Messages in this thread | | | Date | Fri, 5 Feb 2010 08:56:45 +0100 | From | Ingo Molnar <> | Subject | Re: hung bootup with "drm/radeon/kms: move radeon KMS on/off switch out of staging." |
| |
* Matthew Garrett <mjg59@srcf.ucam.org> wrote:
> On Thu, Feb 04, 2010 at 10:05:59PM +0100, Ingo Molnar wrote: > > > Regressions are not limited to 'same config' kernels, last i checked. If that > > has changed (or if i'm misunderstanding it) then it would be nice to hear a > > clarification about that from Linus. > > If an option has *never* worked on a given configuration, then it's not a > regression. [...]
The *main driver* (CONFIG_DRM_RADEON=y, as far as the user is concerned) is many years old and it certainly worked just fine for tens of thousands of test iterations on this box, up until that commit.
That box alone has done in excess of half a million boot iterations in the past 2+ years. About 28% of the tests had CONFIG_DRM_RADEON=y, so the number of successful bootups with CONFIG_DRM_RADEON=y is in excess of one hundred thousand. There was not a single failed bootup in those two years due to the CONFIG_DRM_RADEON driver that i can remember.
If it now does not boot up if all its sub-options are enabled, even of some of those sub-options are new, does that count as a driver regression? Sure it does to me ...
IMHO you are trying to put a narrow technical distinction into it which does not exist for users. You argue that it's a new default-off sub-option of an existing driver, hence it cannot be a regression. The option shows up as a sub-option to an existing driver in make oldconfig, with a fairly innocious sounding help text, so to a user it certainly looks as if it's one unit and that it is the radeon driver that regressed.
*If* we make a driver feature available in a popular driver and make it Kconfig visible without obvious warnings (i.e. lure the user to enable it with the notion that it's just one coherent unit with a trusted, existing driver), we should also hold up the other side of the deal and treat the *bugs* as a coherent unit as well.
I.e. treat the driver as a coherent whole not only when it's convenient to us, but also when it's somewhat inconvenient to do. We cannot have it both ways.
Ingo
| |