Re: [PATCH v2 2/3] bootconfig: Support embedding a bootconfig file in kernel

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Mar 22, 2022 at 5:16 PM Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
>
> On Tue, 22 Mar 2022 20:02:19 +0100
> Padmanabha Srinivasaiah <treasure4paddy@xxxxxxxxx> wrote:
>
> > Hello Masami Hiramatsu,
> >
> > On Tue, Mar 22, 2022 at 12:03:11PM +0900, Masami Hiramatsu wrote:
> > > On Mon, 21 Mar 2022 19:35:00 +0100
> > > Padmanabha Srinivasaiah <treasure4paddy@xxxxxxxxx> wrote:
> > >
> > > > Hello Masami Hiramatsu,
> > > >
> > > > On Fri, Mar 18, 2022 at 10:14:45AM +0900, Masami Hiramatsu wrote:
> > > > > On Wed, 16 Mar 2022 20:16:49 +0100
> > > > > Padmanabha Srinivasaiah <treasure4paddy@xxxxxxxxx> wrote:
> > > > >
> > > > > > Hello Masami Hiramatsu,
> > > > > >
> > > > > > Also noted that a change in default.bconf requries a clean build, is it
> > > > > > expected behaviour?
> > > > >
> > > > > default.bconf will be always updated if CONFIG_EMBED_BOOT_CONFIG=y. So you can
> > > > > do incremental build. (I tested it with the incremental build environment)
> > > > >
> > > >
> > > > Thanks, your observation made me to further experiment ther incremental build.
> > > >
> > > > Below are the observations I have:
> > > >
> > > > When I use GCC for a build; yes, the modified default.conf was observed on
> > > > the target.
> > > >
> > > > But when I use clang; either with FULL or THIN LTO, the modified
> > > > default.conf doesnt get reflected on the target.
> > >
> > > Hmm, curious. So you just add 'CC=clang' on the make command line, right?
> > Yes, CC=clang ARCH=arm64 LLVM=1. As specified here:
> > https://docs.kernel.org/kbuild/llvm.html.

You should just need LLVM=1 (and ARCH=arm64) at this point. LLVM=1
implies CC=clang.

Also, here's the start of the lore thread for folks:
https://lore.kernel.org/linux-doc/164724892075.731226.14103557516176115189.stgit@devnote2/

> >
> > > Can you confirm that following line in your build log,
> > >
> > >   GEN     lib/default.bconf
> > >
> > Yes, I do see above line. Indeed lib/default.bconf will get incremental
> > change.
> >
> >   GEN     lib/default.bconf
> >   CC      lib/bootconfig.o
> >   AR      lib/lib.a
> >
> > > and the timestamp of lib/bootconfig.o is built after lib/default.bconf file?
> > >
> > Yes, verified timestamp for all above artifacts including vmlinux.o.
> >
> > ex:
> > -rw-rw-r-- 1 psrinivasaia psrinivasaia 22K Mar 22 14:50
> > ../out/lib/bootconfig.o
> > -rw-rw-r-- 1 psrinivasaia psrinivasaia 355 Mar 22 14:50
> > ../out/lib/default.bconf
> > -rw-rw-r-- 1 psrinivasaia psrinivasaia 54M Mar 22 14:50 ../out/vmlinux.o
> >
> > As said incremnetal change was refelected in artifact default.bconf.
> > But not in vmlinux.o/vmlinux, used below command to verify.
>
> Interesting! This sounds clang's issue, because the make command rebuilds
> the object file including new default.bconf, but the linker (lld?)
> doesn't link it again correctly.

Sounds like missing FORCE directives in the Makefiles, perhaps?

Sami, do you recall any issues like this when implementing
commit dc5723b02e52 ("kbuild: add support for Clang LTO")
?

>
> >
> > llvm-objdump  -s -j .init.data ../out/vmlinux
> >
> > On target too, /proc/bootconfig shows old data.
> >
> > > And is that related to CONFIG_LTO? What happen if CONFIG_LTO=n?
> > >
> > Yes;  CONFIG_LTO_NONE=y  issue not observed even with LLVM binutils.
>
> And this issue is related to LTO. Maybe LTO ignores the '.init.data'
> section update. (Perhaps, LTO only checks the function code hash or
> something like that instead of the timestamp, and ignore whole object
> file if all of them are not updated.)

Sounds like this is a result of the above issue?
-- 
Thanks,
~Nick Desaulniers



[Index of Archives]     [Kernel Newbies]     [Security]     [Netfilter]     [Bugtraq]     [Linux FS]     [Yosemite Forum]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Samba]     [Video 4 Linux]     [Device Mapper]     [Linux Resources]

  Powered by Linux