[opensuse-kernel] Re: 11.2 kabi vs. PM / yenta fix
On Thu, Nov 12, 2009 at 08:13:34PM +0100, Rafael J. Wysocki wrote:
On Thursday 12 November 2009, Michal Marek wrote:
Hi,
I added the kabi files from the 11.2 goldmaster kernel to the repository, so from now on, every kabi change will be caught by the build-time checks. Unfortunately there were already some changes in the repo that did change the kabi signatures: patches.drivers/alsa-hda-reboot-notify, already reverted by Takashi, "dm snapshot: use unsigned integer chunk size" and "PM / yenta: Split resume into early and late parts (rev. 4)", both came via the 2.6.31.6 update. I reverted the latter two for now, so that the builds don't start failing suddenly, but I guess that's not the right option esp. for the yenta fix. Rafael, you wrote the patch, can you comment here? The problem is the new member in struct pcmci_socket. Can we * just ignore this change (in worst case, some drivers will need a rebuild for the next kernel update)
I'd choose this option.
* hide the change from genksyms (do external drivers ever need to allocate the structure themselves?)
That I don't know. Greg?
I don't know of any that would. I vote for the "ignore this change" as it came from upstream, and any external driver would have to handle it anyway. And note that there almost are no external drivers anymore, with the exception of ati and nvidia, so that reduces the chance of a problem even more. And anyway, I didn't think we were doing ABI stuff for opensuse kernels, is this a change? thanks, greg k-h -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Greg KH napsal(a):
On Thu, Nov 12, 2009 at 08:13:34PM +0100, Rafael J. Wysocki wrote:
On Thursday 12 November 2009, Michal Marek wrote:
problem is the new member in struct pcmci_socket. Can we * just ignore this change (in worst case, some drivers will need a rebuild for the next kernel update) I'd choose this option.
* hide the change from genksyms (do external drivers ever need to allocate the structure themselves?) That I don't know. Greg?
I don't know of any that would. I vote for the "ignore this change" as it came from upstream, and any external driver would have to handle it anyway.
Ok, let's just whitelist it.
And anyway, I didn't think we were doing ABI stuff for opensuse kernels, is this a change?
The checks have always been there, and in cases where a kabi change would require rebuilding and re-releasing some commonly used KMPs (it's not just nvidia & ati, but also some out-of-tree GPL code we distribute) and working around the change is cheap, we usually do it. Michal -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Michal Marek napsal(a):
Greg KH napsal(a):
On Thu, Nov 12, 2009 at 08:13:34PM +0100, Rafael J. Wysocki wrote:
On Thursday 12 November 2009, Michal Marek wrote:
problem is the new member in struct pcmci_socket. Can we * just ignore this change (in worst case, some drivers will need a rebuild for the next kernel update) I'd choose this option.
* hide the change from genksyms (do external drivers ever need to allocate the structure themselves?) That I don't know. Greg? I don't know of any that would. I vote for the "ignore this change" as it came from upstream, and any external driver would have to handle it anyway.
Ok, let's just whitelist it.
Done. Michal -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
On 11/12/2009 02:16 PM, Greg KH wrote:
On Thu, Nov 12, 2009 at 08:13:34PM +0100, Rafael J. Wysocki wrote:
On Thursday 12 November 2009, Michal Marek wrote:
Hi,
I added the kabi files from the 11.2 goldmaster kernel to the repository, so from now on, every kabi change will be caught by the build-time checks. Unfortunately there were already some changes in the repo that did change the kabi signatures: patches.drivers/alsa-hda-reboot-notify, already reverted by Takashi, "dm snapshot: use unsigned integer chunk size" and "PM / yenta: Split resume into early and late parts (rev. 4)", both came via the 2.6.31.6 update. I reverted the latter two for now, so that the builds don't start failing suddenly, but I guess that's not the right option esp. for the yenta fix. Rafael, you wrote the patch, can you comment here? The problem is the new member in struct pcmci_socket. Can we * just ignore this change (in worst case, some drivers will need a rebuild for the next kernel update)
I'd choose this option.
* hide the change from genksyms (do external drivers ever need to allocate the structure themselves?)
That I don't know. Greg?
I don't know of any that would. I vote for the "ignore this change" as it came from upstream, and any external driver would have to handle it anyway.
And note that there almost are no external drivers anymore, with the exception of ati and nvidia, so that reduces the chance of a problem even more.
And anyway, I didn't think we were doing ABI stuff for opensuse kernels, is this a change?
No, we try to preserve the kABI the same way we do for the enterprise kernels because the pain for our users is still the same. The difference is that we regard it as best effort instead of a mandate. In general, we haven't really had to break anything outside of the initial 11.1 branch to SLE11 branch release for openSUSE 11.1. -Jeff (Sorry for the late reply - Thunderbird forgot that it was supposed to check this folder automatically.) -- Jeff Mahoney SUSE Labs -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
participants (3)
-
Greg KH -
Jeff Mahoney -
Michal Marek