feature request: force explicit machine in `uname(2)` and `uname(1)`

marked this issue as related to adelie/packages#1005 (closed)

See also:

Originally introduced for ARM as:

commit 3dd681d944f6d861f12ee03aff17a14342963330
Author: Will Deacon <will@kernel.org>
Date:   Mon Mar 5 11:49:32 2012 +0000

    arm64: 32-bit (compat) applications support
    
    This patch adds support for 32-bit applications. The vectors page is a
    binary blob mapped into the application user space at 0xffff0000 (the
    AArch64 toolchain does not support compilation of AArch32 code). Full
    compatibility with ARMv7 user space is supported. The use of deprecated
    ARMv7 functionality (SWP, CP15 barriers) has been disabled by default on
    AArch64 kernels and unaligned LDM/STM is not supported.
    
    Please note that only the ARM 32-bit EABI is supported, so no OABI
    compatibility.
    
    Signed-off-by: Will Deacon <will.deacon@arm.com>
    Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
    Acked-by: Tony Lindgren <tony@atomide.com>
    Acked-by: Nicolas Pitre <nico@linaro.org>
    Acked-by: Olof Johansson <olof@lixom.net>
    Acked-by: Santosh Shilimkar <santosh.shilimkar@ti.com>

And/or some hack like this, except with variables:

#!/bin/sh
/bin/uname $@ | sed -e 's/armv8l/armv7l/

LD_PRELOAD example from skarnet: unamespoofer.c

Machine strings for 32-bit personality from the kernel source as of writing:

$ grep -r COMPAT_UTS_MACHINE . | grep arch | awk '{print $3}' | sort | uniq
"armv8b\0\0"
"armv8l\0\0"
"i686\0\0"
"mips\0\0\0"
"parisc\0\0"
"ppc\0\0"
"ppcle\0\0"
"riscv\0\0"
"s390\0\0\0\0"
"sparc\0\0"

I think we need:

armv8b --> armv7b
armv8l --> armv7l
i686 --> i586
riscv --> riscv32

Here is an old version of a uname wrapper:

diff --git a/autobuilder b/autobuilder
index 112935e..e9262c6 100755
--- a/autobuilder
+++ b/autobuilder
@@ -981,6 +983,33 @@ export JAVA_HOME=/usr/lib/jvm/java-1.8-openjdk
 export LIBTOOL=slibtool
 EOF

+    ##
+    # Create a 'uname(1)' wrapper to handle 32-bit edge cases.
+    # From the Linux source as of writing:
+    #
+    #   $ grep -r COMPAT_UTS_MACHINE . | grep arch | awk '{print $3}' | sort | uniq
+    #   "armv8b\0\0"
+    #   "armv8l\0\0"
+    #   "i686\0\0"
+    #   "mips\0\0\0"
+    #   "parisc\0\0"
+    #   "ppc\0\0"
+    #   "ppcle\0\0"
+    #   "riscv\0\0"
+    #   "s390\0\0\0\0"
+    #   "sparc\0\0"
+    #
+    cat > "${KEEP}/${k}/${TEMP}/usr/local/bin/uname" <<"EOF"
+#!/bin/sh
+/bin/uname $@ | sed \
+    -e 's/\barmv8b\b/armv7b/g' \
+    -e 's/\barmv8l\b/armv7l/g' \
+    -e 's/\bi686\b/i586/g' \
+    -e 's/\briscv\b/riscv32/g' \
+    ;
+EOF
+    chmod +x "${KEEP}/${k}/${TEMP}/usr/local/bin/uname";
+
     ##
     # Create a small text file with the CPU type for self-ref.
     #

mentioned in commit 98877b44

marked this issue as related to adelie/packages#1036 (closed)

mentioned in issue adelie/packages#1005 (closed)

Additional hacks need to be implemented. For example, /proc/cpuinfo.

See: adelie/packages#1035

Examples: https://gist.github.com/search?q=%2Fproc%2Fcpuinfo

This can probably be accomplished by, before going into the chroot, doing a sed -e ... -e ... blah on the host, and bind-mounting this file.

Example from the HR350A:

root@aarch64-1:/# head -n 9 /proc/cpuinfo
processor       : 0
BogoMIPS        : 80.00
Features        : fp asimd evtstrm aes pmull sha1 sha2 crc32 cpuid
CPU implementer : 0x50
CPU architecture: 8
CPU variant     : 0x3
CPU part        : 0x000
CPU revision    : 2

and

root@aarch64-1:/# setarch linux32 head -n 9 /proc/cpuinfo
processor       : 0
model name      : ARMv8 Processor rev 2 (v8l)
BogoMIPS        : 80.00
Features        : half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt lpae evtstrm aes pmull sha1 sha2 crc32
CPU implementer : 0x50
CPU architecture: 8
CPU variant     : 0x3
CPU part        : 0x000
CPU revision    : 2

so the 32-bit personality updates the Features record, but not the CPU architecture one. Therefore, we must be careful to collect /proc/cpuinfo from the appropriate environment before creating a fake one.

marked this issue as related to adelie/packages#1035

marked this issue as related to adelie/packages#1026

marked this issue as related to #36

closed with commit 9a166c42

mentioned in issue adelie/packages#1227

mentioned in issue #49 (closed)

feature request: force explicit machine in `uname(2)` and `uname(1)`

Child items ...

Activity