qemu m68k/mcf5208: problem with signal handler
Dear kernel hackers,
I have a problem with the signal handling under qemu-system-m68k emulating coldfire mcf5208 evalboard. Following example code provided by Busybox maintainer Denys Vlasenko
shows the problem when running on qemu:
#include <unistd.h>
#include <signal.h>
#include <sys/types.h>
#include <sys/wait.h>
static void sighandler(int sig)
{
write(1, "SIGNAL\n", 7);
}
int main()
{
int pid;
write(1, "VFORK1\n", 7);
pid = vfork();
if (pid == 0) {
write(1, "EXIT1\n", 6);
_exit(1);
}
wait(NULL);
signal(SIGCHLD, sighandler);
write(1, "VFORK2\n", 7);
pid = vfork();
if (pid == 0) {
write(1, "EXIT2\n", 6);
_exit(1);
}
wait(NULL);
write(1, "EXIT\n", 5);
return 0;
} You can generate a bootable image with latest buildroot, which shows the issue:
$ git clone git://git.buildroot.net/buildroot $ cd buildroot; make qemu_m68k_mcf5208_defconfig; make $ qemu-system-m68k -M mcf5208evb -cpu m5208 -kernel output/images/vmlinux -nographic
Every command forked from busybox hush shell will lead into a segmentation fault.
I added following printk to start investigating the problem:
arch/m68k/kernel/signal.c +1 -0
@@ -595,6 +595,7 @@ void __user *fp) {int fsize =frame_extra_sizes (formatvec >>12 ); +printk ("avoid broken signal handler...\n" );if (fsize <0 ) {/* * user process trying to return with weird frame format
But now the problem disappeared. :/
What do you think? Is it a Kernel bug or a C library problem?
Busybox hush otherwise works fine for other noMMU targets as stm32 evalboard with cortex-m4. It also works in Qemu M68k emulating Q800 full MMU system.
Thanks for any ideas,
Waldemar
http://lists.busybox.net/pipermail/busybox/2014-September/081659.html
Hi,
forgot to add Greg in CC.
And sorry for the whitespace fuckup in the example code.
Waldemar Brodkorb wrote,
Quoted 6 lines
Dear kernel hackers,
I have a problem with the signal handling under qemu-system-m68k emulating coldfire mcf5208 evalboard. Following example code provided by Busybox maintainer Denys Vlasenko
shows the problem when running on qemu:
[ .. ]
Quoted 33 lines
You can generate a bootable image with latest buildroot, which shows the issue:
$ git clone git://git.buildroot.net/buildroot $ cd buildroot; make qemu_m68k_mcf5208_defconfig; make $ qemu-system-m68k -M mcf5208evb -cpu m5208 -kernel output/images/vmlinux -nographicEvery command forked from busybox hush shell will lead into a segmentation fault.
I added following printk to start investigating the problem:
diff -Nurlinux-4.5.3.orig/arch/m68k/kernel/signal.clinux-4.5.3/arch/m68k/kernel/signal.carch/m68k/kernel/signal.c +1
-0@@ -595,6 +595,7 @@ void __user *fp) {int fsize =frame_extra_sizes (formatvec >>12 ); +printk ("avoid broken signal handler...\n" );if (fsize <0 ) {/* * user process trying to return with weird frame format But now the problem disappeared. :/
What do you think? Is it a Kernel bug or a C library problem?
Busybox hush otherwise works fine for other noMMU targets as stm32 evalboard with cortex-m4. It also works in Qemu M68k emulating Q800 full MMU system.
Thanks for any ideas,
Waldemarhttp://lists.busybox.net/pipermail/busybox/2014-September/081659.html
Hi Waldemar,
Quoted 36 lines · Waldemar Brodkorb
Hi,
forgot to add Greg in CC.
And sorry for the whitespace fuckup in the example code.Waldemar Brodkorb wrote,
Quoted 6 lines
Dear kernel hackers,
I have a problem with the signal handling under qemu-system-m68k emulating coldfire mcf5208 evalboard. Following example code provided by Busybox maintainer Denys Vlasenko
shows the problem when running on qemu:[ .. ]
Quoted 20 lines
You can generate a bootable image with latest buildroot, which shows the issue:
$ git clone git://git.buildroot.net/buildroot $ cd buildroot; make qemu_m68k_mcf5208_defconfig; make $ qemu-system-m68k -M mcf5208evb -cpu m5208 -kernel output/images/vmlinux -nographicEvery command forked from busybox hush shell will lead into a segmentation fault.
I added following printk to start investigating the problem:
diff -Nurlinux-4.5.3.orig/arch/m68k/kernel/signal.clinux-4.5.3/arch/m68k/kernel/signal.carch/m68k/kernel/signal.c +1
-0@@ -595,6 +595,7 @@ void __user *fp) {int fsize =frame_extra_sizes (formatvec >>12 ); +printk ("avoid broken signal handler...\n" );if (fsize <0 ) {/* * user process trying to return with weird frame format
Quoted 3 lines
Quoted 3 lines
But now the problem disappeared. :/
What do you think? Is it a Kernel bug or a C library problem?
What version of linux kernel?
What version of gcc?
This sounds a lot like the problem I fixed in linux commit a9551799 ("m68k: Use conventional function parameters for do_sigreturn").
Definitely try that first.
Regards
Greg
Quoted 10 lines
Quoted 9 lines
Busybox hush otherwise works fine for other noMMU targets as stm32 evalboard with cortex-m4. It also works in Qemu M68k emulating Q800 full MMU system.
Thanks for any ideas,
Waldemarhttp://lists.busybox.net/pipermail/busybox/2014-September/081659.html
Hi Greg,
Greg Ungerer wrote,
Quoted 3 lines
Quoted 1 line
Quoted 1 line
What do you think? Is it a Kernel bug or a C library problem?
What version of linux kernel?
4.5.3
Quoted 1 line
What version of gcc?
4.9.3
Quoted 4 lines
This sounds a lot like the problem I fixed in linux commit
a9551799("m68k: Use conventional function parameters fordo_sigreturn").Definitely try that first.
Jackpot. Thanks a lot. This works!
May be some candidate for LTS kernels.
best regards
Waldemar