Re: proposal for bus_dma(9) change

[Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>]

Christoph_Egger%gmx.de@localhost wrote:

> > Changing API is really pain than fixing bugs, IMO.
> 
> My proposal is not like adding an additional parameter
> to somewhere.
> My proposal would change the behaviour of one function
> in case of error.

My question is if is there any benefit to change all MD
bus_dmamap_create(9) implementation instead of fixing all drivers.

In most drivers, dmamaps are prepared in softc and they are
initialized to NULL in attach functions so checking it
in failure path is valid.  I guess some guys blindly copied
piece of such code even for uninitialized/reused dmamaps
(mostly for jumbo frames?), so I don't think there are
so many drivers which have the bug.

If the bug is caused by reusing dmamaps, we should explicitly
make dmamap NULL in driver dependent free functions, rather than
MI bus_dmamap_create(9) API.
---
Izumi Tsutsui