diff mbox series

[net-next,8/8] net: phy: use '__packed' instead of '__attribute__((__packed__))'

Message ID 1623393419-2521-9-git-send-email-liweihang@huawei.com
State New
Headers show
Series [net-next,1/8] net: phy: add a blank line after declarations | expand

Commit Message

Weihang Li June 11, 2021, 6:36 a.m. UTC
From: Wenpeng Liang <liangwenpeng@huawei.com>

Prefer __packed over __attribute__((__packed__)).

Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
 drivers/net/phy/mscc/mscc_ptp.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Comments

David Laight June 14, 2021, 2:28 p.m. UTC | #1
From: Weihang Li

> Sent: 11 June 2021 07:37

> 

> Prefer __packed over __attribute__((__packed__)).

> 

> Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>

> Signed-off-by: Weihang Li <liweihang@huawei.com>

> ---

>  drivers/net/phy/mscc/mscc_ptp.h | 4 ++--

>  1 file changed, 2 insertions(+), 2 deletions(-)

> 

> diff --git a/drivers/net/phy/mscc/mscc_ptp.h b/drivers/net/phy/mscc/mscc_ptp.h

> index da34653..01f78b4 100644

...
>  /* Represents an entry in the timestamping FIFO */

>  struct vsc85xx_ts_fifo {

>  	u32 ns;

>  	u64 secs:48;

>  	u8 sig[16];

> -} __attribute__((__packed__));

> +} __packed;


Hmmmm I'd take some convincing that 'u64 secs:48' is anything
other than 'implementation defined'.
So using it to map a hardware structure seems wrong.

If this does map a hardware structure it ought to have
'endianness' annotations.
If it doesn't then why the bitfield and why packed?

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
Weihang Li June 16, 2021, 6:17 a.m. UTC | #2
On 2021/6/14 22:28, David Laight wrote:
> From: Weihang Li

>> Sent: 11 June 2021 07:37

>>

>> Prefer __packed over __attribute__((__packed__)).

>>

>> Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>

>> Signed-off-by: Weihang Li <liweihang@huawei.com>

>> ---

>>  drivers/net/phy/mscc/mscc_ptp.h | 4 ++--

>>  1 file changed, 2 insertions(+), 2 deletions(-)

>>

>> diff --git a/drivers/net/phy/mscc/mscc_ptp.h b/drivers/net/phy/mscc/mscc_ptp.h

>> index da34653..01f78b4 100644

> ...

>>  /* Represents an entry in the timestamping FIFO */

>>  struct vsc85xx_ts_fifo {

>>  	u32 ns;

>>  	u64 secs:48;

>>  	u8 sig[16];

>> -} __attribute__((__packed__));

>> +} __packed;

> 

> Hmmmm I'd take some convincing that 'u64 secs:48' is anything

> other than 'implementation defined'.

> So using it to map a hardware structure seems wrong.

> 

> If this does map a hardware structure it ought to have

> 'endianness' annotations.

> If it doesn't then why the bitfield and why packed?

> 

> 	David

> 

> -

> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK

> Registration No: 1397386 (Wales)

> 

> 


Hi David,

Thank you for your attention. You are right, I found the contents of structure
vsc85xx_ts_fifo is got from hardware. But I'm not sure if any issues or warnings
will be introduced into this driver after just changing 'u64 secs:48' to '__be64
secs:48'.

Let's keep this patch as it is. I cc the developers of the code, maybe they
didn't realize it or had some reasons to define it like that.

Thanks
Weihang
David Laight June 16, 2021, 8:47 a.m. UTC | #3
From: liweihang

> Sent: 16 June 2021 07:17

> 

> On 2021/6/14 22:28, David Laight wrote:

> > From: Weihang Li

> >> Sent: 11 June 2021 07:37

> >>

> >> Prefer __packed over __attribute__((__packed__)).

> >>

> >> Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>

> >> Signed-off-by: Weihang Li <liweihang@huawei.com>

> >> ---

> >>  drivers/net/phy/mscc/mscc_ptp.h | 4 ++--

> >>  1 file changed, 2 insertions(+), 2 deletions(-)

> >>

> >> diff --git a/drivers/net/phy/mscc/mscc_ptp.h b/drivers/net/phy/mscc/mscc_ptp.h

> >> index da34653..01f78b4 100644

> > ...

> >>  /* Represents an entry in the timestamping FIFO */

> >>  struct vsc85xx_ts_fifo {

> >>  	u32 ns;

> >>  	u64 secs:48;

> >>  	u8 sig[16];

> >> -} __attribute__((__packed__));

> >> +} __packed;

> >

> > Hmmmm I'd take some convincing that 'u64 secs:48' is anything

> > other than 'implementation defined'.

> > So using it to map a hardware structure seems wrong.

> >

> > If this does map a hardware structure it ought to have

> > 'endianness' annotations.

> > If it doesn't then why the bitfield and why packed?

> >

> > 	David

> 

> Hi David,

> 

> Thank you for your attention. You are right, I found the contents of structure

> vsc85xx_ts_fifo is got from hardware. But I'm not sure if any issues or warnings

> will be introduced into this driver after just changing 'u64 secs:48' to '__be64

> secs:48'.


I've just checked what this structure looks like - see https://godbolt.org/z/h4EqbMoso

Without any 'packed' annotations  'u64 secs:48' is aligned to an 8 byte
boundary, but is only 6 bytes wide (I don't use bitfields)
so the offset of 'sig' is 6 more than 'secs'.

But the size of the whole structure looks wrong.
I'd expect a hardware fifo so be a power of 2 big.
This one is 26 bytes (as above) or 28 bytes if the 'packed'
is only applied to 'secs' (which removed the 4 byte pad before
it while still allowing aligned 4-byte accesses to the structure.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
diff mbox series

Patch

diff --git a/drivers/net/phy/mscc/mscc_ptp.h b/drivers/net/phy/mscc/mscc_ptp.h
index da34653..01f78b4 100644
--- a/drivers/net/phy/mscc/mscc_ptp.h
+++ b/drivers/net/phy/mscc/mscc_ptp.h
@@ -450,14 +450,14 @@  struct vsc85xx_ptphdr {
 	__be16 seq_id;
 	u8 ctrl;
 	u8 log_interval;
-} __attribute__((__packed__));
+} __packed;
 
 /* Represents an entry in the timestamping FIFO */
 struct vsc85xx_ts_fifo {
 	u32 ns;
 	u64 secs:48;
 	u8 sig[16];
-} __attribute__((__packed__));
+} __packed;
 
 struct vsc85xx_ptp {
 	struct phy_device *phydev;