Next up on creating armv7 MACHINE_ARCH: pre FCP stage
Mark Millard
markmi at dsl-only.net
Thu Jun 15 20:58:23 UTC 2017
[After looking into the details my preliminary
guess seems to be correct: the only dependable
uname output among -m -p -i is for -m for linux.
The uname.c code used varies from distribution
to distribution. Below adds relevant gentoo
uname.c source (a patch source) and gnu source.
Previously I showed the relevant (and distinct)
Ubuntu source.]
On 2017-Jun-15, at 1:15 PM, Mark Millard <markmi at dsl-only.net> wrote:
> On 2017-Jun-15, at 9:58 AM, Mark Millard <markmi at dsl-only.net> wrote:
>
>> On 2017-Jun-15, at 8:40 AM, Mark Millard <markmi at dsl-only.net> wrote:
>>
>>
>>> On 2017-Jun-15, at 5:51 AM, Emmanuel Vadot <manu at bidouilliste.com> wrote:
>>>
>>>> On Thu, 15 Jun 2017 02:08:10 -0700
>>>> Mark Millard <markmi at dsl-only.net> wrote:
>>>>
>>>>> On 2017-Jun-14, at 11:20 PM, Mark Millard <markmi at dsl-only.net> wrote:
>>>>>
>>>>>> On 2017-Jun-14, at 10:22 PM, Warner Losh <imp at bsdimp.com> wrote:
>>>>>>
>>>>>>> . . .
>>>>>>> Comments?
>>>>>>
>>>>>> I booted Ubuntu Mate on a BPI-M3 and tried:
>>>>>>
>>>>>> $ uname -p
>>>>>> armv7l
>>>>>>
>>>>>> $ uname -ap
>>>>>> Linux bpi-iot-ros-ai 3.4.39-BPI-M3-Kernel #1 SMP PREEMPT Tue May 3 13:47:01 UTC 2016 armv7l armv7l armv7l GNU/Linux
>>>>>>
>>>>>> I was actually thinking that a "hf" might
>>>>>> show up in how they name things if it was
>>>>>> a hard float based build. But looking I
>>>>>> see in /lib/ :
>>>>>>
>>>>>> . . .
>>>>>> drwxr-xr-x 3 root root 16384 Nov 4 2016 arm-linux-gnueabihf
>>>>>> . . .
>>>>>> lrwxrwxrwx 1 root root 30 Oct 14 2016 ld-linux-armhf.so.3 -> arm-linux-gnueabihf/ld-2.23.so
>>>>>> lrwxrwxrwx 1 root root 24 Apr 21 2016 ld-linux.so.3 -> /lib/ld-linux-armhf.so.3
>>>>>> . . .
>>>>>>
>>>>>> and in /lib/arm-linux-gnueabihf/ :
>>>>>>
>>>>>> lrwxrwxrwx 1 root root 10 Oct 14 2016 /lib/arm-linux-gnueabihf/ld-linux-armhf.so.3 -> ld-2.23.so
>>>>>>
>>>>>> so it appears armv7l was used for naming a
>>>>>> hard float build in uname -p.
>>>>>>
>>>>>> Of course this does not check how uniform the
>>>>>> various linux distributions are about such
>>>>>> naming.
>>>>>>
>>>>>> Still it may mean that for linux-matching "armv7"
>>>>>> might not be the right name for uname -p output.
>>>>>
>>>>> I tried another linux on the BPI-M3: gentoo .
>>>>>
>>>>> # uname -p
>>>>> ARMv7 Processor rev 5 (v7l)
>>>>>
>>>>> (Wow. Not what I expected.)
>>>>>
>>>>> # uname -pa
>>>>> Linux bananapi 3.4.39-BPI-M3-Kernel #1 SMP PREEMPT Tue May 3 13:47:01 UTC 2016 armv7l ARMv7 Processor rev 5 (v7l) sun8i GNU/Linux
>>>>>
>>>>> # uname -m
>>>>> armv7l
>>>>>
>>>>> # uname -i
>>>>> sun8i
>>>>>
>>>>> # ls -l /lib/ld-*
>>>>> -rwxr-xr-x 1 root root 134192 Mar 26 2016 /lib/ld-2.21.so
>>>>> lrwxrwxrwx 1 root root 10 Mar 26 2016 /lib/ld-linux-armhf.so.3 -> ld-2.21.so
>>>>>
>>>>> So again armv7l seems to be the base name used for
>>>>> a hardfloat little-endian context --although it
>>>>> appears that "uname -m" gives text more likely to
>>>>> be used in testing for how to configure to match
>>>>> the live context. "uname -p" seems far less
>>>>> standardized for its results. The same for
>>>>> "uname -i".
>>>>>
>>>>> ===
>>>>> Mark Millard
>>>>> markmi at dsl-only.net
>>>>
>>>> On both your linux you are using linux-sunxi which is a fork of the
>>>> Allwinner kernel "maintained" by the sunxi community (and it is old).
>>>> To have the proper values of uname one should test running linux
>>>> vanilla kernel.
>>>
>>> They both reported (extracted from the earlier text
>>> that I sent):
>>>
>>> 3.4.39-BPI-M3-Kernel
>>> 3.4.39-BPI-M3-Kernel
>>>
>>> It is the same kernel version from the same group
>>> for the same hardware context as far as what each
>>> reported.
>>>
>>> While they may have varied the kernel for some
>>> reason without changing the version identification
>>> that is not want I would expect.
>>>
>>> I expected it was the Ubuntu vs. Gentoo code that
>>> makes the difference, not the kernel.
>>>
>>> I'm not aware of a modern vanilla kernel for the
>>> BPI-M3.
>>>
>>> From what I can tell for little armv7 boards like
>>> this having older kernels is a common case and
>>> is something ports code would normally deal with
>>> upstream. It is not just sunxi as I understand.
>>>
>>> I may do more experiments and report those too.
>>> My notes are just information for Warner and others
>>> to consider.
>>
>> An FYI:
>>
>> I tried the following on both kernel7.img files
>> (this was via macOS):
>>
>> $ strings /Volumes/BPI-BOOT/kernel7.img | grep -i sun8i | more
>>
>> $ strings /Volumes/BPI-BOOT/kernel7.img | grep -i armv7 | more
>>
>> Both came up empty. The strings reported by uname -p -m -i
>> do not seem to be directly from the kernels.
>
> Summary: for Ubunutu unless HAVE_SYSINFO
> and SI_ARCHITECTURE and SI_PLATFORM are
> defined, it uses struct utsname's machine
> member for all 3 of -m -p -i . This matches
> what Ubuntu Mate reported (all 3 matching).
>
> (I've not looked at Gentoo source yet.)
>
> Supporting details. . .
>
> The following is uname.c source from:
>
> Get:1 http://ports.ubuntu.com/ubuntu-ports xenial/main coreutils 8.25-2ubuntu2 (dsc) [2,071 B]
> Get:2 http://ports.ubuntu.com/ubuntu-ports xenial/main coreutils 8.25-2ubuntu2 (tar) [5,725 kB]
> Get:3 http://ports.ubuntu.com/ubuntu-ports xenial/main coreutils 8.25-2ubuntu2 (diff) [28.0 kB]
>
> Ubuntu's source for uname.c has:
>
> if (toprint
> & (PRINT_KERNEL_NAME | PRINT_NODENAME | PRINT_KERNEL_RELEASE
> | PRINT_KERNEL_VERSION | PRINT_MACHINE))
> {
> struct utsname name; struct utsname name;
> . . .
> if (uname (&name) == -1)
> error (EXIT_FAILURE, errno, _("cannot get system name")); if (toprint & PRINT_MACHINE)
> print_element (name.machine);
> . . .
>
> and later has:
>
> if (toprint & PRINT_PROCESSOR)
> {
> char *element = unknown;
> #if HAVE_SYSINFO && defined SI_ARCHITECTURE
> . . .
> #else
> {
> struct utsname u;
> uname(&u);
> element = u.machine;
> . . .
> }
> #endif
> #ifdef UNAME_PROCESSOR
> if (element == unknown)
> {
> . . .
>
> So it appears that -m and -p are
> explicitly returning the same text
> (No SI_ARCHITECTURE).
>
> Similarly for even later:
>
> if (toprint & PRINT_HARDWARE_PLATFORM)
> {
> char *element = unknown;
> #if HAVE_SYSINFO && defined SI_PLATFORM
> . . .
> #else
> {
> struct utsname u;
> uname(&u);
> element = u.machine;
> if(strlen(element)==4 && element[0]=='i' && element[2]=='8' && element[3
> ]=='6')
> element[1]='3';
> }
> #endif
> #ifdef UNAME_HARDWARE_PLATFORM
> if (element == unknown)
> {
> . . .
>
>
> So it appears that -m and -p and -i
> are explicitly returning the same text
> (No SI_ARCHITECTURE nor SI_PLATFORM).
Turns out that has a patch instead of independent
source. . .
Gentoo patches the gnu uname.c source to be different, including:
(from https://dev.gentoo.org/~polynomial-c/dist/coreutils-8.26-patches-1.1.tar.xz
and its 003_all_coreutils-gentoo-uname.patch )
@@ -250,10 +369,14 @@ main (int argc, char **argv)
if (toprint & PRINT_PROCESSOR)
{
char const *element = unknown;
-#if HAVE_SYSINFO && defined SI_ARCHITECTURE
+#if (HAVE_SYSINFO && defined SI_ARCHITECTURE) || defined (USE_PROCINFO)
{
static char processor[257];
+#if defined (USE_PROCINFO)
+ if (0 <= __linux_procinfo (PROCINFO_PROCESSOR, processor, sizeof processor))
+#else
if (0 <= sysinfo (SI_ARCHITECTURE, processor, sizeof processor))
+#endif
element = processor;
}
#endif
@@ -306,9 +429,13 @@ main (int argc, char **argv)
if (element == unknown)
{
static char hardware_platform[257];
+#if defined (USE_PROCINFO)
+ if (0 <= __linux_procinfo (PROCINFO_HARDWARE_PLATFORM, hardware_platform, sizeof hardware_platform))
+#else
size_t s = sizeof hardware_platform;
static int mib[] = { CTL_HW, UNAME_HARDWARE_PLATFORM };
if (sysctl (mib, 2, hardware_platform, &s, 0, 0) >= 0)
+#endif
element = hardware_platform;
}
#endif
The ubuntu source uname.c source I reported earlier
does not match what is seen via the gnu history at:
http://git.savannah.gnu.org/gitweb/?p=coreutils.git;a=history;f=src/uname.c;h=6378ab7342e5b46e686d719573af99475cc9d69f;hb=3df84f0492a97e07fe4de62bd2ed34e4dd3045ef
I'm using version 8.27 of coreutils source below. . .
if (toprint & PRINT_PROCESSOR)
{
char const *element = unknown;
#if HAVE_SYSINFO && defined SI_ARCHITECTURE
{
static char processor[257];
if (0 <= sysinfo (SI_ARCHITECTURE, processor, sizeof processor))
element = processor;
}
#endif
#ifdef UNAME_PROCESSOR
if (element == unknown)
{
static char processor[257];
size_t s = sizeof processor;
static int mib[] = { CTL_HW, UNAME_PROCESSOR };
if (sysctl (mib, 2, processor, &s, 0, 0) >= 0)
element = processor;
# ifdef __APPLE__
. . .
# endif
}
#endif
if (! (toprint == UINT_MAX && element == unknown))
print_element (element);
}
if (toprint & PRINT_HARDWARE_PLATFORM)
{
char const *element = unknown;
#if HAVE_SYSINFO && defined SI_PLATFORM
{
static char hardware_platform[257];
if (0 <= sysinfo (SI_PLATFORM,
hardware_platform, sizeof hardware_platform))
element = hardware_platform;
}
#endif
#ifdef UNAME_HARDWARE_PLATFORM
if (element == unknown)
{
static char hardware_platform[257];
size_t s = sizeof hardware_platform;
static int mib[] = { CTL_HW, UNAME_HARDWARE_PLATFORM };
if (sysctl (mib, 2, hardware_platform, &s, 0, 0) >= 0)
element = hardware_platform;
}
#endif
if (! (toprint == UINT_MAX && element == unknown))
print_element (element);
}
===
Mark Millard
markmi at dsl-only.net
More information about the freebsd-arm
mailing list