The biggest part of the change is a generic interface to cmpxchg.
The definition I chose for it -- requiring the user to compare the
return value with the expected `old' value -- is largely influenced
by the Sparc64 cas instruction, as that's what it requires. And
given that gcc doesn't give you access to condition codes, it's not
a loss elsewhere either.
I've tested this insofar as it now compiles on alpha.
I don't actually have any interesting 3d hardware.
r~
PS: Thoughts on moving xchg and cmpxchg to atomic.h? They seem
fairly out of place where they are...
--ZGiS0Q5IWpPtfppv
Content-Type: text/plain; charset=us-ascii
Content-Description: drm-patch
Content-Disposition: attachment; filename=z
diff -rup 2.3.30-2-dist/drivers/char/drm/drmP.h 2.3.30-2/drivers/char/drm/drmP.h
--- 2.3.30-2-dist/drivers/char/drm/drmP.h Fri Nov 26 14:17:57 1999
+++ 2.3.30-2/drivers/char/drm/drmP.h Fri Nov 26 13:31:11 1999
@@ -96,21 +96,6 @@ typedef struct wait_queue *wait_queue_he
#define init_waitqueue_head(q) *q = NULL;
#endif
-#define __drm_dummy_lock(lock) (*(__volatile__ unsigned int *)lock)
-#define _DRM_CAS(lock,old,new,__ret) \
- do { \
- int __dummy; /* Can't mark eax as clobbered */ \
- __asm__ __volatile__( \
- "lock ; cmpxchg %4,%1\n\t" \
- "setnz %0" \
- : "=d" (__ret), \
- "=m" (__drm_dummy_lock(lock)), \
- "=a" (__dummy) \
- : "2" (old), \
- "r" (new)); \
- } while (0)
-
-
/* Macros to make printk easier */
#define DRM_ERROR(fmt, arg...) \
diff -rup 2.3.30-2-dist/drivers/char/drm/lists.c 2.3.30-2/drivers/char/drm/lists.c
--- 2.3.30-2-dist/drivers/char/drm/lists.c Tue Sep 7 14:51:28 1999
+++ 2.3.30-2/drivers/char/drm/lists.c Fri Nov 26 13:31:11 1999
@@ -130,9 +130,7 @@ int drm_freelist_destroy(drm_freelist_t
int drm_freelist_put(drm_device_t *dev, drm_freelist_t *bl, drm_buf_t *buf)
{
- unsigned int old;
- unsigned int new;
- char failed;
+ drm_buf_t *old, *prev;
int count = 0;
drm_device_dma_t *dma = dev->dma;
@@ -155,15 +153,14 @@ int drm_freelist_put(drm_device_t *dev,
#endif
buf->list = DRM_LIST_FREE;
do {
- old = (unsigned long)bl->next;
- buf->next = (void *)old;
- new = (unsigned long)buf;
- _DRM_CAS(&bl->next, old, new, failed);
+ old = bl->next;
+ buf->next = old;
+ prev = cmpxchg(&bl->next, old, buf);
if (++count > DRM_LOOPING_LIMIT) {
DRM_ERROR("Looping\n");
return 1;
}
- } while (failed);
+ } while (prev != old);
atomic_inc(&bl->count);
if (atomic_read(&bl->count) > dma->buf_count) {
DRM_ERROR("%d of %d buffers free after addition of %d\n",
@@ -180,9 +177,7 @@ int drm_freelist_put(drm_device_t *dev,
static drm_buf_t *drm_freelist_try(drm_freelist_t *bl)
{
- unsigned int old;
- unsigned int new;
- char failed;
+ drm_buf_t *old, *new, *prev;
drm_buf_t *buf;
int count = 0;
@@ -190,20 +185,20 @@ static drm_buf_t *drm_freelist_try(drm_f
/* Get buffer */
do {
- old = (unsigned int)bl->next;
+ old = bl->next;
if (!old) {
return NULL;
}
- new = (unsigned long)bl->next->next;
- _DRM_CAS(&bl->next, old, new, failed);
+ new = bl->next->next;
+ prev = cmpxchg(&bl->next, old, new);
if (++count > DRM_LOOPING_LIMIT) {
DRM_ERROR("Looping\n");
return NULL;
}
- } while (failed);
+ } while (prev != old);
atomic_dec(&bl->count);
- buf = (drm_buf_t *)old;
+ buf = old;
buf->next = NULL;
buf->list = DRM_LIST_NONE;
DRM_DEBUG("%d, count = %d, wfh = %d, w%d, p%d\n",
diff -rup 2.3.30-2-dist/drivers/char/drm/lock.c 2.3.30-2/drivers/char/drm/lock.c
--- 2.3.30-2-dist/drivers/char/drm/lock.c Tue Sep 7 14:51:28 1999
+++ 2.3.30-2/drivers/char/drm/lock.c Fri Nov 26 13:31:11 1999
@@ -50,15 +50,15 @@ int drm_lock_take(__volatile__ unsigned
{
unsigned int old;
unsigned int new;
- char failed;
+ unsigned int prev;
DRM_DEBUG("%d attempts\n", context);
do {
old = *lock;
if (old & _DRM_LOCK_HELD) new = old | _DRM_LOCK_CONT;
else new = context | _DRM_LOCK_HELD;
- _DRM_CAS(lock, old, new, failed);
- } while (failed);
+ prev = cmpxchg(lock, old, new);
+ } while (prev != old);
if (_DRM_LOCKING_CONTEXT(old) == context) {
if (old & _DRM_LOCK_HELD) {
if (context != DRM_KERNEL_CONTEXT) {
@@ -84,13 +84,13 @@ int drm_lock_transfer(__volatile__ unsig
{
unsigned int old;
unsigned int new;
- char failed;
+ unsigned int prev;
do {
old = *lock;
new = context | _DRM_LOCK_HELD;
- _DRM_CAS(lock, old, new, failed);
- } while (failed);
+ prev = cmpxchg(lock, old, new);
+ } while (prev != old);
DRM_DEBUG("%d => %d\n", _DRM_LOCKING_CONTEXT(old), context);
return 1;
}
@@ -100,14 +100,14 @@ int drm_lock_free(drm_device_t *dev,
{
unsigned int old;
unsigned int new;
- char failed;
+ unsigned int prev;
DRM_DEBUG("%d\n", context);
do {
old = *lock;
new = 0;
- _DRM_CAS(lock, old, new, failed);
- } while (failed);
+ prev = cmpxchg(lock, old, new);
+ } while (prev != old);
if (_DRM_LOCK_IS_HELD(old) && _DRM_LOCKING_CONTEXT(old) != context) {
DRM_ERROR("%d freed heavyweight lock held by %d\n",
context,
diff -rup 2.3.30-2-dist/drivers/char/drm/vm.c 2.3.30-2/drivers/char/drm/vm.c
--- 2.3.30-2-dist/drivers/char/drm/vm.c Fri Nov 26 14:17:52 1999
+++ 2.3.30-2/drivers/char/drm/vm.c Fri Nov 26 14:13:44 1999
@@ -253,7 +253,9 @@ int drm_mmap(struct file *filp, struct v
}
vma->vm_flags |= VM_LOCKED | VM_SHM; /* Don't swap */
if (map->flags & _DRM_READ_ONLY) {
- pgprot_val(vma->vm_page_prot) &= ~_PAGE_RW;
+ /* Ye gads this is ugly. With more thought we could move
+ this up higher and use `protection_map' instead. */
+ vma->vm_page_prot = __pgprot(pte_val(pte_wrprotect(__pte(pgprot_val(vma->vm_page_prot)))));
}
diff -rup 2.3.30-2-dist/include/asm-alpha/system.h 2.3.30-2/include/asm-alpha/system.h
--- 2.3.30-2-dist/include/asm-alpha/system.h Fri Nov 26 14:18:15 1999
+++ 2.3.30-2/include/asm-alpha/system.h Fri Nov 26 13:31:53 1999
@@ -317,12 +317,11 @@ extern void __global_restore_flags(unsig
#define tbia() __tbi(-2, /* no second argument */)
/*
- * Give prototypes to shut up gcc.
+ * Atomic exchange.
*/
-extern __inline__ unsigned long xchg_u32(volatile int *m, unsigned long val);
-extern __inline__ unsigned long xchg_u64(volatile long *m, unsigned long val);
-extern __inline__ unsigned long xchg_u32(volatile int *m, unsigned long val)
+extern __inline__ unsigned long
+__xchg_u32(volatile int *m, unsigned long val)
{
unsigned long dummy;
@@ -341,7 +340,8 @@ extern __inline__ unsigned long xchg_u32
return val;
}
-extern __inline__ unsigned long xchg_u64(volatile long * m, unsigned long val)
+extern __inline__ unsigned long
+__xchg_u64(volatile long *m, unsigned long val)
{
unsigned long dummy;
@@ -360,32 +360,108 @@ extern __inline__ unsigned long xchg_u64
return val;
}
-/*
- * This function doesn't exist, so you'll get a linker error
- * if something tries to do an invalid xchg().
- *
- * This only works if the compiler isn't horribly bad at optimizing.
- * gcc-2.5.8 reportedly can't handle this, but as that doesn't work
- * too well on the alpha anyway..
- */
+/* This function doesn't exist, so you'll get a linker error
+ if something tries to do an invalid xchg(). */
extern void __xchg_called_with_bad_pointer(void);
static __inline__ unsigned long
-__xchg(unsigned long x, volatile void * ptr, int size)
+__xchg(volatile void *ptr, unsigned long x, int size)
{
switch (size) {
case 4:
- return xchg_u32(ptr, x);
+ return __xchg_u32(ptr, x);
case 8:
- return xchg_u64(ptr, x);
+ return __xchg_u64(ptr, x);
}
__xchg_called_with_bad_pointer();
return x;
}
-#define xchg(ptr,x) \
- ((__typeof__(*(ptr)))__xchg((unsigned long)(x),(ptr),sizeof(*(ptr))))
+#define xchg(ptr,x) \
+ ({ \
+ __typeof__(*(ptr)) _x_ = (x); \
+ (__typeof__(*(ptr))) __xchg((ptr), (unsigned long)_x_, sizeof(*(ptr))); \
+ })
+
#define tas(ptr) (xchg((ptr),1))
+
+
+/*
+ * Atomic compare and exchange. Compare OLD with MEM, if identical,
+ * store NEW in MEM. Return the initial value in MEM. Success is
+ * indicated by comparing RETURN with OLD.
+ */
+
+#define __HAVE_ARCH_CMPXCHG 1
+
+extern __inline__ unsigned long
+__cmpxchg_u32(volatile int *m, int old, int new)
+{
+ unsigned long prev, cmp;
+
+ __asm__ __volatile__(
+ "1: ldl_l %0,%2\n"
+ " cmpeq %0,%3,%1\n"
+ " beq %1,2f\n"
+ " mov %4,%1\n"
+ " stl_c %1,%2\n"
+ " beq %1,3f\n"
+ "2: mb\n"
+ ".section .text2,\"ax\"\n"
+ "3: br 1b\n"
+ ".previous"
+ : "=&r"(prev), "=&r"(cmp), "=m"(*m)
+ : "r"((long) old), "r"(new), "m"(*m));
+
+ return prev;
+}
+
+extern __inline__ unsigned long
+__cmpxchg_u64(volatile long *m, unsigned long old, unsigned long new)
+{
+ unsigned long prev, cmp;
+
+ __asm__ __volatile__(
+ "1: ldq_l %0,%2\n"
+ " cmpeq %0,%3,%1\n"
+ " beq %1,2f\n"
+ " mov %4,%1\n"
+ " stq_c %1,%2\n"
+ " beq %1,3f\n"
+ "2: mb\n"
+ ".section .text2,\"ax\"\n"
+ "3: br 1b\n"
+ ".previous"
+ : "=&r"(prev), "=&r"(cmp), "=m"(*m)
+ : "r"((long) old), "r"(new), "m"(*m));
+
+ return prev;
+}
+
+/* This function doesn't exist, so you'll get a linker error
+ if something tries to do an invalid cmpxchg(). */
+extern void __cmpxchg_called_with_bad_pointer(void);
+
+static __inline__ unsigned long
+__cmpxchg(volatile void *ptr, unsigned long old, unsigned long new, int size)
+{
+ switch (size) {
+ case 4:
+ return __cmpxchg_u32(ptr, old, new);
+ case 8:
+ return __cmpxchg_u64(ptr, old, new);
+ }
+ __cmpxchg_called_with_bad_pointer();
+ return old;
+}
+
+#define cmpxchg(ptr,o,n) \
+ ({ \
+ __typeof__(*(ptr)) _o_ = (o); \
+ __typeof__(*(ptr)) _n_ = (n); \
+ (__typeof__(*(ptr))) __cmpxchg((ptr), (unsigned long)_o_, \
+ (unsigned long)_n_, sizeof(*(ptr))); \
+ })
#endif /* __ASSEMBLY__ */
diff -rup 2.3.30-2-dist/include/asm-i386/system.h 2.3.30-2/include/asm-i386/system.h
--- 2.3.30-2-dist/include/asm-i386/system.h Fri Oct 22 12:46:19 1999
+++ 2.3.30-2/include/asm-i386/system.h Fri Nov 26 13:31:11 1999
@@ -158,6 +158,51 @@ static inline unsigned long __xchg(unsig
return x;
}
+/*
+ * Atomic compare and exchange. Compare OLD with MEM, if identical,
+ * store NEW in MEM. Return the initial value in MEM. Success is
+ * indicated by comparing RETURN with OLD.
+ */
+
+#if CPU != 386
+#define __HAVE_ARCH_CMPXCHG 1
+
+static inline unsigned long __cmpxchg(volatile void *ptr, unsigned long old,
+ unsigned long new, int size)
+{
+ unsigned long prev;
+ switch (size) {
+ case 1:
+ __asm__ __volatile__(LOCK_PREFIX "cmpxchgb %b1,%2"
+ : "=a"(prev)
+ : "q"(new), "m"(*__xg(ptr)), "0"(old)
+ : "memory");
+ return prev;
+ case 2:
+ __asm__ __volatile__(LOCK_PREFIX "cmpxchgw %w1,%2"
+ : "=a"(prev)
+ : "q"(new), "m"(*__xg(ptr)), "0"(old)
+ : "memory");
+ return prev;
+ case 4:
+ __asm__ __volatile__(LOCK_PREFIX "cmpxchgl %1,%2"
+ : "=a"(prev)
+ : "q"(new), "m"(*__xg(ptr)), "0"(old)
+ : "memory");
+ return prev;
+ }
+ return old;
+}
+
+#define cmpxchg(ptr,o,n) \
+ ((__typeof__(*(ptr)))__cmpxchg((ptr),(unsigned long)(old), \
+ (unsigned long)(new),sizeof(*(ptr))))
+
+#else
+/* Compiling for a 386 proper. Is it worth implementing via cli/sti? */
+#endif
+
+
/*
* Force strict CPU ordering.
* And yes, this is required on UP too when we're talking
--ZGiS0Q5IWpPtfppv--
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/