Skip to contentSkip to conversation

pps: retrieve generator specific data from framework

PATCH In review [email protected] 2 participants 4 messages Jan 28, 2025 – Jan 29, 2025

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 pps: retrieve generator specific data from framework

From: Subramanian Mohan <[email protected]>

While adapting pps generator driver(tio generator as an example)to the new generator framework, As part of driver registration the pps_gen_device pointer is returned from framework. Due to which there is difficulty in getting generator driver data back in enable function. we won’t be able to use container_of macro as it results in static assert. we might end up in using static pointer. To avoid the same and get back the generator driver data back we are proposing generic approach to add drv_prv_data pointer inside the struct pps_gen_source_info.

Example TIO structure wrapped with pps_gen_device and usage.

struct pps_tio {
	/* Framework Related * /
	struct pps_gen_source_info pps_tio_source_info
	struct pps_gen_device *pps_gen;

	/* TIO Specific Data */
};

static int pps_tio_enable(struct pps_gen_device *pps_gen, bool enable) {

    /* Getting TIO data back */
    /* Note: drv_prv_data will be initialized in our init routine */
    struct pps_tio *tio = pps_gen->info.drv_prv_data;

    /* Access tio members here to set some of the parameters */

    return 0;
}

V1 -> V2:
* Updated reviewers.

---
 include/linux/pps_gen_kernel.h | 1 +
 1 file changed, 1 insertion(+)
include/linux/pps_gen_kernel.h +1 -0
index 022ea0ac4440..8484bb2828ef 100644
@@ -35,6 +35,7 @@ struct pps_gen_source_info {
 	int (*get_time)(struct pps_gen_device *pps_gen,
 					struct timespec64 *time);
 	int (*enable)(struct pps_gen_device *pps_gen, bool enable);
+	void *drv_prv_data;
 
 /* private: internal use only */
 	struct module *owner;
Signature · 1 line
2.35.3
#1 lore raw
Quoted 7 lines · On Tue, Jan 28, 2025 at 07:47:43PM +0530, [email protected] wrote:

From: Subramanian Mohan <[email protected]>

While adapting pps generator driver(tio generator as an example)to the new generator framework, As part of driver registration the pps_gen_device pointer is returned from framework. Due to which there is difficulty in getting generator driver data back in enable function. we won’t be able to use container_of macro as it results in static assert. we might end up

container_of()

Can you be more specific, what exactly happens?

Quoted 11 lines

in using static pointer. To avoid the same and get back the generator driver data back we are proposing generic approach to add drv_prv_data pointer inside the struct pps_gen_source_info.

Example TIO structure wrapped with pps_gen_device and usage.

struct pps_tio {
	/* Framework Related * /
	struct pps_gen_source_info pps_tio_source_info
	struct pps_gen_device *pps_gen;
Quoted 7 lines

};

static int pps_tio_enable(struct pps_gen_device *pps_gen, bool enable) {

    /* Getting TIO data back */
    /* Note: drv_prv_data will be initialized in our init routine */
    struct pps_tio *tio = pps_gen->info.drv_prv_data;

So, what's wrong with the container_of() against pps_gen->info? We have tons of code in the kernel that does it.

Quoted 4 lines
    /* Access tio members here to set some of the parameters */

    return 0;
}

Okay, looking into the implementation I see what's the issue (but it doesn't mean that commit message is good here, you still need to explain better why you can't get the correct address).

The problem is that info is got fully copied. Perhaps we should rely on the fact that it always be provided? I dunno any data structure that defines callbacks and other "info" material should be considered as temporary storage (on stack), it makes not much sense.

...

Quoted 2 lines

V1 -> V2:
* Updated reviewers.

Please, use...

Quoted 2 lines

---

...this place for the comments and changelogs.

...

include/linux/pps_gen_kernel.h struct pps_gen_source_info { · line 38

Quoted 1 line
include/linux/pps_gen_kernel.h +1 -0
+	void *drv_prv_data;

If really needed we should probably use a better name, like "private".

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#2 lore raw

HI Andy,

Quoted 25 lines

-----Original Message-----
From: Andy Shevchenko <[email protected]>
Sent: Wednesday, January 29, 2025 12:24 PM
To: Mohan, Subramanian <[email protected]>
Cc: [email protected]; [email protected]; akpm@linux- foundation.org; [email protected]; [email protected]; Hall, Christopher S <[email protected]>; [email protected]; Dong, Eddie <[email protected]>; N, Pandith <[email protected]>; T R, Thejesh Reddy <[email protected]>; Zage, David <[email protected]>; Chinnadurai, Srinivasan <[email protected]>; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH v2 1/1] pps: retrieve generator specific data from framework

Quoted 8 lines · On Tue, Jan 28, 2025 at 07:47:43PM +0530, [email protected] wrote:

From: Subramanian Mohan <[email protected]>

While adapting pps generator driver(tio generator as an example)to the new generator framework, As part of driver registration the pps_gen_device pointer is returned from framework. Due to which there is difficulty in getting generator driver data back in enable function. we won't be able to use container_of macro as it results in static assert. we might end up

Quoted 3 lines

container_of()

Can you be more specific, what exactly happens?

Let me provide some more info:

/* PPS Gen Device main struct*/

struct pps_gen_device {
	struct pps_gen_source_info info;	/* PSS generator info */
	bool enabled;				/* PSS generator status */
	......
	/* More members */
}

/* TIO struct */

struct pps_tio {
	struct pps_gen_source_info pps_tio_source_info;
	struct pps_gen_device *pps_gen;
	struct hrtimer timer;
	struct device *dev;
	spinlock_t lock;
	........
	/* More members */
};

Once the framework registration is done in probe function through pps_gen_register_source we get back pps_gen pointer and stored in our struct pps_tio.

As part of enable function:
We are trying to get back pps_tio (parent structure) with the help of pps_gen as this is pointer container_of does not work here. Hence proposed add a new generic private pointer inside the pps_gen_source_info structure which is exposed to driver.

static int pps_tio_enable(struct pps_gen_device *pps_gen, bool enable)

{
    /* Trying to get back pps_tio struct is not possible as pps_gen is a  pointer in structure */
     container_of(pps_gen, struct pps_tio, pps_gen);

}

Compilation Error:
In file included from ./include/linux/bitfield.h:10,
from drivers/pps/generators/pps_gen_tio.c:8:
drivers/pps/generators/pps_gen_tio.c: In function 'pps_tio_gen_enable':
./include/linux/build_bug.h:78:41: error: static assertion failed: "pointer type mismatch in container_of()"
78 | #define __static_assert(expr, msg, ...) _Static_assert(expr, msg)

      |                                         ^~~~~~~~~~~~~~
./include/linux/build_bug.h:77:34: note: in expansion of macro '__static_assert'
   77 | #define static_assert(expr, ...) __static_assert(expr, ##__VA_ARGS__, #expr)
      |                                  ^~~~~~~~~~~~~~~
./include/linux/container_of.h:20:2: note: in expansion of macro 'static_assert'
   20 |  static_assert(__same_type(*(ptr), ((type *)0)->member) || \
      |  ^~~~~~~~~~~~~

drivers/pps/generators/pps_gen_tio.c:172:24: note: in expansion of macro 'container_of'
172 | struct pps_tio *tio = container_of(pps_gen,struct pps_tio, pps_gen);

Quoted 17 lines
Quoted 17 lines

in using static pointer. To avoid the same and get back the generator driver data back we are proposing generic approach to add drv_prv_data pointer inside the struct pps_gen_source_info.

Example TIO structure wrapped with pps_gen_device and usage.

struct pps_tio {
	/* Framework Related * /
	struct pps_gen_source_info pps_tio_source_info
	struct pps_gen_device *pps_gen;

};

static int pps_tio_enable(struct pps_gen_device *pps_gen, bool enable) {

Quoted 6 lines
Quoted 3 lines
    /* Getting TIO data back */
    /* Note: drv_prv_data will be initialized in our init routine */
    struct pps_tio *tio = pps_gen->info.drv_prv_data;

So, what's wrong with the container_of() against pps_gen->info? We have tons of code in the kernel that does it.

We are trying to get our TIO/struct pps_tio structure back using pps_gen pointer..

Quoted 9 lines
Quoted 5 lines
    /* Access tio members here to set some of the parameters */

    return 0;
}

Okay, looking into the implementation I see what's the issue (but it doesn't mean that commit message is good here, you still need to explain better why you can't get the correct address).

Ok, Something like this:

"Adding new generic private pointer to the struct pps_gen_source_info which can help in saving the parent driver data.
In the existing adaption using container_of with a pointer to fetch the parent driver data/address is not possible"

"Example as shown above can be given as reference"

Quoted 5 lines

The problem is that info is got fully copied. Perhaps we should rely on the fact that it always be provided? I dunno any data structure that defines callbacks and other "info" material should be considered as temporary storage (on stack), it makes not much sense.

We are trying to get our TIO/struct pps_tio structure back using pps_gen pointer..

Quoted 6 lines

...

Quoted 2 lines

V1 -> V2:
* Updated reviewers.

Quoted 6 lines

Please, use...

Quoted 2 lines

---

...this place for the comments and changelogs.

Ok Will take care in next version.

Quoted 6 lines

...

include/linux/pps_gen_kernel.h struct pps_gen_source_info { · line 38

Quoted 1 line
include/linux/pps_gen_kernel.h +1 -0
+	void *drv_prv_data;

If really needed we should probably use a better name, like "private".

Ok, will change the member name to private.

Thanks,
Subbu

Quoted 5 lines

--
With Best Regards,
Andy Shevchenko

#3 lore raw
Quoted 5 lines · Subramanian
Quoted 5 lines

-----Original Message-----
From: Andy Shevchenko <[email protected]>
Sent: Wednesday, January 29, 2025 12:24 PM
On Tue, Jan 28, 2025 at 07:47:43PM +0530, [email protected] wrote:

...

Quoted 11 lines
Quoted 10 lines
Quoted 6 lines

While adapting pps generator driver(tio generator as an example)to the new generator framework, As part of driver registration the pps_gen_device pointer is returned from framework. Due to which there is difficulty in getting generator driver data back in enable function. we won't be able to use container_of macro as it results in static assert. we might end up

container_of()

Can you be more specific, what exactly happens?

Let me provide some more info:

...

Quoted 3 lines

Once the framework registration is done in probe function through pps_gen_register_source we get back pps_gen pointer and stored in our struct pps_tio.

This is irrelevant.

Quoted 26 lines

As part of enable function:
We are trying to get back pps_tio (parent structure) with the help of pps_gen as this is pointer container_of does not work here.
Hence proposed add a new generic private pointer inside the pps_gen_source_info structure which is exposed to driver.

static int pps_tio_enable(struct pps_gen_device *pps_gen, bool enable)

{
    /* Trying to get back pps_tio struct is not possible as pps_gen is a  pointer in structure */
     container_of(pps_gen, struct pps_tio, pps_gen);

}

Compilation Error:
In file included from ./include/linux/bitfield.h:10,
from drivers/pps/generators/pps_gen_tio.c:8:
drivers/pps/generators/pps_gen_tio.c: In function 'pps_tio_gen_enable':
./include/linux/build_bug.h:78:41: error: static assertion failed: "pointer type mismatch in container_of()"
78 | #define __static_assert(expr, msg, ...) _Static_assert(expr, msg)

      |                                         ^~~~~~~~~~~~~~
./include/linux/build_bug.h:77:34: note: in expansion of macro '__static_assert'
   77 | #define static_assert(expr, ...) __static_assert(expr, ##__VA_ARGS__, #expr)
      |                                  ^~~~~~~~~~~~~~~
./include/linux/container_of.h:20:2: note: in expansion of macro 'static_assert'
   20 |  static_assert(__same_type(*(ptr), ((type *)0)->member) || \
      |  ^~~~~~~~~~~~~
Quoted 2 lines

drivers/pps/generators/pps_gen_tio.c:172:24: note: in expansion of macro 'container_of'
172 | struct pps_tio *tio = container_of(pps_gen,struct pps_tio, pps_gen);

Of course, you shouldn't use pointers, only embedded in members can be used for container_of(). It seems like basic misunderstanding of how container_of() works.

Quoted 34 lines
Quoted 16 lines
Quoted 16 lines

in using static pointer. To avoid the same and get back the generator driver data back we are proposing generic approach to add drv_prv_data pointer inside the struct pps_gen_source_info.

Example TIO structure wrapped with pps_gen_device and usage.

struct pps_tio {
	/* Framework Related * /
	struct pps_gen_source_info pps_tio_source_info
	struct pps_gen_device *pps_gen;

};

static int pps_tio_enable(struct pps_gen_device *pps_gen, bool enable) {

Quoted 6 lines
Quoted 3 lines
    /* Getting TIO data back */
    /* Note: drv_prv_data will be initialized in our init routine */
    struct pps_tio *tio = pps_gen->info.drv_prv_data;

So, what's wrong with the container_of() against pps_gen->info? We have tons of code in the kernel that does it.

We are trying to get our TIO/struct pps_tio structure back using pps_gen pointer..

Quoted 8 lines
Quoted 4 lines
    /* Access tio members here to set some of the parameters */

    return 0;
}

Okay, looking into the implementation I see what's the issue (but it doesn't mean that commit message is good here, you still need to explain better why you can't get the correct address).

Quoted 5 lines

Ok, Something like this:

"Adding new generic private pointer to the struct pps_gen_source_info which can help in saving the parent driver data. In the existing adaption using container_of with a pointer to fetch the parent driver data/address is not

Pay attention for other comments which I gave against commit message text.

Quoted 11 lines

possible"

"Example as shown above can be given as reference"

Quoted 4 lines

The problem is that info is got fully copied. Perhaps we should rely on the fact that it always be provided? I dunno any data structure that defines callbacks and other "info" material should be considered as temporary storage (on stack), it makes not much sense.

We are trying to get our TIO/struct pps_tio structure back using pps_gen pointer..

And..? What I proposed to replace copy of info structure in the struct pps_gen_device by a pointer and make it const. In such a case the users must provide either this from a global constant variable or a heap (if for some most unlike reason they do it dynamically). That will solve your issue and also makes memory foot print better. I.o.w. we have time to modify this in easy way while the framework has not be spread over the kernel.

Signature · 2 lines
With Best Regards,
Andy Shevchenko