On Sat, 15 Nov 2003 20:57:36 +0300
Andrey Borzenkov <[EMAIL PROTECTED]> wrote:
> On Wednesday 12 November 2003 04:44, Olivier Blin wrote:
> > +# CONFIG_M586 is not set
> OK that needs tweaking anyway.
> > -CONFIG_X86_F00F_BUG=y
> F00F code is really small, there is no reason to disable it.
ok, I will keep it
>
> > --
> > -CONFIG_X86_ALIGNMENT_16=y
> > +CONFIG_X86_GOOD_APIC=y
>
> I can't find where it is referenced
in arch/i386/Kconfig
> > CONFIG_X86_INTEL_USERCOPY=y
> > -# CONFIG_HPET_TIMER is not set
>
> anyone has info about how widely it is used? should it be enabled in
> generic kernel?
sorry, I don't know
> > +CONFIG_EDD=m
>
> make CONFIG_EDD module. It provides just sysfs interface to EDD data,
> unless we have applications that access sysfs we do not need it in
> kernel. And those applications can always modprobe edd first.
ok, i'll do it
> > -# CONFIG_MATH_EMULATION is not set
> > +CONFIG_MATH_EMULATION=y
>
> OK that is just stupidity of allmodconfig. Remember it blindly tries
> to enable everything ...
right
> > CONFIG_MTRR=y
> > --
> > -CONFIG_ACPI_AC=y
> > -CONFIG_ACPI_BATTERY=y
> > -CONFIG_ACPI_BUTTON=y
>
> No, please! Everything that can be module should be module. Desktop
> users has no need for most of them and those who need can always load
> them. It would be nice if harddrake could detect needed modules and
> automatically update /etc/sysconfig/acpi.
I agree this should be built as module, I've not made the first
2.6 config file and I haven't had time yet to browse the whole config.
I already planned to do it for next release.
Acutally, I submitted you this diff before I had time to fix most of
obvious things I would have fixed in next release :)
Sorry for that, I should have wait a bit to send you the diff.
> > -CONFIG_APM=y
> > +CONFIG_APM=m
>
> not sure. Probably it should be builtin for legacy reasons - else you
> get "APM missing in Mandrake kernel!". Although technically it can
> just be loaded out of initrd I guess. Given that new systems are
> likely to support ACPI may be we should make it module. And make
> harddrake detect when APM is needed :)
ok, so builtin for now, module when initscripts will be able to load it
> > -CONFIG_CPU_FREQ_TABLE=y
> > +CONFIG_CPU_FREQ_GOV_USERSPACE=m
> > +CONFIG_CPU_FREQ_24_API=y
> > +CONFIG_CPU_FREQ_TABLE=m
>
> I think those need be modules if possible. Anyone has any experience
> with it?
I'll build them as module for next release, if no objection
> > -CONFIG_PARPORT=y
> > -CONFIG_PARPORT_PC=y
> > -CONFIG_PARPORT_PC_CML1=y
> > +CONFIG_PARPORT=m
> > +CONFIG_PARPORT_PC=m
> > +CONFIG_PARPORT_PC_CML1=m
>
> come on, why should it be builtin?
>
same as acpi, I planed to fix that in next release
> > -CONFIG_BLK_DEV_FD=y
>
> ditto
>
> > +CONFIG_BLK_DEV_FD=m
> > +CONFIG_BLK_DEV_PS2=m
> > CONFIG_BLK_DEV_XD=m
> > --
> > -CONFIG_PARIDE_PARPORT=y
> > +CONFIG_PARIDE_PARPORT=m
> >
>
> ditto. I do not have paride device nor I believe do 99% of users.
was planned too
> > #
> > --
> > -CONFIG_BLK_DEV_RAM=y
> > +CONFIG_BLK_DEV_RAM=m
>
> should it be builtin for initrd to work? If not there is no reason to
> have it builtin.
strange, i've quickly browsed the code, it seems it isn't needed.
I'll have a try without.
> > -CONFIG_IDE=y
> > -CONFIG_BLK_DEV_IDE=y
> > +CONFIG_IDE=m
> > +CONFIG_BLK_DEV_IDE=m
> >
> > --
> > -CONFIG_BLK_DEV_IDEDISK=y
> > +CONFIG_BLK_DEV_HD_IDE=y
> > +CONFIG_BLK_DEV_IDEDISK=m
> > CONFIG_IDEDISK_MULTI_MODE=y
>
> nope, module does not work. Even if it possible to fix all missing
> exported symbols, IDE layer does not provide for any sort of module
> reference counting (for chipset drivers) meaning it is simply unsafe.
> Which implies kernel must include all IDE drivers :(
yep :/
That's what i did in test9.6mdk.
Furthermore, ide is initialized before ramdisk is loaded, this will need
some tweaking if we want to build IDE as module.
I don't know what prevents ramdisk from being loaded before initcalls
are done.
Maybe I'll ask on lkml.
> > -CONFIG_SCSI=y
> > +CONFIG_SCSI=m
>
> module, please!
>
> > -CONFIG_BLK_DEV_MD=y
> > +CONFIG_BLK_DEV_MD=m
>
> same.
was planned too
> > -CONFIG_PACKET=y
> > +CONFIG_PACKET=m
>
> module
>
> > -CONFIG_UNIX=y
> > +CONFIG_UNIX=m
>
> module
I was not sure for those.
Almost everybody use it, even if the box isn't networked.
> > -CONFIG_ATM=y
> > -CONFIG_ATM_CLIP=y
> > -# CONFIG_ATM_CLIP_NO_ICMP is not set
> > +CONFIG_ATM=m
> > +CONFIG_ATM_CLIP=m
>
> come on really, should everyone have ATM?
same, planned
> > -CONFIG_INPUT_MOUSEDEV=y
> > +CONFIG_INPUT_MOUSEDEV=m
> > -CONFIG_SERIO=y
> > -CONFIG_SERIO_I8042=y
> > +CONFIG_SERIO=m
> > +CONFIG_SERIO_I8042=m
> > -CONFIG_KEYBOARD_ATKBD=y
> > +CONFIG_KEYBOARD_ATKBD=m
> > -CONFIG_MOUSE_PS2=y
> > +CONFIG_MOUSE_PS2=m
> that is bad. Apparently to have your mouse (and keyboard) autodetected
> you have
> to build it in. OTOH some drivers are still left as module. Meaning -
> there should be some manual configuration. In which case why not
> configure others as well?
ok, so i'll build in all mouse and keyboard drivers
> There is no problem loading them out of initrd. Of course it needs
> harddrake support finally.
>
> WDYT?
How can harddrake detect the mouse/keyboard type ?
It is easy ?
> > -CONFIG_SERIAL_8250=y
> > -CONFIG_SERIAL_8250_CONSOLE=y
> > +CONFIG_SERIAL_8250=m
>
> it needs to be builtin only to have console support ... not sure
> actually how widely it is used. those needing it to debug are likely
> to know how to recompile kernel.
>
> > -CONFIG_SERIAL_CORE=y
> > -CONFIG_SERIAL_CORE_CONSOLE=y
> > +CONFIG_SERIAL_CORE=m
>
> see above
I'll keep it built in for now
> > -CONFIG_AGP=y
> > -CONFIG_AGP_ALI=y
> > +CONFIG_AGP=m
> > +CONFIG_AGP_ALI=m
> > CONFIG_AGP_ATI=m
> > -CONFIG_AGP_AMD=y
> > +CONFIG_AGP_AMD=m
> > CONFIG_AGP_AMD64=m
> > -CONFIG_AGP_INTEL=y
> > +CONFIG_AGP_INTEL=m
> > CONFIG_AGP_NVIDIA=m
> > -CONFIG_AGP_SIS=y
> > -CONFIG_AGP_SWORKS=y
> > -CONFIG_AGP_VIA=y
> > +CONFIG_AGP_SIS=m
> > +CONFIG_AGP_SWORKS=m
> > +CONFIG_AGP_VIA=m
>
> any reason to have this builtin? I think it works just fine as module?
no reason, it was planned to
> > -CONFIG_EXT2_FS=y
> > +CONFIG_EXT2_FS=m
>
> if we switch to initramfs format even this can be module :)
yes :)
i'll make some tests soon
> > -CONFIG_FS_MBCACHE=y
>
> why do you need it in kernel?
config FS_MBCACHE
# Meta block cache for Extended Attributes (ext2/ext3)
tristate
depends on EXT2_FS_XATTR || EXT3_FS_XATTR
default y if EXT2_FS=y || EXT3_FS=y
default m if EXT2_FS=m || EXT3_FS=m
isn't this a sufficient argument ?
> > -CONFIG_QFMT_V2=y
>
> why?
You're right, it should be built as a module.
I guess I would have spot this one too for my next release.
> > -CONFIG_NFS_FS=y
> why? OK diskless stations ... that is the only application that may
> require it. this is rather special case for generic kernel?
i'm not sure, some users may mount /home by NFS
How many bytes is the nfs code ?
> > -CONFIG_NFSD=y
>
> why?
It would have been done in next release too
> > -CONFIG_LOCKD=y
>
> why (unless it is forced by NFS)?
it is forced by NFSD, but since NFSD will be a module, it will be a
module too
> > -CONFIG_EXPORTFS=y
> > -CONFIG_SUNRPC=y
> why?
same as LOCKD
Thanks for your time Andrey :)
--
Olivier Blin
