/usb/common/usb-otg-fsm.c
> +++ b/drivers/usb/common/usb-otg-fsm.c
> @@ -117,7 +117,7 @@ static void otg_leave_state(struct otg_fsm *fsm, enum usb_otg_state old_state)
>  	}
>  }
>  
> -static void otg_hnp_polling_work(struct work_struct *work)
> +void otg_hnp_polling_work(struct work_struct *work)
>  {
>  	struct otg_fsm *fsm = container_of(to_delayed_work(work),
>  				struct otg_fsm, hnp_polling_work);
> @@ -193,11 +193,6 @@ static void otg_start_hnp_polling(struct otg_fsm *fsm)
>  	if (!fsm->host_req_flag)
>  		return;
>  
> -	if (!fsm->hnp_work_inited) {
> -		INIT_DELAYED_WORK(&fsm->hnp_polling_work, otg_hnp_polling_work);
> -		fsm->hnp_work_inited = true;
> -	}
> -
>  	schedule_delayed_work(&fsm->hnp_polling_work,
>  					msecs_to_jiffies(T_HOST_REQ_POLL));
>  }
> diff --git a/drivers/usb/phy/phy-fsl-usb.c b/drivers/usb/phy/phy-fsl-usb.c
> index 79617bb0a70e..4553aaab3761 100644
> --- a/drivers/usb/phy/phy-fsl-usb.c
> +++ b/drivers/usb/phy/phy-fsl-usb.c
> @@ -844,6 +844,7 @@ int usb_otg_start(struct platform_device *pdev)
>  
>  	/* Initialize the state machine structure with default values */
>  	SET_OTG_STATE(otg_trans, OTG_STATE_UNDEFINED);
> +	INIT_DELAYED_WORK(&fsm->hnp_polling_work, otg_hnp_polling_work);
>  	fsm->otg = p_otg->phy.otg;
>  
>  	/* We don't require predefined MEM/IRQ resource index */
> diff --git a/include/linux/usb/otg-fsm.h b/include/linux/usb/otg-fsm.h
> index 6135d076c53d..cc0bc4edf227 100644
> --- a/include/linux/usb/otg-fsm.h
> +++ b/include/linux/usb/otg-fsm.h
> @@ -183,7 +183,6 @@ struct otg_fsm {
>  	struct mutex lock;
>  	u8 *host_req_flag;
>  	struct delayed_work hnp_polling_work;
> -	bool hnp_work_inited;
>  	bool state_changed;
>  };
>  
> -- 
> 2.42.0
> 

-- 

Thanks,
Peter Chen

Re: [PATCH] initialize struct otg_fsm earlier

[Oliver Neukum]

Hi,

On 14.11.23 08:27, Peter Chen wrote:
> On 23-11-13 18:31:20, Oliver Neukum wrote:
>> The earlier fix bf88fef0b6f1 ("usb: otg-fsm: Fix hrtimer list
>> corruption") in effect hid an issue with intialization.
> 
> Typo, hit an issue with initialization.

No, it hides an issue.

>> In effect it replaces the racy continous reinitialization
>> of fsm->hnp_polling_work with a delayed one-time
>> initialization.
>>
>> This just makes no sense. As a single initialization
>> is sufficient, the clean solution is just to do it once
>> and do it early enough.
> 
> The phy-fsl-usb is not the only user, there are other users, eg,
> chipidea.

Interesting.
Are you saying that somebody calls otg_start_hnp_polling() without
a prior call to usb_otg_start()?
If so, how is struct otg_fsm initialized?
Currently struct otg_fsm is partially initialized and that just
makes no sense.

	Regards
		Oliver

Re: [PATCH] initialize struct otg_fsm earlier

[Alan Stern]

On Tue, Nov 14, 2023 at 10:44:36AM +0100, Oliver Neukum wrote:
> Hi,
> 
> On 14.11.23 08:27, Peter Chen wrote:
> > On 23-11-13 18:31:20, Oliver Neukum wrote:
> > > The earlier fix bf88fef0b6f1 ("usb: otg-fsm: Fix hrtimer list
> > > corruption") in effect hid an issue with intialization.
> > 
> > Typo, hit an issue with initialization.
> 
> No, it hides an issue.

The typo was in the last word of the line, not the fourth word.  That is:

	s/intial/initial/, not s/hid/hit/.

Peter had a typo in his correction.  :-)

Alan Stern

Re: [PATCH] initialize struct otg_fsm earlier

[Peter Chen]

On 23-11-14 10:33:55, Alan Stern wrote:
> On Tue, Nov 14, 2023 at 10:44:36AM +0100, Oliver Neukum wrote:
> > Hi,
> > 
> > On 14.11.23 08:27, Peter Chen wrote:
> > > On 23-11-13 18:31:20, Oliver Neukum wrote:
> > > > The earlier fix bf88fef0b6f1 ("usb: otg-fsm: Fix hrtimer list
> > > > corruption") in effect hid an issue with intialization.
> > > 
> > > Typo, hit an issue with initialization.
> > 
> > No, it hides an issue.
> 
> The typo was in the last word of the line, not the fourth word.  That is:
> 
> 	s/intial/initial/, not s/hid/hit/.
> 
> Peter had a typo in his correction.  :-)
> 

Alan, thanks for pointing it.

-- 

Thanks,
Peter Chen

Re: [PATCH] initialize struct otg_fsm earlier

[Peter Chen]

On 23-11-14 10:44:36, Oliver Neukum wrote:
> Hi,
> 
> On 14.11.23 08:27, Peter Chen wrote:
> > On 23-11-13 18:31:20, Oliver Neukum wrote:
> > > The earlier fix bf88fef0b6f1 ("usb: otg-fsm: Fix hrtimer list
> > > corruption") in effect hid an issue with intialization.
> > 
> > Typo, hit an issue with initialization.
> 
> No, it hides an issue.
> 
> > > In effect it replaces the racy continous reinitialization
> > > of fsm->hnp_polling_work with a delayed one-time
> > > initialization.
> > > 
> > > This just makes no sense. As a single initialization
> > > is sufficient, the clean solution is just to do it once
> > > and do it early enough.
> > 
> > The phy-fsl-usb is not the only user, there are other users, eg,
> > chipidea.
> 
> Interesting.
> Are you saying that somebody calls otg_start_hnp_polling() without
> a prior call to usb_otg_start()?

Yes, phy-fsl-usb is the freescale implementation. Others may have
different, see ci_hdrc_otg_fsm_init for detail.

Peter

> If so, how is struct otg_fsm initialized?
> Currently struct otg_fsm is partially initialized and that just
> makes no sense.
> 
> 	Regards
> 		Oliver

-- 

Thanks,
Peter Chen