Date: Sat, 14 Oct 2017 08:28:32 +0200 From: Michal Meloun <melounmichal@gmail.com> To: bob prohaska <fbsd@www.zefox.net> Cc: Mark Millard <markmi@dsl-only.net>, freebsd-arm@freebsd.org, bob prohaska <bob@www.zefox.net> Subject: Re: Difficulty with armv6 to v7 transition. Message-ID: <559584b5-47c6-b9f8-9fae-43ee76fc99cc@freebsd.org> In-Reply-To: <20171014054706.GC75288@www.zefox.net> References: <CANCZdfrKYabu1-bWxX47=Lt=33e%2BFjBXCNBNiGPE7K-83KOAHA@mail.gmail.com> <CANCZdfqHfAe24q=6n8CxsHQv24j58bQLPG3z_7vi_xpGjdQzDg@mail.gmail.com> <20171011030021.GB57571@www.zefox.net> <20171013020604.GA70845@www.zefox.net> <20171013175943.GA74121@www.zefox.net> <F9F4D731-35CD-44E6-8419-D73CB98655E9@dsl-only.net> <20171014010713.GA75288@www.zefox.net> <254A2C41-59A9-4E4E-8982-ADDBAE2B5F91@dsl-only.net> <20171014021133.GB75288@www.zefox.net> <39f6419f-48f3-aaec-dfa4-3048c8a893d8@freebsd.org> <20171014054706.GC75288@www.zefox.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On 14.10.2017 7:47, bob prohaska wrote:
> Hi Michal,
>
> On Sat, Oct 14, 2017 at 06:28:13AM +0200, Michal Meloun wrote:
>> Bob,
>> can you please try this?
>> setenv MACHINE_ARCH armv7; make buildworld TARGET=arm TARGET_ARCH=armv7
>>
> I _believe_ I did try that, with no success, but the scrollback buffer
> isn't long enough to let me verify it.
>
> Right now the machine is running buildworld, using
> BUILD_ARCH!= echo armv7
> in /usr/src/Makefile.inc1 to set BUILD_ARCH. If the
> buildworld fails I'll try setenv, just to verify. Should know
> late Saturday or early Sunday.
> A more immediate puzzle is how to test whether the buildworld command
> produced an armv7 userland, or something else, before running installworld.
> I'd hate to trash the system _again_. Running clang -v ought to give a
> good hint, if I can find the executable. Is there a better test?
>
For not-installed world executable you can dump ABI version encoded in
.note.tag section:
' objdump -s -j .note.tag /<path to new>/bin/sh'
the right binary have something like this:
-----------------------------------------------------
/bin/sh: file format elf32-littlearm
Contents of section .note.tag:
814c 08000000 04000000 01000000 46726565 ............Free
815c 42534400 b24f1200 08000000 04000000 BSD..O..........
816c 02000000 46726565 42534400 00000000 ....FreeBSD.....
817c 08000000 06000000 03000000 46726565 ............Free
818c 42534400 61726d76 37000000 BSD.armv7...
-------------------------------------------------------
where ' FreeBSD.armv7' is ABI version.
For installed world, the next method is 'make -V MACHINE_ARCH -f/dev/null'
And for kernel, 'sysctl kern.supported_archs' and 'sysctl
hw.machine_arch' shoul return armv7
> Thanks!
>
> bob prohaska
>
>
>
>>>> BUILD_ARCH should end up matching what
>>>> you start under, not the end target as
>>>> I understand things.
>>>>
>>>> If I tried a cross build on a host
>>>> other than armv7 that BUILD_ARCH result
>>>> would be wrong. (I use amd64 for cross
>>>> building aarch64 and arm6 . I'm not up to
>>>> an armv7 capable context yet and may not
>>>> be for some time.)
>>>>
>>>> That change is a local workaround that is
>>>> specific to the environment you are using
>>>> for a build.
>>>>
>>>> This is like my 12.0-CURRENT change being
>>>> specific to my not building on 11.x or 10.x .
>>>> It is just a local workaround that applies
>>>> just to a more limited context.
>>>>
>>>> I've not figured out what is going on (how/why)
>>>> for the != uname behavior that I observed. But
>>>> echoing the expected result instead did work.
>>>>
>>>> None of this is appropriate to check-in.
>>>>
>>>>> I'll stop the buildworld in progress and try it.
>>>> Please report if echo of the host type of context
>>>> appears to make your local environment work vs.
>>>> not. If it does work then it suggests that
>>>> != uname -p did not get the expected text for
>>>> some reason.
>>>>
>>>> Older material:
>>>>
>>>> On Fri, Oct 13, 2017 at 11:36:46AM -0700, Mark Millard wrote:
>>>>> On 2017-Oct-13, at 10:59 AM, bob prohaska <fbsd at www.zefox.net> wrote:
>>>>>
>>>>>> It turns out that simply commenting out lines 447-452 in
>>>>>> /usr/src/Makefile.inc1 allows buildworld to run, even with
>>>>>> no /etc/make.conf in place.
>>>>> For reference:
>>>>>
>>>>> 447 .if make(buildworld)
>>>>> 448 BUILD_ARCH!= uname -p
>>>>> 449 .if ${MACHINE_ARCH} != ${BUILD_ARCH}
>>>>> 450 .error To cross-build, set TARGET_ARCH.
>>>>> 451 .endif
>>>>> 452 .endif
>>>>>
>>>>> (I suggest that the .error message include the
>>>>> MACHINE_ARCH text and the BUILD_ARCH text, probably
>>>>> with ""s around each so that empty is easy to see.)
>>>>>
>>>>>
>>>>> I've had problems with Makefiles using != and uname
>>>>> ending up with the MACRO assigned being an empty string
>>>>> despite a command-line uname returning the expected
>>>>> text.
>>>>>
>>>>> For example I've applied the below local work arounds
>>>>> to my /usr/ports/Mk/bsd.port.mk copy as part of setting
>>>>> up to do amd64 -> aarch64 or amd64 -> armv6 cross
>>>>> builds of ports via poudriere (I've not updated to a
>>>>> armv7-targeting vintage sources yet):
>>>>>
>>>>>
>>>>> # Get the operating system type
>>>>> .if !defined(OPSYS)
>>>>> -OPSYS!= ${UNAME} -s
>>>>> +OPSYS!= echo FreeBSD
>>>>> .endif
>>>>> _EXPORTED_VARS+= OPSYS
>>>>>
>>>>> .if !defined(_OSRELEASE)
>>>>> -_OSRELEASE!= ${UNAME} -r
>>>>> +_OSRELEASE!= echo 12.0-CURRENT
>>>>> .endif
>>>>> _EXPORTED_VARS+= _OSRELEASE
>>>>>
>>>>>
>>>>> I was specifically ending up with _OSRELEASE
>>>>> being empty as seen in poudriere prior to the
>>>>> workaround and that was messing up poudriere
>>>>> such that it stopped with an associated
>>>>> message.
>>>>
>>>> ===
>>>> Mark Millard
>>>> markmi at dsl-only.net
>>>>
>>> _______________________________________________
>>> freebsd-arm@freebsd.org mailing list
>>> https://lists.freebsd.org/mailman/listinfo/freebsd-arm
>>> To unsubscribe, send any mail to "freebsd-arm-unsubscribe@freebsd.org"
>>>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?559584b5-47c6-b9f8-9fae-43ee76fc99cc>