From mboxrd@z Thu Jan 1 00:00:00 1970 From: Adolf Belka To: development@lists.ipfire.org Subject: Re: Problem with trying to build qemu-9.0.2 - it requires dtc for all architectures Date: Wed, 04 Sep 2024 15:37:48 +0200 Message-ID: <182681ee-7f75-4117-8ded-978757fadf94@ipfire.org> In-Reply-To: <7DD6AED4-8644-4ACE-9FC6-1AA8FCB3FE91@ipfire.org> MIME-Version: 1.0 Content-Type: multipart/mixed; boundary="===============7785916139924562391==" List-Id: --===============7785916139924562391== Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable Hi Michael, On 04/09/2024 11:34, Michael Tremer wrote: > Hello, >=20 >> On 4 Sep 2024, at 09:22, Adolf Belka wrote: >> >> Hi Michael, >> >> On 04/09/2024 08:58, Michael Tremer wrote: >>> Hello, >>>> On 3 Sep 2024, at 21:06, Adolf Belka wrote: >>>> >>>> Hi All, >>>> >>>> On 03/09/2024 15:25, Adolf Belka wrote: >>>>> Hi Michael, >>>>> >>>>> On 03/09/2024 09:24, Michael Tremer wrote: >>>>>> Hello Adolf, >>>>>> >>>>>> Yes, you can make those changes. >>>>>> >>>>>> dtc is the device tree compiler and it should be quick to build. We ju= st didn=E2=80=99t do that since we didn=E2=80=99t have any use for it apart f= rom aarch64. >>>>>> >>>>>> Move it wherever you want. It should not have any further dependencies= . We can build it for all architectures. >>>>> >>>>> I did that and qemu found libfdt but then said that it was too old and = that a min of 1.5.1 was required but the updated version of dtc I had built w= as version 1.7.1 >>>>> >>>> I have found the section in meson.build where it checks for the fdt libr= ary and it seems to me that the code says that if the system fdt is being use= d then say that it is too old. I can't see any actual check of the version nu= mber of libfdt. >>> Hmm, the code is not the cleanest, but it should perform a few checks bef= ore it throws that error. The crucial part is here: >>>> if fdt.found() and cc.links(''=E2=80=99 >>>> ...''', >>>> dependencies: fdt) >>>> fdt_opt =3D 'system' >>>> elif fdt_opt =3D=3D 'system' >>>> error('system libfdt requested, but it is too old (1.5.1 or newe= r required)') >>>> else >>>> fdt_opt =3D 'internal' >>>> fdt =3D not_found >>>> endif >>> It first of all checks if fdt.found() returns true, it so, it will check = to compile and build the program in cc.links(=E2=80=A6). I assume that the fu= nction fdt_find_max_phandle() does not exitt before version 1.5.1 and therefo= re the program should not build with older versions. Then it will move to the= =E2=80=9Celif fdt_opt =3D=3D =E2=80=99system=E2=80=99=E2=80=9D part and comp= lain that the library that is already installed is incompatible. >>> So the check should work and actually do something. >>> Whenever a program bundles a library, we should remove it. Bundles librar= ies are evil because lets say there is a problem in libfdt, we will patch the= package, but we don=E2=80=99t know what other programs bring their own versi= on which won=E2=80=99t receive the fix. This is particularly bad for anything= security-related as parts of the distribution will still be vulnerable. >> >> Then when I am doing my package updates if I see any bundled stuff then I = will check how to use a system version instead. >>> Shared libraries (like libfdt.so .N) are nicer because= they can be pulled in by any program that needs the functionality and can be= replaced with an improved version without any regard of what programs link a= gainst it - for as long as there are no breaking changes. >>> Rust for example is going the opposite way where everything is statically= linked which is a security nightmare as described above. Every program that = uses a vulnerable crate will have to be rebuilt. But how do you find them all? >>> There is more about bundles libraries here: https://fedoraproject.org/wik= i/Bundled_Libraries >>>> Can you have a check through this following code and see if my interpret= ation is correct or if I have just misunderstood how the code is supposed to = be working. >>>> >>>> It also looks like the check for the fdt code being internal has not bee= n removed from it even though dtc is no longer bundled with qemu. >>> In that case, we should set fdt=3Dsystem just to make sure that QEMU does= the right thing during the build. >> >> Okay, I will try that and see what happens. >=20 > For as long as you have built dtc before QEMU this should work fine. That change was not enough on its own to make the libfdt detection work. I got feedback from Thomas Huth on the qemu gitlab site saying that the error= message might be a bit misleading and that the issue was more likely that th= e libfdt was simply not usable. He said that more info should be in the meson= -logs/meson-log.txt file Commenting out the rm -rf $(DIR_APP) command in the qemu lfs meant that I cou= ld read the build/meson-logs/meson-log.txt and that showed that the #include = command was not finding the libfdt.h file. This turned out to be that the dtc build did not have PREFIX=3D/usr and so th= ose files were being put in /include/ and not /usr/include/ So I added PREFIX=3D/usr to the end of the dtc make install line and qemu the= n successfully built. I will now do a clean and fresh build to confirm that everything stays workin= g but I might be able to submit a patch for qemu-9.0.2 now. Is there any problem with the dtc command being changed from cd $(DIR_APP) && make HOME=3D install to cd $(DIR_APP) && make HOME=3D install PREFIX=3D/usr in terms of the aarch64 requirements? Regards, Adolf. >=20 >>> Did I not merge any patches yesterday? Is that version not affected? >> >> Yes, you did but that was for version 9.0.0 which still had the bundled dt= c. I thought it was still worth updating as there have been 11 update version= s since the 8.2.1 version we currently have in stable. >=20 > Ah, that makes sense. >=20 >> I will also have a look at the other packages they have in the subprojects= directory to see if our build uses any of them. >=20 > Thank you! >=20 > -Michael >=20 >> Regards, >> >> Adolf. >>> -Michael >>>> fdt =3D not_found >>>> fdt_opt =3D get_option('fdt') >>>> if fdt_required.length() > 0 or fdt_opt =3D=3D 'enabled' >>>> if fdt_opt =3D=3D 'disabled' >>>> error('fdt disabled but required by targets ' + ', '.join(fdt_requi= red)) >>>> endif >>>> >>>> if fdt_opt in ['enabled', 'auto', 'system'] >>>> if get_option('wrap_mode') =3D=3D 'nodownload' >>>> fdt_opt =3D 'system' >>>> endif >>>> fdt =3D cc.find_library('fdt', required: fdt_opt =3D=3D 'system') >>>> if fdt.found() and cc.links(''' >>>> #include >>>> #include >>>> int main(void) { fdt_find_max_phandle(NULL, NULL); return 0; }''= ', >>>> dependencies: fdt) >>>> fdt_opt =3D 'system' >>>> elif fdt_opt =3D=3D 'system' >>>> error('system libfdt requested, but it is too old (1.5.1 or newe= r required)') >>>> else >>>> fdt_opt =3D 'internal' >>>> fdt =3D not_found >>>> endif >>>> endif >>>> if not fdt.found() >>>> assert(fdt_opt =3D=3D 'internal') >>>> libfdt_proj =3D subproject('dtc', required: true, >>>> default_options: ['tools=3Dfalse', 'yaml= =3Ddisabled', >>>> 'python=3Ddisabled', 'de= fault_library=3Dstatic']) >>>> fdt =3D libfdt_proj.get_variable('libfdt_dep') >>>> endif >>>> else >>>> fdt_opt =3D 'disabled' >>>> endif >>>> >>>> >>>>> It looks like qemu can correctly detect the presence of libfdt but not = the version number. >>>>> >>>>> I have raised this as an issue in the qemu gitlab site. >>>>> >>>>> Version 9.0.0 still had the dtc subproject included in the source tarba= ll, so I am currently building that and have left the IPFire dtc where it was= for the moment. >>>>> >>>>> This will still update qemu by 11 version numbers. >>>>> >>>>> I will wait to see what response I get back from the qemu team for upda= ting to 9.0.2 >>>>> >>>>> Regards, >>>>> Adolf. >>>>> >>>>>> >>>>>> -Michael >>>>>> >>>>>>> On 2 Sep 2024, at 19:46, Adolf Belka wrote: >>>>>>> >>>>>>> Hi All, >>>>>>> >>>>>>> I am trying to do an update of the qemu package. >>>>>>> >>>>>>> It is failing for a missing fdt library which it says is required by = targets x86_64-softmmu, aarch64-softmmu and riscv64-softmmu. >>>>>>> >>>>>>> The previous version that we currently have, 8.1.2 has a bundled dtc = which includes the fdt library. That has been removed, probably with the chan= ge to the 9.x branch as it is not in 9.0.2 >>>>>>> >>>>>>> We have the dtc package in make but it is after qemu. >>>>>>> >>>>>>> I could move it to before qemu, however I also note that the dtc pack= age is specified only for aarch64 but if we need to have the above softmmu ta= rgets specified then qemu is requiring the fdt library for all our architectu= res. >>>>>>> >>>>>>> Before I move dtc and change it to be for all architectures, I just w= ant to flag this up and confirm that it is the right thing to do. >>>>>>> >>>>>>> Regards, >>>>>>> >>>>>>> Adolf. >=20 >=20 --===============7785916139924562391==--