2024-01-22 18:27:27

by Matthew Gerlach

[permalink] [raw]
Subject: [PATCH] fpga: dfl: afu: update initialization of port_hdr driver

Revision 2 of the Device Feature List (DFL) Port feature has
slightly different requirements than revision 1. Revision 2
does not need the port to reset at driver startup. In fact,
performing a port reset during driver initialization can cause
driver race conditions when the port is connected to a different
PCIe Physical Function (PF) than the management PF performing
the actual port reset.

Signed-off-by: Matthew Gerlach <[email protected]>
---
drivers/fpga/dfl-afu-main.c | 13 ++++++++++++-
1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/drivers/fpga/dfl-afu-main.c b/drivers/fpga/dfl-afu-main.c
index c0a75ca360d6..7d7f80cd264f 100644
--- a/drivers/fpga/dfl-afu-main.c
+++ b/drivers/fpga/dfl-afu-main.c
@@ -417,7 +417,18 @@ static const struct attribute_group port_hdr_group = {
static int port_hdr_init(struct platform_device *pdev,
struct dfl_feature *feature)
{
- port_reset(pdev);
+ void __iomem *base;
+ u8 rev;
+
+ base = dfl_get_feature_ioaddr_by_id(&pdev->dev, PORT_FEATURE_ID_HEADER);
+
+ rev = dfl_feature_revision(base);
+
+ if (rev < 2)
+ port_reset(pdev);
+
+ if (rev > 2)
+ dev_info(&pdev->dev, "unexpected port feature revision, %u\n", rev);

return 0;
}
--
2.34.1



2024-01-23 06:17:07

by Xu Yilun

[permalink] [raw]
Subject: Re: [PATCH] fpga: dfl: afu: update initialization of port_hdr driver

On Mon, Jan 22, 2024 at 09:24:33AM -0800, Matthew Gerlach wrote:
> Revision 2 of the Device Feature List (DFL) Port feature has
> slightly different requirements than revision 1. Revision 2
> does not need the port to reset at driver startup. In fact,

Please help illustrate what's the difference between Revision 1 & 2, and
why revision 2 needs not.

> performing a port reset during driver initialization can cause
> driver race conditions when the port is connected to a different

Please reorganize this part, in this description there seems be a
software racing bug and the patch is a workaround. But the fact is port
reset shouldn't been done for a new HW.

BTW: Is there a way to tell whether the port is connected to a different
PF? Any guarantee that revision 3, 4 ... would need a port reset or not?

Thanks,
Yilun

> PCIe Physical Function (PF) than the management PF performing
> the actual port reset.
>
> Signed-off-by: Matthew Gerlach <[email protected]>
> ---
> drivers/fpga/dfl-afu-main.c | 13 ++++++++++++-
> 1 file changed, 12 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/fpga/dfl-afu-main.c b/drivers/fpga/dfl-afu-main.c
> index c0a75ca360d6..7d7f80cd264f 100644
> --- a/drivers/fpga/dfl-afu-main.c
> +++ b/drivers/fpga/dfl-afu-main.c
> @@ -417,7 +417,18 @@ static const struct attribute_group port_hdr_group = {
> static int port_hdr_init(struct platform_device *pdev,
> struct dfl_feature *feature)
> {
> - port_reset(pdev);
> + void __iomem *base;
> + u8 rev;
> +
> + base = dfl_get_feature_ioaddr_by_id(&pdev->dev, PORT_FEATURE_ID_HEADER);
> +
> + rev = dfl_feature_revision(base);
> +
> + if (rev < 2)
> + port_reset(pdev);
> +
> + if (rev > 2)
> + dev_info(&pdev->dev, "unexpected port feature revision, %u\n", rev);

Remove the print. It is indicating an error but the function returns OK.

Thanks,
Yilun

>
> return 0;
> }
> --
> 2.34.1
>
>

2024-01-24 20:56:08

by Matthew Gerlach

[permalink] [raw]
Subject: Re: [PATCH] fpga: dfl: afu: update initialization of port_hdr driver



On Tue, 23 Jan 2024, Xu Yilun wrote:

> On Mon, Jan 22, 2024 at 09:24:33AM -0800, Matthew Gerlach wrote:
>> Revision 2 of the Device Feature List (DFL) Port feature has
>> slightly different requirements than revision 1. Revision 2
>> does not need the port to reset at driver startup. In fact,
>
> Please help illustrate what's the difference between Revision 1 & 2, and
> why revision 2 needs not.

I will update the commit message to clarify the differences between
revision 1 and 2.

>
>> performing a port reset during driver initialization can cause
>> driver race conditions when the port is connected to a different
>
> Please reorganize this part, in this description there seems be a
> software racing bug and the patch is a workaround. But the fact is port
> reset shouldn't been done for a new HW.

Reorganizing the commit message a bit will help to clarify why port reset
should not be performed during driver initialization with revision 2 of
the hardware.

>
> BTW: Is there a way to tell whether the port is connected to a different
> PF? Any guarantee that revision 3, 4 ... would need a port reset or not?

The use of revision 2 of the port_hdr IP block indicates that the port can
be connected multiple PFs, but there is nothing explicitly stating which
PFs the port is connected to.

It is hard to predict the requirements and implementation of a future
revision of an IP block. If a requirement of a future revision is to work
with existing software, then the future revision would not require a port
reset at driver initialization.

>
> Thanks,
> Yilun
>
>> PCIe Physical Function (PF) than the management PF performing
>> the actual port reset.
>>
>> Signed-off-by: Matthew Gerlach <[email protected]>
>> ---
>> drivers/fpga/dfl-afu-main.c | 13 ++++++++++++-
>> 1 file changed, 12 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/fpga/dfl-afu-main.c b/drivers/fpga/dfl-afu-main.c
>> index c0a75ca360d6..7d7f80cd264f 100644
>> --- a/drivers/fpga/dfl-afu-main.c
>> +++ b/drivers/fpga/dfl-afu-main.c
>> @@ -417,7 +417,18 @@ static const struct attribute_group port_hdr_group = {
>> static int port_hdr_init(struct platform_device *pdev,
>> struct dfl_feature *feature)
>> {
>> - port_reset(pdev);
>> + void __iomem *base;
>> + u8 rev;
>> +
>> + base = dfl_get_feature_ioaddr_by_id(&pdev->dev, PORT_FEATURE_ID_HEADER);
>> +
>> + rev = dfl_feature_revision(base);
>> +
>> + if (rev < 2)
>> + port_reset(pdev);
>> +
>> + if (rev > 2)
>> + dev_info(&pdev->dev, "unexpected port feature revision, %u\n", rev);
>
> Remove the print. It is indicating an error but the function returns OK.

The message is intended to be informational, but I'll remove it because it
could be confusing.

>
> Thanks,
> Yilun

Thanks for the feedback.

>
>>
>> return 0;
>> }
>> --
>> 2.34.1
>>
>>
>

2024-01-30 11:00:04

by Xu Yilun

[permalink] [raw]
Subject: Re: [PATCH] fpga: dfl: afu: update initialization of port_hdr driver

On Wed, Jan 24, 2024 at 11:40:05AM -0800, [email protected] wrote:
>
>
> On Tue, 23 Jan 2024, Xu Yilun wrote:
>
> > On Mon, Jan 22, 2024 at 09:24:33AM -0800, Matthew Gerlach wrote:
> > > Revision 2 of the Device Feature List (DFL) Port feature has
> > > slightly different requirements than revision 1. Revision 2
> > > does not need the port to reset at driver startup. In fact,
> >
> > Please help illustrate what's the difference between Revision 1 & 2, and
> > why revision 2 needs not.
>
> I will update the commit message to clarify the differences between revision
> 1 and 2.
>
> >
> > > performing a port reset during driver initialization can cause
> > > driver race conditions when the port is connected to a different
> >
> > Please reorganize this part, in this description there seems be a
> > software racing bug and the patch is a workaround. But the fact is port
> > reset shouldn't been done for a new HW.
>
> Reorganizing the commit message a bit will help to clarify why port reset
> should not be performed during driver initialization with revision 2 of the
> hardware.
>
> >
> > BTW: Is there a way to tell whether the port is connected to a different
> > PF? Any guarantee that revision 3, 4 ... would need a port reset or not?
>
> The use of revision 2 of the port_hdr IP block indicates that the port can
> be connected multiple PFs, but there is nothing explicitly stating which PFs

Sorry, I mean any specific indicator other than enumerate the revision
number? As you said below, checking revision number may not make further
things right, then you need to amend code each time.

Thanks,
Yilun

> the port is connected to.
>
> It is hard to predict the requirements and implementation of a future
> revision of an IP block. If a requirement of a future revision is to work
> with existing software, then the future revision would not require a port
> reset at driver initialization.
>

2024-01-30 18:30:11

by Matthew Gerlach

[permalink] [raw]
Subject: Re: [PATCH] fpga: dfl: afu: update initialization of port_hdr driver



On Tue, 30 Jan 2024, Xu Yilun wrote:

> On Wed, Jan 24, 2024 at 11:40:05AM -0800, [email protected] wrote:
>>
>>
>> On Tue, 23 Jan 2024, Xu Yilun wrote:
>>
>>> On Mon, Jan 22, 2024 at 09:24:33AM -0800, Matthew Gerlach wrote:
>>>> Revision 2 of the Device Feature List (DFL) Port feature has
>>>> slightly different requirements than revision 1. Revision 2
>>>> does not need the port to reset at driver startup. In fact,
>>>
>>> Please help illustrate what's the difference between Revision 1 & 2, and
>>> why revision 2 needs not.
>>
>> I will update the commit message to clarify the differences between revision
>> 1 and 2.
>>
>>>
>>>> performing a port reset during driver initialization can cause
>>>> driver race conditions when the port is connected to a different
>>>
>>> Please reorganize this part, in this description there seems be a
>>> software racing bug and the patch is a workaround. But the fact is port
>>> reset shouldn't been done for a new HW.
>>
>> Reorganizing the commit message a bit will help to clarify why port reset
>> should not be performed during driver initialization with revision 2 of the
>> hardware.
>>
>>>
>>> BTW: Is there a way to tell whether the port is connected to a different
>>> PF? Any guarantee that revision 3, 4 ... would need a port reset or not?
>>
>> The use of revision 2 of the port_hdr IP block indicates that the port can
>> be connected multiple PFs, but there is nothing explicitly stating which PFs
>
> Sorry, I mean any specific indicator other than enumerate the revision
> number? As you said below, checking revision number may not make further
> things right, then you need to amend code each time.

Using a revision number to indicate the level of functionality for a
particular IP block seems to be a widely used approach. What other
indicator of functionality level did you have in mind?

The revision number of an IP block would change when new functionality is
added to an IP block or the behavior of the IP block changes. It would be
expected that SW might need to change in order to use the new functionality
or to handle the change in behavior of the IP block. Ideally the
new revision of an IP block would be compatible with existing SW, but that
cannot be guaranteed.

Thanks,
Matthew

>
> Thanks,
> Yilun
>
>> the port is connected to.
>>
>> It is hard to predict the requirements and implementation of a future
>> revision of an IP block. If a requirement of a future revision is to work
>> with existing software, then the future revision would not require a port
>> reset at driver initialization.
>>
>

2024-01-31 06:16:11

by Xu Yilun

[permalink] [raw]
Subject: Re: [PATCH] fpga: dfl: afu: update initialization of port_hdr driver

On Tue, Jan 30, 2024 at 09:13:56AM -0800, [email protected] wrote:
>
>
> On Tue, 30 Jan 2024, Xu Yilun wrote:
>
> > On Wed, Jan 24, 2024 at 11:40:05AM -0800, [email protected] wrote:
> > >
> > >
> > > On Tue, 23 Jan 2024, Xu Yilun wrote:
> > >
> > > > On Mon, Jan 22, 2024 at 09:24:33AM -0800, Matthew Gerlach wrote:
> > > > > Revision 2 of the Device Feature List (DFL) Port feature has
> > > > > slightly different requirements than revision 1. Revision 2
> > > > > does not need the port to reset at driver startup. In fact,
> > > >
> > > > Please help illustrate what's the difference between Revision 1 & 2, and
> > > > why revision 2 needs not.
> > >
> > > I will update the commit message to clarify the differences between revision
> > > 1 and 2.
> > >
> > > >
> > > > > performing a port reset during driver initialization can cause
> > > > > driver race conditions when the port is connected to a different
> > > >
> > > > Please reorganize this part, in this description there seems be a
> > > > software racing bug and the patch is a workaround. But the fact is port
> > > > reset shouldn't been done for a new HW.
> > >
> > > Reorganizing the commit message a bit will help to clarify why port reset
> > > should not be performed during driver initialization with revision 2 of the
> > > hardware.
> > >
> > > >
> > > > BTW: Is there a way to tell whether the port is connected to a different
> > > > PF? Any guarantee that revision 3, 4 ... would need a port reset or not?
> > >
> > > The use of revision 2 of the port_hdr IP block indicates that the port can
> > > be connected multiple PFs, but there is nothing explicitly stating which PFs
> >
> > Sorry, I mean any specific indicator other than enumerate the revision
> > number? As you said below, checking revision number may not make further
> > things right, then you need to amend code each time.
>
> Using a revision number to indicate the level of functionality for a
> particular IP block seems to be a widely used approach. What other indicator

If you still want to make the existing driver work, some capability indication
would have more compatibility. That's more reasonable approach. Or you
need to change existing behavior for each new revision, that's not
actually widely used.

> of functionality level did you have in mind?

I'm not trying to make the design. You tell me.

If finally no indicator could be used, we have to use revision number. That's
OK but make SW work harder, so I'm asking if anything could be done to
avoid that.

>
> The revision number of an IP block would change when new functionality is
> added to an IP block or the behavior of the IP block changes. It would be
> expected that SW might need to change in order to use the new functionality
> or to handle the change in behavior of the IP block. Ideally the new
> revision of an IP block would be compatible with existing SW, but that
> cannot be guaranteed.

People make the IP block, and be compatible should be the concern if it
want upstream support.

Thanks,
Yilun

>
> Thanks,
> Matthew
>
> >
> > Thanks,
> > Yilun
> >
> > > the port is connected to.
> > >
> > > It is hard to predict the requirements and implementation of a future
> > > revision of an IP block. If a requirement of a future revision is to work
> > > with existing software, then the future revision would not require a port
> > > reset at driver initialization.
> > >
> >

2024-02-01 02:21:14

by Matthew Gerlach

[permalink] [raw]
Subject: Re: [PATCH] fpga: dfl: afu: update initialization of port_hdr driver



On Wed, 31 Jan 2024, Xu Yilun wrote:

> On Tue, Jan 30, 2024 at 09:13:56AM -0800, [email protected] wrote:
>>
>>
>> On Tue, 30 Jan 2024, Xu Yilun wrote:
>>
>>> On Wed, Jan 24, 2024 at 11:40:05AM -0800, [email protected] wrote:
>>>>
>>>>
>>>> On Tue, 23 Jan 2024, Xu Yilun wrote:
>>>>
>>>>> On Mon, Jan 22, 2024 at 09:24:33AM -0800, Matthew Gerlach wrote:
>>>>>> Revision 2 of the Device Feature List (DFL) Port feature has
>>>>>> slightly different requirements than revision 1. Revision 2
>>>>>> does not need the port to reset at driver startup. In fact,
>>>>>
>>>>> Please help illustrate what's the difference between Revision 1 & 2, and
>>>>> why revision 2 needs not.
>>>>
>>>> I will update the commit message to clarify the differences between revision
>>>> 1 and 2.
>>>>
>>>>>
>>>>>> performing a port reset during driver initialization can cause
>>>>>> driver race conditions when the port is connected to a different
>>>>>
>>>>> Please reorganize this part, in this description there seems be a
>>>>> software racing bug and the patch is a workaround. But the fact is port
>>>>> reset shouldn't been done for a new HW.
>>>>
>>>> Reorganizing the commit message a bit will help to clarify why port reset
>>>> should not be performed during driver initialization with revision 2 of the
>>>> hardware.
>>>>
>>>>>
>>>>> BTW: Is there a way to tell whether the port is connected to a different
>>>>> PF? Any guarantee that revision 3, 4 ... would need a port reset or not?
>>>>
>>>> The use of revision 2 of the port_hdr IP block indicates that the port can
>>>> be connected multiple PFs, but there is nothing explicitly stating which PFs
>>>
>>> Sorry, I mean any specific indicator other than enumerate the revision
>>> number? As you said below, checking revision number may not make further
>>> things right, then you need to amend code each time.
>>
>> Using a revision number to indicate the level of functionality for a
>> particular IP block seems to be a widely used approach. What other indicator
>
> If you still want to make the existing driver work, some capability indication
> would have more compatibility. That's more reasonable approach. Or you
> need to change existing behavior for each new revision, that's not
> actually widely used.

I understand some capability indication would be better for compatibility
implementation. A revision number change is not as explicit or precise as
capability lists.

>
>> of functionality level did you have in mind?
>
> I'm not trying to make the design. You tell me.

One could use parameter blocks introduced in version 1 of the Device
Feature Header (DFH), or capability registers could be added the IP block.
In this particular case it seems the least impact to upstreamed software is
to keep the DFH and the register map unchanged, except for an incremented
revision number field.

>
> If finally no indicator could be used, we have to use revision number. That's
> OK but make SW work harder, so I'm asking if anything could be done to
> avoid that.

In this case, I don't think anything else can be done without bigger
impacts to the SW.

>
>>
>> The revision number of an IP block would change when new functionality is
>> added to an IP block or the behavior of the IP block changes. It would be
>> expected that SW might need to change in order to use the new functionality
>> or to handle the change in behavior of the IP block. Ideally the new
>> revision of an IP block would be compatible with existing SW, but that
>> cannot be guaranteed.
>
> People make the IP block, and be compatible should be the concern if it
> want upstream support.

Agreed, and making sure some capability mechanism exists when an IP is
created would be a great start.

Thanks,
Matthew

>
> Thanks,
> Yilun
>
>>
>> Thanks,
>> Matthew
>>
>>>
>>> Thanks,
>>> Yilun
>>>
>>>> the port is connected to.
>>>>
>>>> It is hard to predict the requirements and implementation of a future
>>>> revision of an IP block. If a requirement of a future revision is to work
>>>> with existing software, then the future revision would not require a port
>>>> reset at driver initialization.
>>>>
>>>
>
>

2024-02-05 02:55:57

by Xu Yilun

[permalink] [raw]
Subject: Re: [PATCH] fpga: dfl: afu: update initialization of port_hdr driver

On Wed, Jan 31, 2024 at 03:53:23PM -0800, [email protected] wrote:
>
>
> On Wed, 31 Jan 2024, Xu Yilun wrote:
>
> > On Tue, Jan 30, 2024 at 09:13:56AM -0800, [email protected] wrote:
> > >
> > >
> > > On Tue, 30 Jan 2024, Xu Yilun wrote:
> > >
> > > > On Wed, Jan 24, 2024 at 11:40:05AM -0800, [email protected] wrote:
> > > > >
> > > > >
> > > > > On Tue, 23 Jan 2024, Xu Yilun wrote:
> > > > >
> > > > > > On Mon, Jan 22, 2024 at 09:24:33AM -0800, Matthew Gerlach wrote:
> > > > > > > Revision 2 of the Device Feature List (DFL) Port feature has
> > > > > > > slightly different requirements than revision 1. Revision 2
> > > > > > > does not need the port to reset at driver startup. In fact,
> > > > > >
> > > > > > Please help illustrate what's the difference between Revision 1 & 2, and
> > > > > > why revision 2 needs not.
> > > > >
> > > > > I will update the commit message to clarify the differences between revision
> > > > > 1 and 2.
> > > > >
> > > > > >
> > > > > > > performing a port reset during driver initialization can cause
> > > > > > > driver race conditions when the port is connected to a different
> > > > > >
> > > > > > Please reorganize this part, in this description there seems be a
> > > > > > software racing bug and the patch is a workaround. But the fact is port
> > > > > > reset shouldn't been done for a new HW.
> > > > >
> > > > > Reorganizing the commit message a bit will help to clarify why port reset
> > > > > should not be performed during driver initialization with revision 2 of the
> > > > > hardware.
> > > > >
> > > > > >
> > > > > > BTW: Is there a way to tell whether the port is connected to a different
> > > > > > PF? Any guarantee that revision 3, 4 ... would need a port reset or not?
> > > > >
> > > > > The use of revision 2 of the port_hdr IP block indicates that the port can
> > > > > be connected multiple PFs, but there is nothing explicitly stating which PFs
> > > >
> > > > Sorry, I mean any specific indicator other than enumerate the revision
> > > > number? As you said below, checking revision number may not make further
> > > > things right, then you need to amend code each time.
> > >
> > > Using a revision number to indicate the level of functionality for a
> > > particular IP block seems to be a widely used approach. What other indicator
> >
> > If you still want to make the existing driver work, some capability indication
> > would have more compatibility. That's more reasonable approach. Or you
> > need to change existing behavior for each new revision, that's not
> > actually widely used.
>
> I understand some capability indication would be better for compatibility
> implementation. A revision number change is not as explicit or precise as
> capability lists.
>
> >
> > > of functionality level did you have in mind?
> >
> > I'm not trying to make the design. You tell me.
>
> One could use parameter blocks introduced in version 1 of the Device Feature
> Header (DFH), or capability registers could be added the IP block.
> In this particular case it seems the least impact to upstreamed software is
> to keep the DFH and the register map unchanged, except for an incremented
> revision number field.
>
> >
> > If finally no indicator could be used, we have to use revision number. That's
> > OK but make SW work harder, so I'm asking if anything could be done to
> > avoid that.
>
> In this case, I don't think anything else can be done without bigger impacts
> to the SW.

Changing the existing SW is not a problem, repeat the same change every time
is a problem. So if we make sure port reset is no longer needed after
version 1, then this patch is OK. Otherwise, please re-evaluate.

Thanks,
Yilun

>
> >
> > >
> > > The revision number of an IP block would change when new functionality is
> > > added to an IP block or the behavior of the IP block changes. It would be
> > > expected that SW might need to change in order to use the new functionality
> > > or to handle the change in behavior of the IP block. Ideally the new
> > > revision of an IP block would be compatible with existing SW, but that
> > > cannot be guaranteed.
> >
> > People make the IP block, and be compatible should be the concern if it
> > want upstream support.
>
> Agreed, and making sure some capability mechanism exists when an IP is
> created would be a great start.
>
> Thanks,
> Matthew
>
> >
> > Thanks,
> > Yilun
> >
> > >
> > > Thanks,
> > > Matthew
> > >
> > > >
> > > > Thanks,
> > > > Yilun
> > > >
> > > > > the port is connected to.
> > > > >
> > > > > It is hard to predict the requirements and implementation of a future
> > > > > revision of an IP block. If a requirement of a future revision is to work
> > > > > with existing software, then the future revision would not require a port
> > > > > reset at driver initialization.
> > > > >
> > > >
> >
> >

2024-02-05 23:31:16

by Matthew Gerlach

[permalink] [raw]
Subject: Re: [PATCH] fpga: dfl: afu: update initialization of port_hdr driver



On Mon, 5 Feb 2024, Xu Yilun wrote:

> On Wed, Jan 31, 2024 at 03:53:23PM -0800, [email protected] wrote:
>>
>>
>> On Wed, 31 Jan 2024, Xu Yilun wrote:
>>
>>> On Tue, Jan 30, 2024 at 09:13:56AM -0800, [email protected] wrote:
>>>>
>>>>
>>>> On Tue, 30 Jan 2024, Xu Yilun wrote:
>>>>
>>>>> On Wed, Jan 24, 2024 at 11:40:05AM -0800, [email protected] wrote:
>>>>>>
>>>>>>
>>>>>> On Tue, 23 Jan 2024, Xu Yilun wrote:
>>>>>>
>>>>>>> On Mon, Jan 22, 2024 at 09:24:33AM -0800, Matthew Gerlach wrote:
>>>>>>>> Revision 2 of the Device Feature List (DFL) Port feature has
>>>>>>>> slightly different requirements than revision 1. Revision 2
>>>>>>>> does not need the port to reset at driver startup. In fact,
>>>>>>>
>>>>>>> Please help illustrate what's the difference between Revision 1 & 2, and
>>>>>>> why revision 2 needs not.
>>>>>>
>>>>>> I will update the commit message to clarify the differences between revision
>>>>>> 1 and 2.
>>>>>>
>>>>>>>
>>>>>>>> performing a port reset during driver initialization can cause
>>>>>>>> driver race conditions when the port is connected to a different
>>>>>>>
>>>>>>> Please reorganize this part, in this description there seems be a
>>>>>>> software racing bug and the patch is a workaround. But the fact is port
>>>>>>> reset shouldn't been done for a new HW.
>>>>>>
>>>>>> Reorganizing the commit message a bit will help to clarify why port reset
>>>>>> should not be performed during driver initialization with revision 2 of the
>>>>>> hardware.
>>>>>>
>>>>>>>
>>>>>>> BTW: Is there a way to tell whether the port is connected to a different
>>>>>>> PF? Any guarantee that revision 3, 4 ... would need a port reset or not?
>>>>>>
>>>>>> The use of revision 2 of the port_hdr IP block indicates that the port can
>>>>>> be connected multiple PFs, but there is nothing explicitly stating which PFs
>>>>>
>>>>> Sorry, I mean any specific indicator other than enumerate the revision
>>>>> number? As you said below, checking revision number may not make further
>>>>> things right, then you need to amend code each time.
>>>>
>>>> Using a revision number to indicate the level of functionality for a
>>>> particular IP block seems to be a widely used approach. What other indicator
>>>
>>> If you still want to make the existing driver work, some capability indication
>>> would have more compatibility. That's more reasonable approach. Or you
>>> need to change existing behavior for each new revision, that's not
>>> actually widely used.
>>
>> I understand some capability indication would be better for compatibility
>> implementation. A revision number change is not as explicit or precise as
>> capability lists.
>>
>>>
>>>> of functionality level did you have in mind?
>>>
>>> I'm not trying to make the design. You tell me.
>>
>> One could use parameter blocks introduced in version 1 of the Device Feature
>> Header (DFH), or capability registers could be added the IP block.
>> In this particular case it seems the least impact to upstreamed software is
>> to keep the DFH and the register map unchanged, except for an incremented
>> revision number field.
>>
>>>
>>> If finally no indicator could be used, we have to use revision number. That's
>>> OK but make SW work harder, so I'm asking if anything could be done to
>>> avoid that.
>>
>> In this case, I don't think anything else can be done without bigger impacts
>> to the SW.
>
> Changing the existing SW is not a problem, repeat the same change every time
> is a problem. So if we make sure port reset is no longer needed after
> version 1, then this patch is OK. Otherwise, please re-evaluate.

The initial port reset will no longer be after version 1. The requirement
for the initial state of the logic in side the port to be valid will remain.

Thanks,
Matthew

>
> Thanks,
> Yilun
>
>>
>>>
>>>>
>>>> The revision number of an IP block would change when new functionality is
>>>> added to an IP block or the behavior of the IP block changes. It would be
>>>> expected that SW might need to change in order to use the new functionality
>>>> or to handle the change in behavior of the IP block. Ideally the new
>>>> revision of an IP block would be compatible with existing SW, but that
>>>> cannot be guaranteed.
>>>
>>> People make the IP block, and be compatible should be the concern if it
>>> want upstream support.
>>
>> Agreed, and making sure some capability mechanism exists when an IP is
>> created would be a great start.
>>
>> Thanks,
>> Matthew
>>
>>>
>>> Thanks,
>>> Yilun
>>>
>>>>
>>>> Thanks,
>>>> Matthew
>>>>
>>>>>
>>>>> Thanks,
>>>>> Yilun
>>>>>
>>>>>> the port is connected to.
>>>>>>
>>>>>> It is hard to predict the requirements and implementation of a future
>>>>>> revision of an IP block. If a requirement of a future revision is to work
>>>>>> with existing software, then the future revision would not require a port
>>>>>> reset at driver initialization.
>>>>>>
>>>>>
>>>
>>>
>