Re: linux-next: Tree for Sep 19

[Randy Dunlap <rdunlap@infradead.org>]

On 09/19/14 17:15, Guenter Roeck wrote:
> On 09/19/2014 03:21 PM, Randy Dunlap wrote:
>> On 09/19/14 14:14, Guenter Roeck wrote:
>>> On Fri, Sep 19, 2014 at 04:58:17PM +1000, Stephen Rothwell wrote:
>>>> Hi all,
>>>>
>>>> Changes since 20140917:
>>>>
>>>> The fsl tree still had its build failure so I used the version from
>>>> next-20140917.
>>>>
>>>> The v4l-dvb tree lost its build failure.
>>>>
>>>> The security tree gained a conflict against the file-locks tree.
>>>>
>>>> Non-merge commits (relative to Linus' tree): 6014
>>>>   5488 files changed, 217522 insertions(+), 129375 deletions(-)
>>>>
>>>> ----------------------------------------------------------------------------
>>>>
>>> Guess this is most difficult one.
>>>
>>> mips:nlm_xlp_defconfig:
>>>
>>> warning: (SCSI_BNX2X_FCOE && LIBFCOE && TCM_QLA2XXX) selects LIBFC which has
>>> unmet direct dependencies (SCSI_LOWLEVEL && SCSI && SCSI_FC_ATTRS)
>>> warning: (SCSI_BNX2X_FCOE && LIBFCOE && TCM_QLA2XXX) selects LIBFC which has
>>> unmet direct dependencies (SCSI_LOWLEVEL && SCSI && SCSI_FC_ATTRS)
>>
>> Yes, I have a patch sitting on my hard drive that makes LIBFCOE and TCM_QLA2XXX
>> and SCSI_BNX2X_FCOE depend on SCSI_FC_ATTRS, but I'm not entirely happy about
>> having to hunt these down (even with the help of kconfig warnings).
>>
> 
> One key problem is that nlm_xlp_defconfig does not configure NET anymore.
> this is after 'scsi_netlink : Make SCSI_NETLINK dependent on NET instead
> of selecting NET' was applied.
> 
> In fact, there are many more affected configurations; the change from
> "select XXX" to "depends XXX" has far reaching consequences, as many
> configurations are no longer valid. For mips alone I find that this commit
> changes the following configurations.
> 
> e55_defconfig
> gpr_defconfig
> ip27_defconfig
> jazz_defconfig
> loongson3_defconfig
> malta_defconfig
> malta_kvm_defconfig
> malta_kvm_guest_defconfig
> mtx1_defconfig
> nlm_xlp_defconfig
> nlm_xlr_defconfig
> rm200_defconfig
> 
> e55_defconfig is almost the same as before, but only because CONFIG_NET
> was not configured for it to start with. For all others, CONFIG_NET is
> no longer set. This effectively means that out of 55 mips configurations,
> 11 or 20% are now bad. This is 3.17-rc5. vs. next-20140919.
> 
> Frankly, I don't think this change was really helpful. On the contrary,
> we will be in a lot of trouble when this change makes it upstream.
> It might be be a good idea to be much more careful when making such
> changes. Cleaning up configuration dependencies is a laudable goal,
> but when it breaks configurations all over the place it does more damage
> than it is worth.

Yes, I suspect that we should just drop this effort and redo the original
problem's "fix".

Dave?

Anish, if you can remind me of the original problem and hopefully a kernel
.config file for it, I'll try to help with it.

-- 
~Randy