Hi, On Thu, Sep 01, 2016 at 04:00:35PM -0700, David Daney wrote: > On 09/01/2016 01:43 PM, Aaro Koskinen wrote: > >PHY access through the board helper should be impossible with the > >current drivers, so delete this code and add BUG_ON(). > > > >Signed-off-by: Aaro Koskinen <aaro.koskinen@xxxxxx> > >--- > > .../cavium-octeon/executive/cvmx-helper-board.c | 110 +-------------------- > > 1 file changed, 5 insertions(+), 105 deletions(-) > > > >diff --git a/arch/mips/cavium-octeon/executive/cvmx-helper-board.c b/arch/mips/cavium-octeon/executive/cvmx-helper-board.c > >index 5572e39..8d75242 100644 > >--- a/arch/mips/cavium-octeon/executive/cvmx-helper-board.c > >+++ b/arch/mips/cavium-octeon/executive/cvmx-helper-board.c > [...] > >+ BUG_ON(cvmx_helper_board_get_mii_address(ipd_port) != -1); > > > > Can we do WARN_ON instead? > > No need to crash the kernel for this I think. I could change it, but I really think it is 100% unreachable case with the current mainline: If we know the mii address then the in-tree DTB pruning code would have set up the phy node properly, and this code never gets called. So reaching the BUG_ON condition really would mean a fatal bug somewhere. The only case you can hit the BUG_ON is: 1) you add new switch case CVMX_BOARD_TYPE_* into cvmx_helper_board_get_mii_address() 2) you pass your own DTB without proper phy-handle (or fixed-link) But we should not allow 1) as there is no reason to add new CVMX_BOARD_TYPE_* types anymore, and the best way to tell people not to do that is BUG_ON... A.
