linux-user: Clean up handling of clone() argument order

Linux manages to have three separate orderings of the arguments to
the clone() syscall on different architectures. In the kernel these
are selected via CONFIG_CLONE_BACKWARDS and CONFIG_CLONE_BACKWARDS2.
Clean up our implementation of this to use similar #define names
rather than a TARGET_* ifdef ladder.

This includes behaviour changes fixing bugs on cris, x86-64, m68k,
openrisc and unicore32.  cris had explicit but wrong handling; the
others were just incorrectly using QEMU's default, which happened
to be the equivalent of CONFIG_CLONE_BACKWARDS. (unicore32 appears
to be broken in the mainline kernel in that it tries to use arg3 for
both parent_tidptr and newtls simultaneously -- we don't attempt
to emulate this bug...)

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>

1 files changed, 7 insertions, 0 deletions

diff --git a/linux-user/sparc64/syscall.h b/linux-user/sparc64/syscall.h
index 81a816d..e60bf31 100644
--- a/linux-user/sparc64/syscall.h
+++ b/linux-user/sparc64/syscall.h
@@ -8,3 +8,10 @@ struct target_pt_regs {
 };
 
 #define UNAME_MACHINE "sun4u"
+
+/* SPARC kernels don't define this in their Kconfig, but they have the
+ * same ABI as if they did, implemented by sparc-specific code which fishes
+ * directly in the u_regs() struct for half the parameters in sparc_do_fork()
+ * and copy_thread().
+ */
+#define TARGET_CLONE_BACKWARDS