On 4/24/12, Andrew Lunn [off-list ref] wrote:

Sorry, but still wrong.

The clock is optional. If we can find a clock, turn it on. If not,
keep going....

You patch causes the missing clock to become a fatal error.

This sata_mv exists in multiple forms. It can be part of a SoC. It can
also be on a PCI bus. In the PCI form, it is unlikely to have a clk
which can be controlled. When built into a SoC, namely one of the
Orion family, dove, orion5x, mv78xx0 do not have a clock which can be
controlled. However kirkwood does have a clock.

So, kirkwood will provide a clock and expects that sata_mv will turn
it on. All the other ways of using sata_mv will not provide a clock,
but still expect the driver to be happy.

Hmm. What this code does now is:
If HAVE_CLK is selected, then there must be a clock for the device. Otherwise
it will always pass.

You want not to return error if a platform does have HAVE_CLK, but doesn't
have a clock for sata? That would be simple to fix, but want to confirm if this
is actually required.

@Russell: Can we have your view also?

--
viresh