Android: create a generic android_ucontext.h

sandbox/linux/services/android_arm_ucontext.h

Index: sandbox/linux/services/android_arm_ucontext.h
diff --git a/sandbox/linux/services/android_arm_ucontext.h b/sandbox/linux/services/android_arm_ucontext.h
index 0c21e2092796d8ce83100bf8e9876c0a009cc00e..e05e8a8ceba5e2e6a6075d09fee21b108b7a894f 100644
--- a/sandbox/linux/services/android_arm_ucontext.h
+++ b/sandbox/linux/services/android_arm_ucontext.h
@@ -5,10 +5,19 @@
 #ifndef SANDBOX_LINUX_SERVICES_ANDROID_ARM_UCONTEXT_H_
 #define SANDBOX_LINUX_SERVICES_ANDROID_ARM_UCONTEXT_H_
 
+#if !defined(__BIONIC_HAVE_UCONTEXT_T)
 #include <asm/sigcontext.h>
 
-typedef long int greg_t;
-typedef unsigned long sigset_t;
+// We also need greg_t for the sandbox, include it in this header as well.
+typedef unsigned long greg_t;

[palmer, 2013/01/17 01:33:46]

So, it was supposed to have been unsigned before?

[jln (very slow on Chromium), 2013/01/17 01:53:28]

It was wrong before. Someone nice added it to green the tree at some point, not
me :)

+
+// sigset_t from arch/arm/include/asm/signal.h in the Linux kernel.
+typedef struct {
+  unsigned long sig[2];  // 2 is actually _NSIG_WORDS.

[Markus (顧孟勤), 2013/01/17 01:31:34]

Is this correct for 64bit architectures. I thought, the Linux kernel always has
exactly 64bit signals. But an "unsigned long" would likely be 64bit on a 64bit
system. So, it would give us 128 signals.

Or do we just not care and assume that for the foreseeable future, all of the
Android ARM devices are going to be 32bit?

[palmer, 2013/01/17 01:33:46]

Style nit: I'd align these comments. Or, preferably, put them above (rather than
to the right of) the code they comment on.

[palmer, 2013/01/17 01:33:46]

I'd be surprised if that assumption holds for long.

[jln (very slow on Chromium), 2013/01/17 01:53:28]

Yeah I tried to copy exactly what they were doing on arm 32 bits.
arm64 bits will be a new architecture.

On arm64 bits, they'll use the generic UAPI one from
include/uapi/asm-generic/signal.h

I changed it to the UAPI version, since it's more generic.

Moreover, as you can see below, it doesn't matter for the structure definition
anyways (we always build a 32 bytes sigset_t) and I specifically changed the
name of sigset_t to something obscure.

+} android_header_sigset_t;  // Change name to avoid collissions if sigset_t
+                            // gets defined nower Android headers.

[palmer, 2013/01/17 01:33:46]

collisions (one s)

nower --> in newer

[jln (very slow on Chromium), 2013/01/17 01:53:28]

Done.

+
+//typedef unsigned long sigset_t;
 typedef struct ucontext {

[palmer, 2013/01/17 01:33:46]

Style nit: Since we are poking this code and not blindly copying it from
upstream, it'd be nice to format it according to Chromium style guide. The
upstream style is kind of nasty...

[jln (very slow on Chromium), 2013/01/17 01:53:28]

I'll keep it as is if you don't mind I would really rather touch it as little as
I can.
Also it makes it easier for people to compare to the original etc..

   unsigned long   uc_flags;
   struct ucontext  *uc_link;
@@ -16,9 +25,14 @@ typedef struct ucontext {
   struct sigcontext uc_mcontext;
   sigset_t    uc_sigmask;
   /* Allow for uc_sigmask growth.  Glibc uses a 1024-bit sigset_t.  */
-  int     __not_used[32 - (sizeof (sigset_t) / sizeof (int))];
+  int     __not_used[32 - (sizeof (android_header_sigset_t) / sizeof (int))];
   /* Last for extensibility.  Eight byte aligned because some
      coprocessors require eight byte alignment.  */
   unsigned long   uc_regspace[128] __attribute__((__aligned__(8)));
 } ucontext_t;
+
+#else
+#include <sys/ucontext.h>
+#endif  // __BIONIC_HAVE_UCONTEXT_T
+
 #endif  // SANDBOX_LINUX_SERVICES_ANDROID_ARM_UCONTEXT_H_