Re: [PATCH] Wrong unpacked structure for epoll_event on qemu-or1k (openrisc user-space)

[Luca Bonissi <qemu@bonslack.org>]

On 19/07/23 10:49, Laurent Vivier wrote:
> 
> According to linux/glibc sourced, epoll is only packed for x86_64.

And, in recent glibc, also for i386, even it seems not necessary: even 
if the __alignof__(long long) is 8, structures like epoll_event are only 
12 bytes, maybe "packed" for historical reasons. Ancient i386 gcc[s] 
(<3.0.0) have 4 bytes for __alignof__(long long).

> Perhaps the default alignment of long is not correctly defined in qemu 
> for openrisc?

__alignof__(long long):
- 8 bytes: all 64 bit targets + arm, hppa, mips, ppc, sparc, xtensa, x86
- 4 bytes: microblaze, nios2, or1k, sh4
- 2 bytes: m68k
- 1 byte : cris

offsetof(struct epoll_event,data):
- 8: all 64 bit targets + arm, hppa, mips, ppc, sparc, xtensa
- 4: cris, m68k, microblaze, nios2, or1k, sh4, x86

So, epoll_event is "naturally" packed on the following targets (checked 
in linux-user container and/or with cross-compiler):
- cris, m68k, microblaze, nios2, or1k, sh4, x86 (32bit)

> See include/exec/user/abitypes.h to update the value.

OK, abitypes.h should be updated with the following patch (discard the 
previous patch on syscall_defs.h):

Signed-off-by: Luca Bonissi <qemu@bonslack.org>
---

diff -up a/include/exec/user/abitypes.h b/include/exec/user/abitypes.h
--- a/include/exec/user/abitypes.h	2023-03-27 15:41:42.511916232 +0200
+++ b/include/exec/user/abitypes.h	2023-07-19 12:09:03.001687788 +0200
@@ -15,7 +15,15 @@
  #define ABI_LLONG_ALIGNMENT 2
  #endif

+#ifdef TARGET_CRIS
+#define ABI_SHORT_ALIGNMENT 1
+#define ABI_INT_ALIGNMENT 1
+#define ABI_LONG_ALIGNMENT 1
+#define ABI_LLONG_ALIGNMENT 1
+#endif
+
-#if (defined(TARGET_I386) && !defined(TARGET_X86_64)) || 
defined(TARGET_SH4)
+#if (defined(TARGET_I386) && !defined(TARGET_X86_64)) || 
defined(TARGET_SH4) || \
+    defined(TARGET_OPENRISC) || defined(TARGET_NIOS2) || 
defined(TARGET_MICROBLAZE)
  #define ABI_LLONG_ALIGNMENT 4
  #endif