Merge remote-tracking branch 'osvm/Cog' into Issue3571

build.macos64x64/gdbarm64/conf.COG

@@ -10,5 +10,5 @@ done
 for d in sim/common sim/aarch64; do
 	(cd $d;
 	 ../$S/$d/configure COG=1 CFLAGS="$CFLAGS" $CONFFLAGS \
-		--disable-sim-trace --without-profile)
+		--disable-sim-trace)
 done

platforms/Cross/plugins/SqueakFFIPrims/sqFFITestFuncs.c

@@ -51,7 +51,7 @@ typedef struct Si2 {int a,b;} ffiTestSi2;
 typedef struct Ss2 {short a,b;} ffiTestSs2;
 typedef struct Ssi {short a; int b;} ffiTestSsi;
 typedef struct Sfi {float a; int b;} ffiTestSfi;
-typedef struct Sfd {float a; double b} ffiTestSfd;
+typedef struct Sfd {float a; double b;} ffiTestSfd;
 typedef struct Sdi {double a; int b;} ffiTestSdi;
 typedef struct Ssf {short a; float b;} ffiTestSsf;
 typedef struct SsSsi {short a; struct Ssi b;} ffiTestSsSsi;
@@ -66,6 +66,13 @@ typedef struct Sf4 {float a,b,c,d;} ffiTestSf4;
 typedef struct Ss4 {short a,b,c,d;} ffiTestSs4;
 typedef struct SSdi5 {struct Sdi a,b,c,d,e;} ffiTestSSdi5; /* a structure longer than 8 eightBytes */
 
+typedef union  Ufi {float a; int b;} ffiTestUfi;
+typedef union  Ufd {float a; double b;} ffiTestUfd;
+typedef union  UdSi2 {double a; struct Si2 b;} ffiTestUdSi2;
+
+typedef struct SUfdUfi {union Ufd a; union Ufi b;} ffiTestSUfdUfi;
+typedef struct SUfdUdSi2 {union Ufd a; union UdSi2 b;} ffiTestSUfdUdSi2;
+
 #pragma export on
 
 EXPORT(ffiTestSd2)  ffiTestInitSd2  (double    a,double    b) { ffiTestSd2 v = {a,b}; return v; }
@@ -90,6 +97,15 @@ EXPORT(ffiTestSf4)  ffiTestInitSf4  (float     a,float     b, float  c,float d)
 EXPORT(ffiTestSs4)  ffiTestInitSs4  (short     a,short     b, short  c,short d) { ffiTestSs4 v = {a,b,c,d}; return v; }
 EXPORT(ffiTestSSdi5)ffiTestInitSSdi5(struct Sdi a,struct Sdi b,struct Sdi c, struct Sdi d,struct Sdi e) { ffiTestSSdi5 v = {a,b,c,d,e}; return v; }
 
+EXPORT(ffiTestUfi)  ffiTestInitUfi_f(float  a) { ffiTestUfi v; v.a=a; return v; }
+EXPORT(ffiTestUfi)  ffiTestInitUfi_i(int    b) { ffiTestUfi v; v.b=b; return v; }
+EXPORT(ffiTestUfd)  ffiTestInitUfd_f(float  a) { ffiTestUfd v; v.a=a; return v; }
+EXPORT(ffiTestUfd)  ffiTestInitUfd_d(double b) { ffiTestUfd v; v.b=b; return v; }
+EXPORT(ffiTestUdSi2) ffiTestInitUdSi2_d(double a) { ffiTestUdSi2 v; v.a=a; return v; }
+EXPORT(ffiTestUdSi2) ffiTestInitUdSi2_ii(int a,int b) { ffiTestUdSi2 v; v.b.a=a; v.b.b=b; return v; }
+EXPORT(ffiTestSUfdUfi)  ffiTestInitSUfdUfi  (union Ufd a,union Ufi b) { ffiTestSUfdUfi v = {a,b}; return v; }
+EXPORT(ffiTestSUfdUdSi2)  ffiTestInitSUfdUdSi2  (union Ufd a,union UdSi2 b) { ffiTestSUfdUdSi2 v = {a,b}; return v; }
+
 EXPORT(ffiTestSd2)  ffiTestReturnSd2  () { return ffiTestInitSd2(1.0 ,2.0 ); }
 EXPORT(ffiTestSf2)  ffiTestReturnSf2  () { return ffiTestInitSf2(1.0f,2.0f); }
 EXPORT(ffiTestSl2)  ffiTestReturnSl2  () { return ffiTestInitSl2(1LL,2LL); }
@@ -131,6 +147,8 @@ EXPORT(double) ffiTestSumSslf(ffiTestSslf x) { return (double) x.a + (double) x.
 EXPORT(double) ffiTestSumSslf_2(ffiTestSslf x,ffiTestSslf y) { return ffiTestSumSslf(x) + ffiTestSumSslf(y); }
 EXPORT(double) ffiTestSumSslf_4(ffiTestSslf x,ffiTestSslf y,ffiTestSslf z,ffiTestSslf t) { return ffiTestSumSslf_2(x,y) + ffiTestSumSslf_2(z,t); }
 
+EXPORT(double) ffiTestSumSUfdUfi_f(ffiTestSUfdUfi x) {return (double) x.a.a + (double) x.b.a; }; /* sum the float parts */
+EXPORT(double) ffiTestSumSUfdUdSi2_d(ffiTestSUfdUdSi2 x) {return x.a.b + x.b.a; }; /* sum the double parts */
 
 EXPORT(char) ffiTestChars(char c1, char c2, char c3, char c4);
 EXPORT(short) ffiTestShorts(short c1, short c2, short c3, short c4);

platforms/Cross/vm/sq.h

@@ -102,7 +102,7 @@ extern void sqDeallocateMemorySegmentAtOfSize(void *addr, sqInt sz);
 #endif /* SPURVM */
 #if COGVM
 extern void sqMakeMemoryExecutableFromToCodeToDataDelta(usqInt, usqInt, sqInt*);
-extern void sqMakeMemoryNotExecutableFromTo(unsigned long, unsigned long);
+extern void sqMakeMemoryNotExecutableFromTo(usqInt, usqInt);
 #endif
 
 /* Platform-dependent memory size adjustment macro. */

platforms/Cross/vm/sqCogStackAlignment.h

@@ -92,7 +92,7 @@
 #	define setsp(spval) __asm volatile ("mov sp, %0"  : : "r"(spval))
 
 #  endif
-# elif defined(__arm32__) || defined(ARM32)
+# elif defined(__arm__) || defined(__arm32__) || defined(ARM32)
 	/* http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0041c/Cegbidie.html
 	 * ARM DUI 0041C Page 9-7
 	 */

platforms/Mac OS/vm/sqMacMemory.c

@@ -121,26 +121,89 @@ sqMacMemoryFree()
 }
 
 #if COGVM
+#   if DUAL_MAPPED_CODE_ZONE
+/* We are indebted to Chris Wellons who designed this elegant API for dual
+ * mapping which we depend on for fine-grained code modification (classical
+ * Deutsch/Schiffman style inline cacheing and derivatives).  Chris's code is:
+	https://nullprogram.com/blog/2016/04/10/
+ *
+ * To cope with modern OSs that disallow executing code in writable memory we
+ * dual-map the code zone, one mapping with read/write permissions and the other
+ * with read/execute permissions. In such a configuration the code zone has
+ * already been alloated and is not included in (what is no longer) the initial
+ * alloc.
+ */
+static void
+memory_alias_map(size_t size, size_t naddr, void **addrs)
+{
+extern char  *exeName;
+	char path[128];
+	snprintf(path, sizeof(path), "/%s(%lu,%p)",
+			 exeName ? exeName : __FUNCTION__, (long)getpid(), addrs);
+	int fd = shm_open(path, O_RDWR | O_CREAT | O_EXCL, 0600);
+	if (fd == -1) {
+		perror("memory_alias_map: shm_open");
+		exit(0666);
+	}
+	shm_unlink(path);
+	ftruncate(fd, size);
+	for (size_t i = 0; i < naddr; i++) {
+		addrs[i] = mmap(addrs[i], size, PROT_READ | PROT_WRITE,
+								addrs[i] ? MAP_FIXED | MAP_SHARED : MAP_SHARED,
+								fd, 0);
+		if (addrs[i] == MAP_FAILED) {
+			perror("memory_alias_map: mmap(addrs[i]...");
+			exit(0667);
+		}
+	}
+	close(fd);
+	return;
+}
+#   endif /* DUAL_MAPPED_CODE_ZONE */
 void
-sqMakeMemoryExecutableFromTo(unsigned long startAddr, unsigned long endAddr)
+sqMakeMemoryExecutableFromToCodeToDataDelta(usqInt startAddr,
+											usqInt endAddr,
+											sqInt *codeToDataDelta)
 {
-	unsigned long firstPage = roundDownToPage(startAddr);
+	usqInt firstPage = roundDownToPage(startAddr);
+	usqInt size = endAddr - firstPage;
+
+#  if DUAL_MAPPED_CODE_ZONE
+	usqInt mappings[2];
+
+	mappings[0] = firstPage;
+	mappings[1] = 0;
+
+	memory_alias_map(size, 2, (void **)mappings);
+	assert(mappings[0] == firstPage);
+	*codeToDataDelta = mappings[1] - startAddr;
+
 	if (mprotect((void *)firstPage,
-				 endAddr - firstPage + 1,
+				 size,
+				 PROT_READ | PROT_EXEC) < 0)
+		perror("mprotect(x,y,PROT_READ | PROT_EXEC)");
+
+#  else /* DUAL_MAPPED_CODE_ZONE */
+
+	if (mprotect((void *)firstPage,
+				 size,
 				 PROT_READ | PROT_WRITE | PROT_EXEC) < 0)
 		perror("mprotect(x,y,PROT_READ | PROT_WRITE | PROT_EXEC)");
+
+	*codeToDataDelta = 0;
+#  endif
 }
 
 void
-sqMakeMemoryNotExecutableFromTo(unsigned long startAddr, unsigned long endAddr)
+sqMakeMemoryNotExecutableFromTo(usqInt startAddr, usqInt endAddr)
 {
 # if 0
 	/* We get EACCESS on 10.6.3 when trying to disable exec perm; Why? */
 	/* Arguably this is pointless since allocated memory always does include
 	 * write permission.  Annoyingly the mprotect call fails on both linux &
 	 * mac os x.  So make the whole thing a nop.
 	 */
-	unsigned long firstPage = roundDownToPage(startAddr);
+	usqInt firstPage = roundDownToPage(startAddr);
 	if (mprotect((void *)firstPage,
 				 endAddr - firstPage + 1,
 				 PROT_READ | PROT_WRITE) < 0

platforms/iOS/vm/Common/sqMacV2Memory.c

@@ -188,19 +188,20 @@ sqImageFileReadEntireImage(void *ptr, size_t elementSize, size_t count, sqImageF
 # define roundUpToPage(v) (((v)+pageSize-1)&pageMask)
 #if COGVM || defined(HAVE_NATIVEBOOST) 
 void
-sqMakeMemoryExecutableFromTo(unsigned long startAddr, unsigned long endAddr)
+sqMakeMemoryExecutableFromToCodeToDataDelta(usqInt startAddr, usqInt endAddr, sqInt *codeToDataDelta)
 {
-	unsigned long firstPage = roundDownToPage(startAddr);
+	usqInt firstPage = roundDownToPage(startAddr);
 	if (mprotect((void *)firstPage,
 				 roundUpToPage(endAddr - firstPage),
 				 PROT_READ | PROT_WRITE | PROT_EXEC) < 0)
 		perror("mprotect(x,y,PROT_READ | PROT_WRITE | PROT_EXEC)");
+	*codeToDataDelta = 0;
 }
 
 void
-sqMakeMemoryNotExecutableFromTo(unsigned long startAddr, unsigned long endAddr)
+sqMakeMemoryNotExecutableFromTo(usqInt startAddr, usqInt endAddr)
 {
-	unsigned long firstPage = roundDownToPage(startAddr);
+	usqInt firstPage = roundDownToPage(startAddr);
 	if (mprotect((void *)firstPage,
 				 roundUpToPage(endAddr - firstPage),
 				 PROT_READ | PROT_WRITE) < 0)

platforms/minheadless/unix/sqPlatformSpecific-Unix.h

@@ -42,9 +42,6 @@ extern sqInt sqGrowMemoryBy(sqInt oldLimit, sqInt delta);
 extern sqInt sqShrinkMemoryBy(sqInt oldLimit, sqInt delta);
 extern sqInt sqMemoryExtraBytesLeft(sqInt includingSwap);
 #if COGVM
-extern void sqMakeMemoryExecutableFromTo(unsigned long, unsigned long);
-extern void sqMakeMemoryNotExecutableFromTo(unsigned long, unsigned long);
-
 extern int osCogStackPageHeadroom(void);
 extern void reportMinimumUnusedHeadroom(void);
 #endif

platforms/minheadless/unix/sqUnixMemory.c

@@ -322,17 +322,20 @@ sqInt sqMemoryExtraBytesLeft(sqInt includingSwap)			{ return uxMemoryExtraBytesL
 #define roundUpToPage(v) (((v)+pageSize-1)&pageMask)
 #if COGVM
 void
-sqMakeMemoryExecutableFromTo(unsigned long startAddr, unsigned long endAddr)
+sqMakeMemoryExecutableFromToCodeToDataDelta(usqInt startAddr,
+											usqInt endAddr,
+											sqInt *codeToDataDelta)
 {
-	unsigned long firstPage = roundDownToPage(startAddr);
+	usqInt firstPage = roundDownToPage(startAddr);
 	if (mprotect((void *)firstPage,
 				 endAddr - firstPage + 1,
 				 PROT_READ | PROT_WRITE | PROT_EXEC) < 0)
 		perror("mprotect(x,y,PROT_READ | PROT_WRITE | PROT_EXEC)");
+	*codeToDataDelta = 0;
 }
 
 void
-sqMakeMemoryNotExecutableFromTo(unsigned long startAddr, unsigned long endAddr)
+sqMakeMemoryNotExecutableFromTo(usqInt startAddr, usqInt endAddr)
 {
 # if 0
 	unsigned long firstPage = roundDownToPage(startAddr);

platforms/minheadless/unix/sqUnixSpurMemory.c

@@ -240,10 +240,10 @@ sqMakeMemoryExecutableFromToCodeToDataDelta(usqInt startAddr,
 }
 
 void
-sqMakeMemoryNotExecutableFromTo(unsigned long startAddr, unsigned long endAddr)
+sqMakeMemoryNotExecutableFromTo(usqInt startAddr, usqInt endAddr)
 {
-	unsigned long firstPage = roundDownToPage(startAddr);
-	unsigned long size = endAddr - firstPage;
+	usqInt firstPage = roundDownToPage(startAddr);
+	usqInt size = endAddr - firstPage;
 	/* Arguably this is pointless since allocated memory always includes write
 	 * permission by default.  Annoyingly the mprotect call fails on both linux
 	 * and mac os x.  So make the whole thing a nop.

platforms/minheadless/windows/sqPlatformSpecific-Win32.h

@@ -134,9 +134,6 @@ sqInt ioSetCursorARGB(sqInt bitsIndex, sqInt w, sqInt h, sqInt x, sqInt y);
 #define PROF_THREAD_PRIORITY THREAD_PRIORITY_TIME_CRITICAL
 
 #if COGVM
-extern void sqMakeMemoryExecutableFromTo(usqIntptr_t, usqIntptr_t);
-extern void sqMakeMemoryNotExecutableFromTo(usqIntptr_t, usqIntptr_t);
-
 extern int osCogStackPageHeadroom(void);
 extern void reportMinimumUnusedHeadroom(void);
 #endif

platforms/minheadless/windows/sqWin32Alloc.c

@@ -181,7 +181,7 @@ sqMemoryExtraBytesLeft(int includingSwap) {
 
 # if COGVM
 void
-sqMakeMemoryExecutableFromTo(usqIntptr_t startAddr, usqIntptr_t endAddr)
+sqMakeMemoryExecutableFromToCodeToDataDelta(usqInt startAddr, usqInt endAddr, sqInt *codeToDataDelta)
 {
 	DWORD previous;
 
@@ -190,10 +190,11 @@ sqMakeMemoryExecutableFromTo(usqIntptr_t startAddr, usqIntptr_t endAddr)
 						PAGE_EXECUTE_READWRITE,
 						&previous))
 		sqWin32PrintLastError("VirtualProtect(x,y,PAGE_EXECUTE_READWRITE)");
+	*codeToDataDelta = 0;
 }
 
 void
-sqMakeMemoryNotExecutableFromTo(usqIntptr_t startAddr, usqIntptr_t endAddr)
+sqMakeMemoryNotExecutableFromTo(usqInt startAddr, usqInt endAddr)
 {
 	DWORD previous;
 

platforms/minheadless/windows/sqWin32SpurAlloc.c

@@ -214,7 +214,7 @@ sqMakeMemoryExecutableFromToCodeToDataDelta(usqInt startAddr,
 }
 
 void
-sqMakeMemoryNotExecutableFromTo(usqIntptr_t startAddr, usqIntptr_t endAddr)
+sqMakeMemoryNotExecutableFromTo(usqInt startAddr, usqInt endAddr)
 {
 	DWORD previous;
     SIZE_T size;

platforms/unix/vm/sqUnixMain.c

@@ -942,12 +942,12 @@ reportStackState(char *msg, char *date, int printAll, ucontext_t *uap)
 # elif __sun__ && __i386__
 			void *fp = (void *)(uap ? uap->uc_mcontext.gregs[REG_FP] : 0);
 			void *sp = (void *)(uap ? uap->uc_mcontext.gregs[REG_SP] : 0);
-# elif defined(__arm32__) || defined(ARM32)
-			void *fp = (void *)(uap ? uap->uc_mcontext.arm_fp : 0);
-			void *sp = (void *)(uap ? uap->uc_mcontext.arm_sp : 0);
 # elif defined(__arm64__) || defined(__aarch64__) || defined(ARM64)
 			void *fp = (void *)(uap ? uap->uc_mcontext.regs[29] : 0);
 			void *sp = (void *)(uap ? uap->uc_mcontext.sp : 0);
+# elif defined(__arm__) || defined(__arm32__) || defined(ARM32)
+			void *fp = (void *)(uap ? uap->uc_mcontext.arm_fp : 0);
+			void *sp = (void *)(uap ? uap->uc_mcontext.arm_sp : 0);
 # else
 #	error need to implement extracting pc from a ucontext_t on this system
 # endif