[PATCH] mhi: pci_generic: Ensure device readiness before starting MHI

[PATCH] mhi: pci_generic: Ensure device readiness before starting MHI

[Loic Poulain]

The PCI device may have not been bound from cold boot and be in
undefined state, or simply not yet ready for MHI operations. This
change ensures that the MHI layer is reset to initial state and
ready for MHI initialization and power up.

Signed-off-by: Loic Poulain <loic.poulain@linaro.org>
---
 drivers/bus/mhi/pci_generic.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/drivers/bus/mhi/pci_generic.c b/drivers/bus/mhi/pci_generic.c
index c20f59e..bfa0a1e 100644
--- a/drivers/bus/mhi/pci_generic.c
+++ b/drivers/bus/mhi/pci_generic.c
@@ -17,6 +17,8 @@
 #include <linux/timer.h>
 #include <linux/workqueue.h>
 
+#include "core/internal.h"
+
 #define MHI_PCI_DEFAULT_BAR_NUM 0
 
 #define MHI_POST_RESET_DELAY_MS 500
@@ -391,6 +393,22 @@ static void health_check(struct timer_list *t)
 	mod_timer(&mhi_pdev->health_check_timer, jiffies + HEALTH_CHECK_PERIOD);
 }
 
+static void __mhi_sw_reset(struct mhi_controller *mhi_cntrl)
+{
+	unsigned int max_wait_ready = 200;
+
+	mhi_pci_write_reg(mhi_cntrl, mhi_cntrl->regs + MHICTRL,
+			  MHICTRL_RESET_MASK);
+
+	while (mhi_get_mhi_state(mhi_cntrl) != MHI_STATE_READY) {
+		if (!max_wait_ready--) {
+			dev_warn(mhi_cntrl->cntrl_dev, "Not ready\n");
+			break;
+		}
+		msleep(50);
+	}
+}
+
 static int mhi_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 {
 	const struct mhi_pci_dev_info *info = (struct mhi_pci_dev_info *) id->driver_data;
@@ -451,6 +469,9 @@ static int mhi_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 		goto err_unregister;
 	}
 
+	/* Before starting MHI, ensure device is in good initial state */
+	__mhi_sw_reset(mhi_cntrl);
+
 	err = mhi_sync_power_up(mhi_cntrl);
 	if (err) {
 		dev_err(&pdev->dev, "failed to power up MHI controller\n");
@@ -532,6 +553,8 @@ static void mhi_pci_reset_done(struct pci_dev *pdev)
 		return;
 	}
 
+	__mhi_sw_reset(mhi_cntrl);
+
 	err = mhi_sync_power_up(mhi_cntrl);
 	if (err) {
 		dev_err(&pdev->dev, "failed to power up MHI controller\n");
-- 
2.7.4

Re: [PATCH] mhi: pci_generic: Ensure device readiness before starting MHI

[Jeffrey Hugo]

On 2/11/2021 10:08 AM, Loic Poulain wrote:
> The PCI device may have not been bound from cold boot and be in
> undefined state, or simply not yet ready for MHI operations. This
> change ensures that the MHI layer is reset to initial state and
> ready for MHI initialization and power up.
> 
> Signed-off-by: Loic Poulain <loic.poulain@linaro.org>
> ---
>   drivers/bus/mhi/pci_generic.c | 23 +++++++++++++++++++++++
>   1 file changed, 23 insertions(+)
> 
> diff --git a/drivers/bus/mhi/pci_generic.c b/drivers/bus/mhi/pci_generic.c
> index c20f59e..bfa0a1e 100644
> --- a/drivers/bus/mhi/pci_generic.c
> +++ b/drivers/bus/mhi/pci_generic.c
> @@ -17,6 +17,8 @@
>   #include <linux/timer.h>
>   #include <linux/workqueue.h>
>   
> +#include "core/internal.h"
> +
>   #define MHI_PCI_DEFAULT_BAR_NUM 0
>   
>   #define MHI_POST_RESET_DELAY_MS 500
> @@ -391,6 +393,22 @@ static void health_check(struct timer_list *t)
>   	mod_timer(&mhi_pdev->health_check_timer, jiffies + HEALTH_CHECK_PERIOD);
>   }
>   
> +static void __mhi_sw_reset(struct mhi_controller *mhi_cntrl)
> +{
> +	unsigned int max_wait_ready = 200;
> +
> +	mhi_pci_write_reg(mhi_cntrl, mhi_cntrl->regs + MHICTRL,
> +			  MHICTRL_RESET_MASK);
> +
> +	while (mhi_get_mhi_state(mhi_cntrl) != MHI_STATE_READY) {
> +		if (!max_wait_ready--) {
> +			dev_warn(mhi_cntrl->cntrl_dev, "Not ready\n");
> +			break;
> +		}
> +		msleep(50);
> +	}
> +}
> +
>   static int mhi_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
>   {
>   	const struct mhi_pci_dev_info *info = (struct mhi_pci_dev_info *) id->driver_data;
> @@ -451,6 +469,9 @@ static int mhi_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
>   		goto err_unregister;
>   	}
>   
> +	/* Before starting MHI, ensure device is in good initial state */
> +	__mhi_sw_reset(mhi_cntrl);
> +
>   	err = mhi_sync_power_up(mhi_cntrl);
>   	if (err) {
>   		dev_err(&pdev->dev, "failed to power up MHI controller\n");
> @@ -532,6 +553,8 @@ static void mhi_pci_reset_done(struct pci_dev *pdev)
>   		return;
>   	}
>   
> +	__mhi_sw_reset(mhi_cntrl);
> +
>   	err = mhi_sync_power_up(mhi_cntrl);
>   	if (err) {
>   		dev_err(&pdev->dev, "failed to power up MHI controller\n");
> 

So, I'm curious, how does this actually work?

 From what I can see, you define SBL images.  If those get loaded by the 
PBL, it doesn't happen over MHI.  PBL will not move MHI to ready state, 
except in the specific instance of a fatal error.

Your above change works if the device comes up straight in mission mode 
(AMSS), but if it comes up in PBL, you are going to hit the timeout and 
dev_warn() every time.

-- 
Jeffrey Hugo
Qualcomm Technologies, Inc. is a member of the
Code Aurora Forum, a Linux Foundation Collaborative Project.

Re: [PATCH] mhi: pci_generic: Ensure device readiness before starting MHI

[Loic Poulain]

On Thu, 11 Feb 2021 at 18:13, Jeffrey Hugo <jhugo@codeaurora.org> wrote:
>
> On 2/11/2021 10:08 AM, Loic Poulain wrote:
> > The PCI device may have not been bound from cold boot and be in
> > undefined state, or simply not yet ready for MHI operations. This
> > change ensures that the MHI layer is reset to initial state and
> > ready for MHI initialization and power up.
> >
> > Signed-off-by: Loic Poulain <loic.poulain@linaro.org>
> > ---
> >   drivers/bus/mhi/pci_generic.c | 23 +++++++++++++++++++++++
> >   1 file changed, 23 insertions(+)
> >
> > diff --git a/drivers/bus/mhi/pci_generic.c b/drivers/bus/mhi/pci_generic.c
> > index c20f59e..bfa0a1e 100644
> > --- a/drivers/bus/mhi/pci_generic.c
> > +++ b/drivers/bus/mhi/pci_generic.c
> > @@ -17,6 +17,8 @@
> >   #include <linux/timer.h>
> >   #include <linux/workqueue.h>
> >
> > +#include "core/internal.h"
> > +
> >   #define MHI_PCI_DEFAULT_BAR_NUM 0
> >
> >   #define MHI_POST_RESET_DELAY_MS 500
> > @@ -391,6 +393,22 @@ static void health_check(struct timer_list *t)
> >       mod_timer(&mhi_pdev->health_check_timer, jiffies + HEALTH_CHECK_PERIOD);
> >   }
> >
> > +static void __mhi_sw_reset(struct mhi_controller *mhi_cntrl)
> > +{
> > +     unsigned int max_wait_ready = 200;
> > +
> > +     mhi_pci_write_reg(mhi_cntrl, mhi_cntrl->regs + MHICTRL,
> > +                       MHICTRL_RESET_MASK);
> > +
> > +     while (mhi_get_mhi_state(mhi_cntrl) != MHI_STATE_READY) {
> > +             if (!max_wait_ready--) {
> > +                     dev_warn(mhi_cntrl->cntrl_dev, "Not ready\n");
> > +                     break;
> > +             }
> > +             msleep(50);
> > +     }
> > +}
> > +
> >   static int mhi_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
> >   {
> >       const struct mhi_pci_dev_info *info = (struct mhi_pci_dev_info *) id->driver_data;
> > @@ -451,6 +469,9 @@ static int mhi_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
> >               goto err_unregister;
> >       }
> >
> > +     /* Before starting MHI, ensure device is in good initial state */
> > +     __mhi_sw_reset(mhi_cntrl);
> > +
> >       err = mhi_sync_power_up(mhi_cntrl);
> >       if (err) {
> >               dev_err(&pdev->dev, "failed to power up MHI controller\n");
> > @@ -532,6 +553,8 @@ static void mhi_pci_reset_done(struct pci_dev *pdev)
> >               return;
> >       }
> >
> > +     __mhi_sw_reset(mhi_cntrl);
> > +
> >       err = mhi_sync_power_up(mhi_cntrl);
> >       if (err) {
> >               dev_err(&pdev->dev, "failed to power up MHI controller\n");
> >
>
> So, I'm curious, how does this actually work?
>
>  From what I can see, you define SBL images.  If those get loaded by the
> PBL, it doesn't happen over MHI.  PBL will not move MHI to ready state,
> except in the specific instance of a fatal error.

I defined generic SBL images for flashless controller versions, but
mine is not, and so it boots directly in mission mode.

>
> Your above change works if the device comes up straight in mission mode
> (AMSS), but if it comes up in PBL, you are going to hit the timeout and
> dev_warn() every time.

Ok, I thought we should get into MHI ready state, whatever the
'execution environment'... So I definitely need to take that into
consideration. thanks.

Regards,
Loic

Re: [PATCH] mhi: pci_generic: Ensure device readiness before starting MHI

[Jeffrey Hugo]

On 2/11/2021 10:47 AM, Loic Poulain wrote:
> On Thu, 11 Feb 2021 at 18:13, Jeffrey Hugo <jhugo@codeaurora.org> wrote:
>>
>> On 2/11/2021 10:08 AM, Loic Poulain wrote:
>>> The PCI device may have not been bound from cold boot and be in
>>> undefined state, or simply not yet ready for MHI operations. This
>>> change ensures that the MHI layer is reset to initial state and
>>> ready for MHI initialization and power up.
>>>
>>> Signed-off-by: Loic Poulain <loic.poulain@linaro.org>
>>> ---
>>>    drivers/bus/mhi/pci_generic.c | 23 +++++++++++++++++++++++
>>>    1 file changed, 23 insertions(+)
>>>
>>> diff --git a/drivers/bus/mhi/pci_generic.c b/drivers/bus/mhi/pci_generic.c
>>> index c20f59e..bfa0a1e 100644
>>> --- a/drivers/bus/mhi/pci_generic.c
>>> +++ b/drivers/bus/mhi/pci_generic.c
>>> @@ -17,6 +17,8 @@
>>>    #include <linux/timer.h>
>>>    #include <linux/workqueue.h>
>>>
>>> +#include "core/internal.h"
>>> +
>>>    #define MHI_PCI_DEFAULT_BAR_NUM 0
>>>
>>>    #define MHI_POST_RESET_DELAY_MS 500
>>> @@ -391,6 +393,22 @@ static void health_check(struct timer_list *t)
>>>        mod_timer(&mhi_pdev->health_check_timer, jiffies + HEALTH_CHECK_PERIOD);
>>>    }
>>>
>>> +static void __mhi_sw_reset(struct mhi_controller *mhi_cntrl)
>>> +{
>>> +     unsigned int max_wait_ready = 200;
>>> +
>>> +     mhi_pci_write_reg(mhi_cntrl, mhi_cntrl->regs + MHICTRL,
>>> +                       MHICTRL_RESET_MASK);
>>> +
>>> +     while (mhi_get_mhi_state(mhi_cntrl) != MHI_STATE_READY) {
>>> +             if (!max_wait_ready--) {
>>> +                     dev_warn(mhi_cntrl->cntrl_dev, "Not ready\n");
>>> +                     break;
>>> +             }
>>> +             msleep(50);
>>> +     }
>>> +}
>>> +
>>>    static int mhi_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
>>>    {
>>>        const struct mhi_pci_dev_info *info = (struct mhi_pci_dev_info *) id->driver_data;
>>> @@ -451,6 +469,9 @@ static int mhi_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
>>>                goto err_unregister;
>>>        }
>>>
>>> +     /* Before starting MHI, ensure device is in good initial state */
>>> +     __mhi_sw_reset(mhi_cntrl);
>>> +
>>>        err = mhi_sync_power_up(mhi_cntrl);
>>>        if (err) {
>>>                dev_err(&pdev->dev, "failed to power up MHI controller\n");
>>> @@ -532,6 +553,8 @@ static void mhi_pci_reset_done(struct pci_dev *pdev)
>>>                return;
>>>        }
>>>
>>> +     __mhi_sw_reset(mhi_cntrl);
>>> +
>>>        err = mhi_sync_power_up(mhi_cntrl);
>>>        if (err) {
>>>                dev_err(&pdev->dev, "failed to power up MHI controller\n");
>>>
>>
>> So, I'm curious, how does this actually work?
>>
>>   From what I can see, you define SBL images.  If those get loaded by the
>> PBL, it doesn't happen over MHI.  PBL will not move MHI to ready state,
>> except in the specific instance of a fatal error.
> 
> I defined generic SBL images for flashless controller versions, but
> mine is not, and so it boots directly in mission mode.
> 
>>
>> Your above change works if the device comes up straight in mission mode
>> (AMSS), but if it comes up in PBL, you are going to hit the timeout and
>> dev_warn() every time.
> 
> Ok, I thought we should get into MHI ready state, whatever the
> 'execution environment'... So I definitely need to take that into
> consideration. thanks.

I could see where you could think that, which is why I commented.  I 
didn't want you to run into issues later, assuming those issues are 
valid to you.

MHI only gets into the ready state via EEs which drive MHI.  PBL 
famously does not drive MHI because PBL is encoded into hardware and 
extremely difficult to fix, so it is generally designed with the mantra 
of "simpler is more reliable".

Hopefully I didn't throw a wrench in things for you.  Just trying to 
save you some pain later.

-- 
Jeffrey Hugo
Qualcomm Technologies, Inc. is a member of the
Code Aurora Forum, a Linux Foundation Collaborative Project.

Re: [PATCH] mhi: pci_generic: Ensure device readiness before starting MHI

[Loic Poulain]

On Thu, 11 Feb 2021 at 18:55, Jeffrey Hugo <jhugo@codeaurora.org> wrote:
>
> On 2/11/2021 10:47 AM, Loic Poulain wrote:
> > On Thu, 11 Feb 2021 at 18:13, Jeffrey Hugo <jhugo@codeaurora.org> wrote:
> >>
> >> On 2/11/2021 10:08 AM, Loic Poulain wrote:
> >>> The PCI device may have not been bound from cold boot and be in
> >>> undefined state, or simply not yet ready for MHI operations. This
> >>> change ensures that the MHI layer is reset to initial state and
> >>> ready for MHI initialization and power up.
> >>>
> >>> Signed-off-by: Loic Poulain <loic.poulain@linaro.org>
> >>> ---
> >>>    drivers/bus/mhi/pci_generic.c | 23 +++++++++++++++++++++++
> >>>    1 file changed, 23 insertions(+)
> >>>
> >>> diff --git a/drivers/bus/mhi/pci_generic.c b/drivers/bus/mhi/pci_generic.c
> >>> index c20f59e..bfa0a1e 100644
> >>> --- a/drivers/bus/mhi/pci_generic.c
> >>> +++ b/drivers/bus/mhi/pci_generic.c
> >>> @@ -17,6 +17,8 @@
> >>>    #include <linux/timer.h>
> >>>    #include <linux/workqueue.h>
> >>>
> >>> +#include "core/internal.h"
> >>> +
> >>>    #define MHI_PCI_DEFAULT_BAR_NUM 0
> >>>
> >>>    #define MHI_POST_RESET_DELAY_MS 500
> >>> @@ -391,6 +393,22 @@ static void health_check(struct timer_list *t)
> >>>        mod_timer(&mhi_pdev->health_check_timer, jiffies + HEALTH_CHECK_PERIOD);
> >>>    }
> >>>
> >>> +static void __mhi_sw_reset(struct mhi_controller *mhi_cntrl)
> >>> +{
> >>> +     unsigned int max_wait_ready = 200;
> >>> +
> >>> +     mhi_pci_write_reg(mhi_cntrl, mhi_cntrl->regs + MHICTRL,
> >>> +                       MHICTRL_RESET_MASK);
> >>> +
> >>> +     while (mhi_get_mhi_state(mhi_cntrl) != MHI_STATE_READY) {
> >>> +             if (!max_wait_ready--) {
> >>> +                     dev_warn(mhi_cntrl->cntrl_dev, "Not ready\n");
> >>> +                     break;
> >>> +             }
> >>> +             msleep(50);
> >>> +     }
> >>> +}
> >>> +
> >>>    static int mhi_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
> >>>    {
> >>>        const struct mhi_pci_dev_info *info = (struct mhi_pci_dev_info *) id->driver_data;
> >>> @@ -451,6 +469,9 @@ static int mhi_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
> >>>                goto err_unregister;
> >>>        }
> >>>
> >>> +     /* Before starting MHI, ensure device is in good initial state */
> >>> +     __mhi_sw_reset(mhi_cntrl);
> >>> +
> >>>        err = mhi_sync_power_up(mhi_cntrl);
> >>>        if (err) {
> >>>                dev_err(&pdev->dev, "failed to power up MHI controller\n");
> >>> @@ -532,6 +553,8 @@ static void mhi_pci_reset_done(struct pci_dev *pdev)
> >>>                return;
> >>>        }
> >>>
> >>> +     __mhi_sw_reset(mhi_cntrl);
> >>> +
> >>>        err = mhi_sync_power_up(mhi_cntrl);
> >>>        if (err) {
> >>>                dev_err(&pdev->dev, "failed to power up MHI controller\n");
> >>>
> >>
> >> So, I'm curious, how does this actually work?
> >>
> >>   From what I can see, you define SBL images.  If those get loaded by the
> >> PBL, it doesn't happen over MHI.  PBL will not move MHI to ready state,
> >> except in the specific instance of a fatal error.
> >
> > I defined generic SBL images for flashless controller versions, but
> > mine is not, and so it boots directly in mission mode.
> >
> >>
> >> Your above change works if the device comes up straight in mission mode
> >> (AMSS), but if it comes up in PBL, you are going to hit the timeout and
> >> dev_warn() every time.
> >
> > Ok, I thought we should get into MHI ready state, whatever the
> > 'execution environment'... So I definitely need to take that into
> > consideration. thanks.
>
> I could see where you could think that, which is why I commented.  I
> didn't want you to run into issues later, assuming those issues are
> valid to you.
>
> MHI only gets into the ready state via EEs which drive MHI.  PBL
> famously does not drive MHI because PBL is encoded into hardware and
> extremely difficult to fix, so it is generally designed with the mantra
> of "simpler is more reliable".

Ok understood.

>
> Hopefully I didn't throw a wrench in things for you.  Just trying to
> save you some pain later.

Yes, that's perfectly fine and valid, so I'm going to rework that.

Regards,
Loic