Coccinelle Archive on lore.kernel.org
 help / color / Atom feed
* Re: [Cocci] Proposal for a new checkpatch check; matching _set_drvdata() & _get_drvdata()
       [not found] ` <CAHp75VcT5hZH6m0Dri1h_EFjc7=4+1XoE7sRuQyfO75k9A0GKA@mail.gmail.com>
@ 2020-11-20  0:11   ` Joe Perches
  2020-11-20 10:47     ` Julia Lawall
  0 siblings, 1 reply; 7+ messages in thread
From: Joe Perches @ 2020-11-20  0:11 UTC (permalink / raw)
  To: Andy Shevchenko, Alexandru Ardelean
  Cc: Robo Bot, Julia Lawall, Alexandru Ardelean, LKML, cocci

On Thu, 2020-11-19 at 17:16 +0200, Andy Shevchenko wrote:
> On Thu, Nov 19, 2020 at 4:09 PM Alexandru Ardelean
> <ardeleanalex@gmail.com> wrote:
> > 
> > Hey,
> > 
> > So, I stumbled on a new check that could be added to checkpatch.
> > Since it's in Perl, I'm reluctant to try it.
> > 
> > Seems many drivers got to a point where they now call (let's say)
> > spi_set_drvdata(), but never access that information via
> > spi_get_drvdata().
> > Reasons for this seem to be:
> > 1. They got converted to device-managed functions and there is no
> > longer a remove hook to require the _get_drvdata() access
> > 2. They look like they were copied from a driver that had a
> > _set_drvdata() and when the code got finalized, the _set_drvdata() was
> > omitted
> > 
> > There are a few false positives that I can notice at a quick look,
> > like the data being set via some xxx_set_drvdata() and retrieved via a
> > dev_get_drvdata().
> 
> I can say quite a few. And this makes a difference.
> So, basically all drivers that are using PM callbacks would rather use
> dev_get_drvdata() rather than bus specific.
> 
> > I think checkpatch reporting these as well would be acceptable simply
> > from a reviewability perspective.
> > 
> > I did a shell script to quickly check these. See below.
> > It's pretty badly written but it is enough for me to gather a list.
> > And I wrote it in 5 minutes :P
> > I initially noticed this in some IIO drivers, and then I suspected
> > that this may be more widespread.
> 
> It seems more suitable for coccinelle.

To me as well.


_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [Cocci] Proposal for a new checkpatch check; matching _set_drvdata() & _get_drvdata()
  2020-11-20  0:11   ` [Cocci] Proposal for a new checkpatch check; matching _set_drvdata() & _get_drvdata() Joe Perches
@ 2020-11-20 10:47     ` Julia Lawall
  2020-11-20 11:54       ` Alexandru Ardelean
  0 siblings, 1 reply; 7+ messages in thread
From: Julia Lawall @ 2020-11-20 10:47 UTC (permalink / raw)
  To: Joe Perches
  Cc: LKML, Alexandru Ardelean, Andy Shevchenko, Robo Bot,
	Alexandru Ardelean, cocci



On Thu, 19 Nov 2020, Joe Perches wrote:

> On Thu, 2020-11-19 at 17:16 +0200, Andy Shevchenko wrote:
> > On Thu, Nov 19, 2020 at 4:09 PM Alexandru Ardelean
> > <ardeleanalex@gmail.com> wrote:
> > >
> > > Hey,
> > >
> > > So, I stumbled on a new check that could be added to checkpatch.
> > > Since it's in Perl, I'm reluctant to try it.
> > >
> > > Seems many drivers got to a point where they now call (let's say)
> > > spi_set_drvdata(), but never access that information via
> > > spi_get_drvdata().
> > > Reasons for this seem to be:
> > > 1. They got converted to device-managed functions and there is no
> > > longer a remove hook to require the _get_drvdata() access
> > > 2. They look like they were copied from a driver that had a
> > > _set_drvdata() and when the code got finalized, the _set_drvdata() was
> > > omitted
> > >
> > > There are a few false positives that I can notice at a quick look,
> > > like the data being set via some xxx_set_drvdata() and retrieved via a
> > > dev_get_drvdata().
> >
> > I can say quite a few. And this makes a difference.
> > So, basically all drivers that are using PM callbacks would rather use
> > dev_get_drvdata() rather than bus specific.
> >
> > > I think checkpatch reporting these as well would be acceptable simply
> > > from a reviewability perspective.
> > >
> > > I did a shell script to quickly check these. See below.
> > > It's pretty badly written but it is enough for me to gather a list.
> > > And I wrote it in 5 minutes :P
> > > I initially noticed this in some IIO drivers, and then I suspected
> > > that this may be more widespread.
> >
> > It seems more suitable for coccinelle.
>
> To me as well.

To me as well, since it seems to involve nonlocal information.

I'm not sure to understand the original shell script. Is there
something interesting about pci_set_drvdata?

julia
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [Cocci] Proposal for a new checkpatch check; matching _set_drvdata() & _get_drvdata()
  2020-11-20 10:47     ` Julia Lawall
@ 2020-11-20 11:54       ` Alexandru Ardelean
  2020-11-20 11:57         ` Julia Lawall
  2020-11-20 13:16         ` Lars-Peter Clausen
  0 siblings, 2 replies; 7+ messages in thread
From: Alexandru Ardelean @ 2020-11-20 11:54 UTC (permalink / raw)
  To: Julia Lawall
  Cc: LKML, Robo Bot, Andy Shevchenko, Joe Perches, Alexandru Ardelean, cocci

On Fri, Nov 20, 2020 at 12:47 PM Julia Lawall <julia.lawall@inria.fr> wrote:
>
>
>
> On Thu, 19 Nov 2020, Joe Perches wrote:
>
> > On Thu, 2020-11-19 at 17:16 +0200, Andy Shevchenko wrote:
> > > On Thu, Nov 19, 2020 at 4:09 PM Alexandru Ardelean
> > > <ardeleanalex@gmail.com> wrote:
> > > >
> > > > Hey,
> > > >
> > > > So, I stumbled on a new check that could be added to checkpatch.
> > > > Since it's in Perl, I'm reluctant to try it.
> > > >
> > > > Seems many drivers got to a point where they now call (let's say)
> > > > spi_set_drvdata(), but never access that information via
> > > > spi_get_drvdata().
> > > > Reasons for this seem to be:
> > > > 1. They got converted to device-managed functions and there is no
> > > > longer a remove hook to require the _get_drvdata() access
> > > > 2. They look like they were copied from a driver that had a
> > > > _set_drvdata() and when the code got finalized, the _set_drvdata() was
> > > > omitted
> > > >
> > > > There are a few false positives that I can notice at a quick look,
> > > > like the data being set via some xxx_set_drvdata() and retrieved via a
> > > > dev_get_drvdata().
> > >
> > > I can say quite a few. And this makes a difference.
> > > So, basically all drivers that are using PM callbacks would rather use
> > > dev_get_drvdata() rather than bus specific.
> > >
> > > > I think checkpatch reporting these as well would be acceptable simply
> > > > from a reviewability perspective.
> > > >
> > > > I did a shell script to quickly check these. See below.
> > > > It's pretty badly written but it is enough for me to gather a list.
> > > > And I wrote it in 5 minutes :P
> > > > I initially noticed this in some IIO drivers, and then I suspected
> > > > that this may be more widespread.
> > >
> > > It seems more suitable for coccinelle.
> >
> > To me as well.
>
> To me as well, since it seems to involve nonlocal information.
>
> I'm not sure to understand the original shell script. Is there
> something interesting about pci_set_drvdata?

Ah, it's a stupid script I wrote in 5 minutes, so I did not bother to
make things smart.
In the text-matching I did in shell, there are some entries that come
from comments and docs.
It's only about 3-4 entries, so I just did a visual/manual ignore.

In essence:
The script searches for all strings that contain _set_drvdata.
The separators are whitespace.
It creates a list of all  xxxx_set_drvdata functions.
For each xxxx_set_drvdata function:
    It checks all files that have a xxxx_set_drvdata entry, but no
xxxx_get_drvdata

I piped this output into a file and started manually checking the drivers.
There is one [I forget which function] that is xxxx_set_drvdata() but
equivalent is xxxx_drvdata()

As Andy said, some precautions must be taken in places where
xxxx_set_drvdata() is called but dev_get_drvdata() is used.
Cases like PM suspend/resume calls.
And there may be some cases outside this context.


>
> julia
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [Cocci] Proposal for a new checkpatch check; matching _set_drvdata() & _get_drvdata()
  2020-11-20 11:54       ` Alexandru Ardelean
@ 2020-11-20 11:57         ` Julia Lawall
  2020-11-20 11:58           ` Alexandru Ardelean
  2020-11-20 13:16         ` Lars-Peter Clausen
  1 sibling, 1 reply; 7+ messages in thread
From: Julia Lawall @ 2020-11-20 11:57 UTC (permalink / raw)
  To: Alexandru Ardelean
  Cc: LKML, Robo Bot, Andy Shevchenko, Joe Perches, Alexandru Ardelean, cocci



On Fri, 20 Nov 2020, Alexandru Ardelean wrote:

> On Fri, Nov 20, 2020 at 12:47 PM Julia Lawall <julia.lawall@inria.fr> wrote:
> >
> >
> >
> > On Thu, 19 Nov 2020, Joe Perches wrote:
> >
> > > On Thu, 2020-11-19 at 17:16 +0200, Andy Shevchenko wrote:
> > > > On Thu, Nov 19, 2020 at 4:09 PM Alexandru Ardelean
> > > > <ardeleanalex@gmail.com> wrote:
> > > > >
> > > > > Hey,
> > > > >
> > > > > So, I stumbled on a new check that could be added to checkpatch.
> > > > > Since it's in Perl, I'm reluctant to try it.
> > > > >
> > > > > Seems many drivers got to a point where they now call (let's say)
> > > > > spi_set_drvdata(), but never access that information via
> > > > > spi_get_drvdata().
> > > > > Reasons for this seem to be:
> > > > > 1. They got converted to device-managed functions and there is no
> > > > > longer a remove hook to require the _get_drvdata() access
> > > > > 2. They look like they were copied from a driver that had a
> > > > > _set_drvdata() and when the code got finalized, the _set_drvdata() was
> > > > > omitted
> > > > >
> > > > > There are a few false positives that I can notice at a quick look,
> > > > > like the data being set via some xxx_set_drvdata() and retrieved via a
> > > > > dev_get_drvdata().
> > > >
> > > > I can say quite a few. And this makes a difference.
> > > > So, basically all drivers that are using PM callbacks would rather use
> > > > dev_get_drvdata() rather than bus specific.
> > > >
> > > > > I think checkpatch reporting these as well would be acceptable simply
> > > > > from a reviewability perspective.
> > > > >
> > > > > I did a shell script to quickly check these. See below.
> > > > > It's pretty badly written but it is enough for me to gather a list.
> > > > > And I wrote it in 5 minutes :P
> > > > > I initially noticed this in some IIO drivers, and then I suspected
> > > > > that this may be more widespread.
> > > >
> > > > It seems more suitable for coccinelle.
> > >
> > > To me as well.
> >
> > To me as well, since it seems to involve nonlocal information.
> >
> > I'm not sure to understand the original shell script. Is there
> > something interesting about pci_set_drvdata?
>
> Ah, it's a stupid script I wrote in 5 minutes, so I did not bother to
> make things smart.
> In the text-matching I did in shell, there are some entries that come
> from comments and docs.
> It's only about 3-4 entries, so I just did a visual/manual ignore.
>
> In essence:
> The script searches for all strings that contain _set_drvdata.
> The separators are whitespace.
> It creates a list of all  xxxx_set_drvdata functions.
> For each xxxx_set_drvdata function:
>     It checks all files that have a xxxx_set_drvdata entry, but no
> xxxx_get_drvdata

OK, but I have the impression that you want to ignore pci_set_drvdata for
some reason?  Or did I misunderstand?

julia

>
> I piped this output into a file and started manually checking the drivers.
> There is one [I forget which function] that is xxxx_set_drvdata() but
> equivalent is xxxx_drvdata()
>
> As Andy said, some precautions must be taken in places where
> xxxx_set_drvdata() is called but dev_get_drvdata() is used.
> Cases like PM suspend/resume calls.
> And there may be some cases outside this context.
>
>
> >
> > julia
>
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [Cocci] Proposal for a new checkpatch check; matching _set_drvdata() & _get_drvdata()
  2020-11-20 11:57         ` Julia Lawall
@ 2020-11-20 11:58           ` Alexandru Ardelean
  0 siblings, 0 replies; 7+ messages in thread
From: Alexandru Ardelean @ 2020-11-20 11:58 UTC (permalink / raw)
  To: Julia Lawall
  Cc: LKML, Robo Bot, Andy Shevchenko, Joe Perches, Alexandru Ardelean, cocci

On Fri, Nov 20, 2020 at 1:57 PM Julia Lawall <julia.lawall@inria.fr> wrote:
>
>
>
> On Fri, 20 Nov 2020, Alexandru Ardelean wrote:
>
> > On Fri, Nov 20, 2020 at 12:47 PM Julia Lawall <julia.lawall@inria.fr> wrote:
> > >
> > >
> > >
> > > On Thu, 19 Nov 2020, Joe Perches wrote:
> > >
> > > > On Thu, 2020-11-19 at 17:16 +0200, Andy Shevchenko wrote:
> > > > > On Thu, Nov 19, 2020 at 4:09 PM Alexandru Ardelean
> > > > > <ardeleanalex@gmail.com> wrote:
> > > > > >
> > > > > > Hey,
> > > > > >
> > > > > > So, I stumbled on a new check that could be added to checkpatch.
> > > > > > Since it's in Perl, I'm reluctant to try it.
> > > > > >
> > > > > > Seems many drivers got to a point where they now call (let's say)
> > > > > > spi_set_drvdata(), but never access that information via
> > > > > > spi_get_drvdata().
> > > > > > Reasons for this seem to be:
> > > > > > 1. They got converted to device-managed functions and there is no
> > > > > > longer a remove hook to require the _get_drvdata() access
> > > > > > 2. They look like they were copied from a driver that had a
> > > > > > _set_drvdata() and when the code got finalized, the _set_drvdata() was
> > > > > > omitted
> > > > > >
> > > > > > There are a few false positives that I can notice at a quick look,
> > > > > > like the data being set via some xxx_set_drvdata() and retrieved via a
> > > > > > dev_get_drvdata().
> > > > >
> > > > > I can say quite a few. And this makes a difference.
> > > > > So, basically all drivers that are using PM callbacks would rather use
> > > > > dev_get_drvdata() rather than bus specific.
> > > > >
> > > > > > I think checkpatch reporting these as well would be acceptable simply
> > > > > > from a reviewability perspective.
> > > > > >
> > > > > > I did a shell script to quickly check these. See below.
> > > > > > It's pretty badly written but it is enough for me to gather a list.
> > > > > > And I wrote it in 5 minutes :P
> > > > > > I initially noticed this in some IIO drivers, and then I suspected
> > > > > > that this may be more widespread.
> > > > >
> > > > > It seems more suitable for coccinelle.
> > > >
> > > > To me as well.
> > >
> > > To me as well, since it seems to involve nonlocal information.
> > >
> > > I'm not sure to understand the original shell script. Is there
> > > something interesting about pci_set_drvdata?
> >
> > Ah, it's a stupid script I wrote in 5 minutes, so I did not bother to
> > make things smart.
> > In the text-matching I did in shell, there are some entries that come
> > from comments and docs.
> > It's only about 3-4 entries, so I just did a visual/manual ignore.
> >
> > In essence:
> > The script searches for all strings that contain _set_drvdata.
> > The separators are whitespace.
> > It creates a list of all  xxxx_set_drvdata functions.
> > For each xxxx_set_drvdata function:
> >     It checks all files that have a xxxx_set_drvdata entry, but no
> > xxxx_get_drvdata
>
> OK, but I have the impression that you want to ignore pci_set_drvdata for
> some reason?  Or did I misunderstand?

Yes. See difficultly visible double quote :P
'  "pci_set_drvdata   '
Apologies for the confusion

        if [ "$fn" == '"pci_set_drvdata' ] ; then
                continue
        fi


>
> julia
>
> >
> > I piped this output into a file and started manually checking the drivers.
> > There is one [I forget which function] that is xxxx_set_drvdata() but
> > equivalent is xxxx_drvdata()
> >
> > As Andy said, some precautions must be taken in places where
> > xxxx_set_drvdata() is called but dev_get_drvdata() is used.
> > Cases like PM suspend/resume calls.
> > And there may be some cases outside this context.
> >
> >
> > >
> > > julia
> >
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [Cocci] Proposal for a new checkpatch check; matching _set_drvdata() & _get_drvdata()
  2020-11-20 11:54       ` Alexandru Ardelean
  2020-11-20 11:57         ` Julia Lawall
@ 2020-11-20 13:16         ` Lars-Peter Clausen
  2020-11-20 13:27           ` Alexandru Ardelean
  1 sibling, 1 reply; 7+ messages in thread
From: Lars-Peter Clausen @ 2020-11-20 13:16 UTC (permalink / raw)
  To: Alexandru Ardelean, Julia Lawall
  Cc: LKML, Robo Bot, Andy Shevchenko, Joe Perches, Alexandru Ardelean, cocci

On 11/20/20 12:54 PM, Alexandru Ardelean wrote:
> On Fri, Nov 20, 2020 at 12:47 PM Julia Lawall <julia.lawall@inria.fr> wrote:
>>
>>
>> On Thu, 19 Nov 2020, Joe Perches wrote:
>>
>>> On Thu, 2020-11-19 at 17:16 +0200, Andy Shevchenko wrote:
>>>> On Thu, Nov 19, 2020 at 4:09 PM Alexandru Ardelean
>>>> <ardeleanalex@gmail.com> wrote:
>>>>> Hey,
>>>>>
>>>>> So, I stumbled on a new check that could be added to checkpatch.
>>>>> Since it's in Perl, I'm reluctant to try it.
>>>>>
>>>>> Seems many drivers got to a point where they now call (let's say)
>>>>> spi_set_drvdata(), but never access that information via
>>>>> spi_get_drvdata().
>>>>> Reasons for this seem to be:
>>>>> 1. They got converted to device-managed functions and there is no
>>>>> longer a remove hook to require the _get_drvdata() access
>>>>> 2. They look like they were copied from a driver that had a
>>>>> _set_drvdata() and when the code got finalized, the _set_drvdata() was
>>>>> omitted
>>>>>
>>>>> There are a few false positives that I can notice at a quick look,
>>>>> like the data being set via some xxx_set_drvdata() and retrieved via a
>>>>> dev_get_drvdata().
>>>> I can say quite a few. And this makes a difference.
>>>> So, basically all drivers that are using PM callbacks would rather use
>>>> dev_get_drvdata() rather than bus specific.
>>>>
>>>>> I think checkpatch reporting these as well would be acceptable simply
>>>>> from a reviewability perspective.
>>>>>
>>>>> I did a shell script to quickly check these. See below.
>>>>> It's pretty badly written but it is enough for me to gather a list.
>>>>> And I wrote it in 5 minutes :P
>>>>> I initially noticed this in some IIO drivers, and then I suspected
>>>>> that this may be more widespread.
>>>> It seems more suitable for coccinelle.
>>> To me as well.
>> To me as well, since it seems to involve nonlocal information.
>>
>> I'm not sure to understand the original shell script. Is there
>> something interesting about pci_set_drvdata?
> Ah, it's a stupid script I wrote in 5 minutes, so I did not bother to
> make things smart.
> In the text-matching I did in shell, there are some entries that come
> from comments and docs.
> It's only about 3-4 entries, so I just did a visual/manual ignore.
>
> In essence:
> The script searches for all strings that contain _set_drvdata.
> The separators are whitespace.
> It creates a list of all  xxxx_set_drvdata functions.
> For each xxxx_set_drvdata function:
>      It checks all files that have a xxxx_set_drvdata entry, but no
> xxxx_get_drvdata
>
> I piped this output into a file and started manually checking the drivers.
> There is one [I forget which function] that is xxxx_set_drvdata() but
> equivalent is xxxx_drvdata()
>
> As Andy said, some precautions must be taken in places where
> xxxx_set_drvdata() is called but dev_get_drvdata() is used.
> Cases like PM suspend/resume calls.
> And there may be some cases outside this context.
>
Doing something like this with coccinelle is fairly easy.

But I'd be very cautious about putting such a script into the kernel. It 
will result in too many false positive drive-by patches. Such a script 
will not detect cases such as:

  * Driver is split over multiple files. One file does 
..._set_drvdata(), another does ..._get_drvdata().

  * Framework uses drvdata to exchange data with the driver. E.g driver 
is expected to call set_drvdata() and then the framework uses 
get_drvdata() to retrieve the data. This is not a very good pattern, but 
there are some palces int he kernel where this is used. I believe for 
example V4L2 uses this.

- Lars

_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [Cocci] Proposal for a new checkpatch check; matching _set_drvdata() & _get_drvdata()
  2020-11-20 13:16         ` Lars-Peter Clausen
@ 2020-11-20 13:27           ` Alexandru Ardelean
  0 siblings, 0 replies; 7+ messages in thread
From: Alexandru Ardelean @ 2020-11-20 13:27 UTC (permalink / raw)
  To: Lars-Peter Clausen
  Cc: LKML, Robo Bot, Andy Shevchenko, Joe Perches, Alexandru Ardelean, cocci

On Fri, Nov 20, 2020 at 3:16 PM Lars-Peter Clausen <lars@metafoo.de> wrote:
>
> On 11/20/20 12:54 PM, Alexandru Ardelean wrote:
> > On Fri, Nov 20, 2020 at 12:47 PM Julia Lawall <julia.lawall@inria.fr> wrote:
> >>
> >>
> >> On Thu, 19 Nov 2020, Joe Perches wrote:
> >>
> >>> On Thu, 2020-11-19 at 17:16 +0200, Andy Shevchenko wrote:
> >>>> On Thu, Nov 19, 2020 at 4:09 PM Alexandru Ardelean
> >>>> <ardeleanalex@gmail.com> wrote:
> >>>>> Hey,
> >>>>>
> >>>>> So, I stumbled on a new check that could be added to checkpatch.
> >>>>> Since it's in Perl, I'm reluctant to try it.
> >>>>>
> >>>>> Seems many drivers got to a point where they now call (let's say)
> >>>>> spi_set_drvdata(), but never access that information via
> >>>>> spi_get_drvdata().
> >>>>> Reasons for this seem to be:
> >>>>> 1. They got converted to device-managed functions and there is no
> >>>>> longer a remove hook to require the _get_drvdata() access
> >>>>> 2. They look like they were copied from a driver that had a
> >>>>> _set_drvdata() and when the code got finalized, the _set_drvdata() was
> >>>>> omitted
> >>>>>
> >>>>> There are a few false positives that I can notice at a quick look,
> >>>>> like the data being set via some xxx_set_drvdata() and retrieved via a
> >>>>> dev_get_drvdata().
> >>>> I can say quite a few. And this makes a difference.
> >>>> So, basically all drivers that are using PM callbacks would rather use
> >>>> dev_get_drvdata() rather than bus specific.
> >>>>
> >>>>> I think checkpatch reporting these as well would be acceptable simply
> >>>>> from a reviewability perspective.
> >>>>>
> >>>>> I did a shell script to quickly check these. See below.
> >>>>> It's pretty badly written but it is enough for me to gather a list.
> >>>>> And I wrote it in 5 minutes :P
> >>>>> I initially noticed this in some IIO drivers, and then I suspected
> >>>>> that this may be more widespread.
> >>>> It seems more suitable for coccinelle.
> >>> To me as well.
> >> To me as well, since it seems to involve nonlocal information.
> >>
> >> I'm not sure to understand the original shell script. Is there
> >> something interesting about pci_set_drvdata?
> > Ah, it's a stupid script I wrote in 5 minutes, so I did not bother to
> > make things smart.
> > In the text-matching I did in shell, there are some entries that come
> > from comments and docs.
> > It's only about 3-4 entries, so I just did a visual/manual ignore.
> >
> > In essence:
> > The script searches for all strings that contain _set_drvdata.
> > The separators are whitespace.
> > It creates a list of all  xxxx_set_drvdata functions.
> > For each xxxx_set_drvdata function:
> >      It checks all files that have a xxxx_set_drvdata entry, but no
> > xxxx_get_drvdata
> >
> > I piped this output into a file and started manually checking the drivers.
> > There is one [I forget which function] that is xxxx_set_drvdata() but
> > equivalent is xxxx_drvdata()
> >
> > As Andy said, some precautions must be taken in places where
> > xxxx_set_drvdata() is called but dev_get_drvdata() is used.
> > Cases like PM suspend/resume calls.
> > And there may be some cases outside this context.
> >
> Doing something like this with coccinelle is fairly easy.
>
> But I'd be very cautious about putting such a script into the kernel. It
> will result in too many false positive drive-by patches. Such a script
> will not detect cases such as:

Yeah, it would probably be a good idea to start with a few simple
checks, then scale it.
If we go for the existing _set_drvdata() / _get_drvdata() pair checks,
there is a risk of breaking things.

>
>   * Driver is split over multiple files. One file does
> ..._set_drvdata(), another does ..._get_drvdata().
>
>   * Framework uses drvdata to exchange data with the driver. E.g driver
> is expected to call set_drvdata() and then the framework uses
> get_drvdata() to retrieve the data. This is not a very good pattern, but
> there are some palces int he kernel where this is used. I believe for
> example V4L2 uses this.
>
> - Lars
>
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci

^ permalink raw reply	[flat|nested] 7+ messages in thread

end of thread, back to index

Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
     [not found] <CA+U=Dspy5+RE9agcLr6eY9DCMa1c5+++0JLeugMMBRXz4YLj1w@mail.gmail.com>
     [not found] ` <CAHp75VcT5hZH6m0Dri1h_EFjc7=4+1XoE7sRuQyfO75k9A0GKA@mail.gmail.com>
2020-11-20  0:11   ` [Cocci] Proposal for a new checkpatch check; matching _set_drvdata() & _get_drvdata() Joe Perches
2020-11-20 10:47     ` Julia Lawall
2020-11-20 11:54       ` Alexandru Ardelean
2020-11-20 11:57         ` Julia Lawall
2020-11-20 11:58           ` Alexandru Ardelean
2020-11-20 13:16         ` Lars-Peter Clausen
2020-11-20 13:27           ` Alexandru Ardelean

Coccinelle Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/cocci/0 cocci/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 cocci cocci/ https://lore.kernel.org/cocci \
		cocci@systeme.lip6.fr
	public-inbox-index cocci

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/fr.lip6.systeme.cocci


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git