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

From: liweihang

Sent: 16 June 2021 07:17

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

quoted

From: Weihang Li

quoted

Sent: 11 June 2021 07:37

Prefer __packed over __attribute__((__packed__)).

Signed-off-by: Wenpeng Liang <redacted>
Signed-off-by: Weihang Li <redacted>
---
 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

...

quoted

 /* 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)