Re: [PATCH] net: phy: Fix phy_modify() semantic difference fallout

[Russell King - ARM Linux]

On Thu, Jan 11, 2018 at 05:00:03PM +0100, Geert Uytterhoeven wrote:
> On Thu, Jan 11, 2018 at 4:54 PM, Geert Uytterhoeven
> <geert@xxxxxxxxxxxxxx> wrote:
> > On Thu, Jan 11, 2018 at 4:53 PM, Russell King - ARM Linux
> > <linux@xxxxxxxxxxxxxxx> wrote:
> >> On Thu, Jan 11, 2018 at 10:48:35AM -0500, David Miller wrote:
> >>> From: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
> >>> Date: Tue,  9 Jan 2018 12:11:21 +0100
> >>>
> >>> > In case of success, the return values of (__)phy_write() and
> >>> > (__)phy_modify() are not compatible: (__)phy_write() returns 0, while
> >>> > (__)phy_modify() returns the old PHY register value.
> >>> >
> >>> > Apparently this change was catered for in drivers/net/phy/marvell.c, but
> >>> > not in other source files.
> >>> >
> >>> > Hence genphy_restart_aneg() now returns 4416 instead zero, which is
> >>> > considered an error:
> >>> >
> >>> >     ravb e6800000.ethernet eth0: failed to connect PHY
> >>> >     IP-Config: Failed to open eth0
> >>> >     IP-Config: No network devices available
> >>> >
> >>> > Fix this by converting positive values to zero in all callers of
> >>> > phy_modify().
> >>> >
> >>> > Fixes: fea23fb591cce995 ("net: phy: convert read-modify-write to phy_modify()")
> >>> > Signed-off-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
> >>> > ---
> >>> > Alternatively, __phy_modify() could be changed to follow __phy_write()
> >>> > semantics?
> >>>
> >>> I really want a resolution to this quickly, this broke lots of stuff
> >>> for people.
> >>>
> >>> __phy_modify() wants to return multiple values, so it should be coded
> >>> up to do so explicitly rather than trying to encode two values from
> >>> overlapping value spaces in one return value.
> >>>
> >>> That means the original value should be returned by-reference.  And
> >>> this will make the error/no-error return value unambiguous.
> >>>
> >>> int __phy_modify(struct phy_device *phydev, u32 regnum, u16 mask, u16 set,
> >>>                u16 *orig_val);
> >>
> >> I'm sorry I have no time to work on this right now due to the meltdown
> >> and spectre stuff that hit last week.  If you need to do something,
> >> please revert both the mvneta series and the series containing this
> >> patch.
> >
> > I'll have a look into it...
> 
> Sorry, the phy_restore_page() semantics are driving me crazy.
> Let's revert.

I don't have any other solution for the phy_restore_page() stuff other
than open coding those exact same semantics _everywhere_ it's used.

Unfortunately, not having those race fixes in place blocks converting
any network driver to use phylink and SFP.  If people find
phy_restore_page() distasteful, I have no way to progress phylink and
SFP any further.

I guess phylink and SFP stuff will also need to be reverted unless
someone can find a solution to this.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
According to speedtest.net: 8.21Mbps down 510kbps up