Merge pull request #84 from snaury/fix-sparc

Fix stack switching on sparc

greenlet.c

@@ -308,6 +308,50 @@ static int (*g_initialstub)(void*);
 } while(0)
 #endif
 
+/*
+ * the following macros are spliced into the OS/compiler
+ * specific code, in order to simplify maintenance.
+ */
+
+#define SLP_SAVE_STATE(stackref, stsizediff)            \
+	stackref += STACK_MAGIC;                        \
+	if (slp_save_state((char*)stackref)) return -1; \
+	if (!PyGreenlet_ACTIVE(ts_target)) return 1;    \
+	stsizediff = ts_target->stack_start - (char*)stackref
+
+#define SLP_RESTORE_STATE()                             \
+	slp_restore_state()
+
+
+#define SLP_EVAL
+#define slp_switch GREENLET_NOINLINE(slp_switch)
+#include "slp_platformselect.h"
+#undef slp_switch
+
+#ifndef STACK_MAGIC
+#error "greenlet needs to be ported to this platform, or teached how to detect your compiler properly."
+#endif /* !STACK_MAGIC */
+
+#ifdef EXTERNAL_ASM
+/* CCP addition: Make these functions, to be called from assembler.
+ * The token include file for the given platform should enable the
+ * EXTERNAL_ASM define so that this is included.
+ */
+
+intptr_t slp_save_state_asm(intptr_t *ref) {
+	intptr_t diff;
+	SLP_SAVE_STATE(ref, diff);
+	return diff;
+}
+
+void slp_restore_state_asm(void) {
+	SLP_RESTORE_STATE();
+}
+
+extern int slp_switch(void);
+
+#endif
+
 /***********************************************************/
 
 static int g_save(PyGreenlet* g, char* stop)
@@ -345,6 +389,10 @@ static void GREENLET_NOINLINE(slp_restore_state)(void)
 	PyGreenlet* g = ts_target;
 	PyGreenlet* owner = ts_current;
 
+#ifdef SLP_BEFORE_RESTORE_STATE
+	SLP_BEFORE_RESTORE_STATE();
+#endif
+
 	/* Restore the heap copy back into the C stack */
 	if (g->stack_saved != 0) {
 		memcpy(g->stack_start, g->stack_copy, g->stack_saved);
@@ -370,6 +418,10 @@ static int GREENLET_NOINLINE(slp_save_state)(char* stackref)
 	else
 		owner->stack_start = stackref;
 
+#ifdef SLP_BEFORE_SAVE_STATE
+	SLP_BEFORE_SAVE_STATE();
+#endif
+
 	while (owner->stack_stop < target_stop) {
 		/* ts_current is entierely within the area to free */
 		if (g_save(owner, owner->stack_stop))
@@ -383,51 +435,6 @@ static int GREENLET_NOINLINE(slp_save_state)(char* stackref)
 	return 0;
 }
 
-
-/*
- * the following macros are spliced into the OS/compiler
- * specific code, in order to simplify maintenance.
- */
-
-#define SLP_SAVE_STATE(stackref, stsizediff)            \
-	stackref += STACK_MAGIC;                        \
-	if (slp_save_state((char*)stackref)) return -1; \
-	if (!PyGreenlet_ACTIVE(ts_target)) return 1;    \
-	stsizediff = ts_target->stack_start - (char*)stackref
-
-#define SLP_RESTORE_STATE()                             \
-	slp_restore_state()
-
-
-#define SLP_EVAL
-#define slp_switch GREENLET_NOINLINE(slp_switch)
-#include "slp_platformselect.h"
-#undef slp_switch
-
-#ifndef STACK_MAGIC
-#error "greenlet needs to be ported to this platform, or teached how to detect your compiler properly."
-#endif /* !STACK_MAGIC */
-
-#ifdef EXTERNAL_ASM
-/* CCP addition: Make these functions, to be called from assembler.
- * The token include file for the given platform should enable the
- * EXTERNAL_ASM define so that this is included.
- */
-
-intptr_t slp_save_state_asm(intptr_t *ref) {
-	intptr_t diff;
-	SLP_SAVE_STATE(ref, diff);
-	return diff;
-}
-
-void slp_restore_state_asm(void) {
-	SLP_RESTORE_STATE();
-}
-
-extern int slp_switch(void);
-
-#endif
-
 static int g_switchstack(void)
 {
 	/* Perform a stack switch according to some global variables

platform/switch_sparc_sun_gcc.h

@@ -2,6 +2,8 @@
  * this is the internal transfer function.
  *
  * HISTORY
+ * 16-May-15  Alexey Borzenkov <snaury@gmail.com>
+ *      Move stack spilling code inside save/restore functions
  * 30-Aug-13  Floris Bruynooghe <flub@devork.be>
         Clean the register windows again before returning.
         This does not clobber the PIC register as it leaves
@@ -33,32 +35,29 @@
 
 
 #define STACK_MAGIC 0
-#define ST_FLUSH_WINDOWS 0x03
-#define ST_CLEAN_WINDOWS 0x04
+
+
+#if defined(__sparcv9)
+#define SLP_FLUSHW __asm__ volatile ("flushw")
+#else
+#define SLP_FLUSHW __asm__ volatile ("ta 3") /* ST_FLUSH_WINDOWS */
+#endif
+
+/* On sparc we need to spill register windows inside save/restore functions */
+#define SLP_BEFORE_SAVE_STATE() SLP_FLUSHW
+#define SLP_BEFORE_RESTORE_STATE() SLP_FLUSHW
+
 
 static int
 slp_switch(void)
 {
     register int err;
     register int *stackref, stsizediff;
 
-    /* Flush SPARC register windows onto the stack, so they can be used to
-     * restore the registers after the stack has been switched out and
-     * restored.  This also ensures the current window (pointed at by
-     * the CWP register) is the only window left in the registers
-     * (CANSAVE=0, CANRESTORE=0), that means the registers of our
-     * caller are no longer there and when we return they will always
-     * be loaded from the stack by a window underflow/fill trap.
-     *
-     * On SPARC v9 and above it might be more efficient to use the
-     * FLUSHW instruction instead of TA ST_FLUSH_WINDOWS.  But that
-     * requires the correct -mcpu flag to gcc.
-     *
-     * Then put the stack pointer into stackref. */
-    __asm__ volatile (
-        "ta %1\n\t"
-        "mov %%sp, %0"
-        : "=r" (stackref) :  "i" (ST_FLUSH_WINDOWS));
+    /* Put current stack pointer into stackref.
+     * Register spilling is done in save/restore.
+     */
+    __asm__ volatile ("mov %%sp, %0" : "=r" (stackref));
 
     {
         /* Thou shalt put SLP_SAVE_STATE into a local block */
@@ -74,13 +73,6 @@ slp_switch(void)
         /* Copy new stack from it's save store on the heap */
         SLP_RESTORE_STATE();
 
-        /* No need to set the return value register, the return
-         * statement below does this just fine.  After returning a restore
-         * instruction is given and a fill-trap will load all the registers
-         * from the stack if needed.  However in a multi-threaded environment
-         * we can't guarantee the other register windows are fine to use by
-         * their threads anymore, so tell the CPU to clean them. */
-        __asm__ volatile ("ta %0" : : "i" (ST_CLEAN_WINDOWS));
         __asm__ volatile ("mov %1, %0" : "=r" (err) : "i" (0));
         return err;
     }