Skip to contentSkip to conversation

sparc64: remove byteshifting from out* helpers

PATCH Applied Harvey Harrison 2 participants 5 messages Oct 22, 2008 – Oct 29, 2008

lore
Share & tools
lore

Harvey Harrison

author
#0 lore raw

PATCH 1/1 sparc64: remove byteshifting from out* helpers

__raw_write avoids the need to byteswap twice, as we are reading from a host-endian (big) area, just deref the pointers directly, taking care of alignment.

As before, outsw must still be called with a 2-byte aligned pointer.

Signed-off-by Harvey Harrison <[email protected]>
---
 arch/sparc64/lib/PeeCeeI.c |  138 +++++++++++++++----------------------------
 1 files changed, 48 insertions(+), 90 deletions(-)
arch/sparc64/lib/PeeCeeI.c +50 -92
index 8b313f1..a4b6246 100644
@@ -20,30 +20,9 @@ void outsw(unsigned long __addr, const void *src, unsigned long count)
 {
 	void __iomem *addr = (void __iomem *) __addr;
 
-	if (count) {
-		u16 *ps = (u16 *)src;
-		u32 *pi;
-
-		if (((u64)src) & 0x2) {
-			u16 val = le16_to_cpup(ps);
-			outw(val, addr);
-			ps++;
-			count--;
-		}
-		pi = (u32 *)ps;
-		while (count >= 2) {
-			u32 w = le32_to_cpup(pi);
-
-			pi++;
-			outw(w >> 0, addr);
-			outw(w >> 16, addr);
-			count -= 2;
-		}
-		ps = (u16 *)pi;
-		if (count) {
-			u16 val = le16_to_cpup(ps);
-			outw(val, addr);
-		}
+	while (count--) {
+		__raw_writew(*(u16 *)src, addr);
+		src += sizeof(u16);
 	}
 }
 

@@ -51,76 +30,55 @@ void outsl(unsigned long __addr, const void *src, unsigned long count)
 {
 	void __iomem *addr = (void __iomem *) __addr;
 
-	if (count) {
-		if ((((u64)src) & 0x3) = 0) {
-			u32 *p = (u32 *)src;
-			while (count--) {
-				u32 val = cpu_to_le32p(p);
-				outl(val, addr);
-				p++;
-			}
-		} else {
-			u8 *pb;
-			u16 *ps = (u16 *)src;
-			u32 l = 0, l2;
-			u32 *pi;
-
-			switch (((u64)src) & 0x3) {
-			case 0x2:
-				count -= 1;
-				l = cpu_to_le16p(ps) << 16;
-				ps++;
-				pi = (u32 *)ps;
-				while (count--) {
-					l2 = cpu_to_le32p(pi);
-					pi++;
-					outl(((l >> 16) | (l2 << 16)), addr);
-					l = l2;
-				}
-				ps = (u16 *)pi;
-				l2 = cpu_to_le16p(ps);
-				outl(((l >> 16) | (l2 << 16)), addr);
-				break;
-
-			case 0x1:
-				count -= 1;
-				pb = (u8 *)src;
-				l = (*pb++ << 8);
-				ps = (u16 *)pb;
-				l2 = cpu_to_le16p(ps);
-				ps++;
-				l |= (l2 << 16);
-				pi = (u32 *)ps;
-				while (count--) {
-					l2 = cpu_to_le32p(pi);
-					pi++;
-					outl(((l >> 8) | (l2 << 24)), addr);
-					l = l2;
-				}
-				pb = (u8 *)pi;
-				outl(((l >> 8) | (*pb << 24)), addr);
-				break;
+	if (!count)
+		return;
 
-			case 0x3:
-				count -= 1;
-				pb = (u8 *)src;
-				l = (*pb++ << 24);
-				pi = (u32 *)pb;
-				while (count--) {
-					l2 = cpu_to_le32p(pi);
-					pi++;
-					outl(((l >> 24) | (l2 << 8)), addr);
-					l = l2;
-				}
-				ps = (u16 *)pi;
-				l2 = cpu_to_le16p(ps);
-				ps++;
-				pb = (u8 *)ps;
-				l2 |= (*pb << 16);
-				outl(((l >> 24) | (l2 << 8)), addr);
-				break;
-			}
+	switch (((unsigned long)src) & 0x3) {
+	case 0x0:
+		/* src is naturally aligned */
+		while (count--) {
+			__raw_writel(*(u32 *)src, addr);
+			src += sizeof(u32);
+		}
+		break;
+	case 0x2:
+		/* 2-byte alignment */
+		while (count--) {
+			u32 l = (*(u16 *)src) << 16;
+			l |= *(u16 *)(src + sizeof(u16));
+			__raw_writel(l, addr);
+			src += sizeof(u32);
+		}
+		break;
+	case 0x1:
+		/* Hold three bytes in l each time, grab a byte from l2 */
+		u32 l, l2;
+
+		l = (*(u8 *)src) << 24;
+		l |= (*(u16 *)(src + sizeof(u8))) << 8;
+		src += sizeof(u8) + sizeof(u16);
+		while (count--) {
+			l2 = *(u32 *)src;
+			l |= (l2 >> 24);
+			__raw_writel(l, addr);
+			l = l2 << 8;
+			src += sizeof(u32);
+		}
+		break;
+	case 0x3:
+		/* Hold a byte in l each time, grab 3 bytes from l2 */
+		u32 l, l2;
+
+		l = (*(u8 *)src) << 24;
+		src += sizeof(u8);
+		while (count--) {
+			l2 = *(u32 *)src;
+			l |= (l2 >> 8);
+			__raw_writel(l, addr);
+			l = l2 << 24;
+			src += sizeof(u32);
 		}
+		break;
 	}
 }
Signature · 1 line
1.6.0.2.824.geb4d2
Applied
#1 lore raw

From: Harvey Harrison <[email protected]>
Date: Wed, 22 Oct 2008 11:50:24 -0700

Quoted 7 lines

__raw_write avoids the need to byteswap twice, as we are reading from a host-endian (big) area, just deref the pointers directly, taking care of alignment.

As before, outsw must still be called with a 2-byte aligned pointer.

Signed-off-by Harvey Harrison <[email protected]>

Applied, thanks Harvey.

#2 lore raw
In reply to Harvey Harrison

From: David Miller <[email protected]>
Date: Thu, 23 Oct 2008 14:08:19 -0700 (PDT)

Quoted 7 lines

From: Harvey Harrison <[email protected]>
Date: Wed, 22 Oct 2008 11:50:24 -0700

Quoted 4 lines

__raw_write avoids the need to byteswap twice, as we are reading from a host-endian (big) area, just deref the pointers directly, taking care of alignment.

Quoted 5 lines
Quoted 3 lines

As before, outsw must still be called with a 2-byte aligned pointer.

Signed-off-by Harvey Harrison <[email protected]>

Applied, thanks Harvey.

Sorry, I had to revert this, it doesn't even build.

arch/sparc64/lib/PeeCeeI.c: In function ‘outsl’:
arch/sparc64/lib/PeeCeeI.c:57: error: expected expression before ‘u32’
arch/sparc64/lib/PeeCeeI.c:59: error: ‘l’ undeclared (first use in this function) arch/sparc64/lib/PeeCeeI.c:59: error: (Each undeclared identifier is reported only once arch/sparc64/lib/PeeCeeI.c:59: error: for each function it appears in.)
arch/sparc64/lib/PeeCeeI.c:63: error: ‘l2’ undeclared (first use in this function) arch/sparc64/lib/PeeCeeI.c:72: error: expected expression before ‘u32’
make[1]: *** [arch/sparc64/lib/PeeCeeI.o] Error 1 ÿôèº{.nÇ+‰·Ÿ®‰­†+%ŠËÿ±éݶ¥Šwÿº{.nÇ+‰·¬¥ªÜ–)îÇø§¶›¡Ü¨}©ž²Æ zÚ&j:+v‰¨þø¯ù®w¥þŠà2ŠÞ™¨è­Ú&¢)ß¡«a¶Úÿÿûàz¿äz¹Þ—ú+ƒùšŽŠÝ¢jÿŠwèþf

author
#3 lore raw
In reply to Harvey Harrison

__raw_write avoids the need to byteswap, as we are reading from a host-endian area, just deref the pointers directly, taking care of alignment.

As before, outsw must be called with a 2-byte aligned pointer.

Signed-off-by Harvey Harrison <[email protected]>

---
Sorry about the build breakage, I did a last-minute move of the variable declarations into the individual case statements where they were used. Should have retested.

This is a build-tested version.

 arch/sparc64/lib/PeeCeeI.c |  139 +++++++++++++++-----------------------------
 1 files changed, 47 insertions(+), 92 deletions(-)
arch/sparc64/lib/PeeCeeI.c +47 -92
index 8b313f1..46053e6 100644
@@ -20,107 +20,62 @@ void outsw(unsigned long __addr, const void *src, unsigned long count)
 {
 	void __iomem *addr = (void __iomem *) __addr;
 
-	if (count) {
-		u16 *ps = (u16 *)src;
-		u32 *pi;
-
-		if (((u64)src) & 0x2) {
-			u16 val = le16_to_cpup(ps);
-			outw(val, addr);
-			ps++;
-			count--;
-		}
-		pi = (u32 *)ps;
-		while (count >= 2) {
-			u32 w = le32_to_cpup(pi);
-
-			pi++;
-			outw(w >> 0, addr);
-			outw(w >> 16, addr);
-			count -= 2;
-		}
-		ps = (u16 *)pi;
-		if (count) {
-			u16 val = le16_to_cpup(ps);
-			outw(val, addr);
-		}
+	while (count--) {
+		__raw_writew(*(u16 *)src, addr);
+		src += sizeof(u16);
 	}
 }
 
 void outsl(unsigned long __addr, const void *src, unsigned long count)
 {
 	void __iomem *addr = (void __iomem *) __addr;
+	u32 l, l2;
 
-	if (count) {
-		if ((((u64)src) & 0x3) = 0) {
-			u32 *p = (u32 *)src;
-			while (count--) {
-				u32 val = cpu_to_le32p(p);
-				outl(val, addr);
-				p++;
-			}
-		} else {
-			u8 *pb;
-			u16 *ps = (u16 *)src;
-			u32 l = 0, l2;
-			u32 *pi;
-
-			switch (((u64)src) & 0x3) {
-			case 0x2:
-				count -= 1;
-				l = cpu_to_le16p(ps) << 16;
-				ps++;
-				pi = (u32 *)ps;
-				while (count--) {
-					l2 = cpu_to_le32p(pi);
-					pi++;
-					outl(((l >> 16) | (l2 << 16)), addr);
-					l = l2;
-				}
-				ps = (u16 *)pi;
-				l2 = cpu_to_le16p(ps);
-				outl(((l >> 16) | (l2 << 16)), addr);
-				break;
-
-			case 0x1:
-				count -= 1;
-				pb = (u8 *)src;
-				l = (*pb++ << 8);
-				ps = (u16 *)pb;
-				l2 = cpu_to_le16p(ps);
-				ps++;
-				l |= (l2 << 16);
-				pi = (u32 *)ps;
-				while (count--) {
-					l2 = cpu_to_le32p(pi);
-					pi++;
-					outl(((l >> 8) | (l2 << 24)), addr);
-					l = l2;
-				}
-				pb = (u8 *)pi;
-				outl(((l >> 8) | (*pb << 24)), addr);
-				break;
+	if (!count)
+		return;
 
-			case 0x3:
-				count -= 1;
-				pb = (u8 *)src;
-				l = (*pb++ << 24);
-				pi = (u32 *)pb;
-				while (count--) {
-					l2 = cpu_to_le32p(pi);
-					pi++;
-					outl(((l >> 24) | (l2 << 8)), addr);
-					l = l2;
-				}
-				ps = (u16 *)pi;
-				l2 = cpu_to_le16p(ps);
-				ps++;
-				pb = (u8 *)ps;
-				l2 |= (*pb << 16);
-				outl(((l >> 24) | (l2 << 8)), addr);
-				break;
-			}
+	switch (((unsigned long)src) & 0x3) {
+	case 0x0:
+		/* src is naturally aligned */
+		while (count--) {
+			__raw_writel(*(u32 *)src, addr);
+			src += sizeof(u32);
+		}
+		break;
+	case 0x2:
+		/* 2-byte alignment */
+		while (count--) {
+			l = (*(u16 *)src) << 16;
+			l |= *(u16 *)(src + sizeof(u16));
+			__raw_writel(l, addr);
+			src += sizeof(u32);
+		}
+		break;
+	case 0x1:
+		/* Hold three bytes in l each time, grab a byte from l2 */
+		l = (*(u8 *)src) << 24;
+		l |= (*(u16 *)(src + sizeof(u8))) << 8;
+		src += sizeof(u8) + sizeof(u16);
+		while (count--) {
+			l2 = *(u32 *)src;
+			l |= (l2 >> 24);
+			__raw_writel(l, addr);
+			l = l2 << 8;
+			src += sizeof(u32);
+		}
+		break;
+	case 0x3:
+		/* Hold a byte in l each time, grab 3 bytes from l2 */
+		l = (*(u8 *)src) << 24;
+		src += sizeof(u8);
+		while (count--) {
+			l2 = *(u32 *)src;
+			l |= (l2 >> 8);
+			__raw_writel(l, addr);
+			l = l2 << 24;
+			src += sizeof(u32);
 		}
+		break;
 	}
 }
Signature · 1 line
1.6.0.3.723.g757e