From: Alexey Brodkin <Alexey.Brodkin@synopsys.com>
Date: Mon, 14 Dec 2015 14:15:13 +0000 (+0300)
Subject: arc: cache - utilize IO coherency (AKA IOC) engine
X-Git-Url: http://git.ipfire.org/?p=people%2Fms%2Fu-boot.git;a=commitdiff_plain;h=db6ce2312dcae87619136457d1f9df56789f630a;ds=sidebyside

arc: cache - utilize IO coherency (AKA IOC) engine

With release of ARC HS38 v2.1 new IO coherency engine could be built-in
ARC core. This hardware module ensures coherency between DMA-ed data
from peripherals and L2 cache.

With L2 and IOC enabled there's no overhead for L2 cache manual
maintenance which results in significantly improved IO bandwidth.

Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
---

diff --git a/arch/arc/include/asm/arcregs.h b/arch/arc/include/asm/arcregs.h
index 667f218bd8..cf999b0280 100644
--- a/arch/arc/include/asm/arcregs.h
+++ b/arch/arc/include/asm/arcregs.h
@@ -53,6 +53,13 @@
 #define ARC_AUX_SLC_INVALIDATE	0x905
 #define ARC_AUX_SLC_IVDL	0x910
 #define ARC_AUX_SLC_FLDL	0x912
+#define ARC_BCR_CLUSTER		0xcf
+
+/* IO coherency related auxiliary registers */
+#define ARC_AUX_IO_COH_ENABLE	0x500
+#define ARC_AUX_IO_COH_PARTIAL	0x501
+#define ARC_AUX_IO_COH_AP0_BASE	0x508
+#define ARC_AUX_IO_COH_AP0_SIZE	0x509
 
 #ifndef __ASSEMBLY__
 /* Accessors for auxiliary registers */
diff --git a/arch/arc/lib/cache.c b/arch/arc/lib/cache.c
index 8089f9df1e..56988dda18 100644
--- a/arch/arc/lib/cache.c
+++ b/arch/arc/lib/cache.c
@@ -39,6 +39,7 @@ int icache_exists __section(".data");
 #ifdef CONFIG_ISA_ARCV2
 int slc_line_sz __section(".data");
 int slc_exists __section(".data");
+int ioc_exists __section(".data");
 
 static unsigned int __before_slc_op(const int op)
 {
@@ -147,6 +148,21 @@ static void read_decode_cache_bcr_arcv2(void)
 		slc_exists = 1;
 		slc_line_sz = (slc_cfg.fields.lsz == 0) ? 128 : 64;
 	}
+
+	union {
+		struct bcr_clust_cfg {
+#ifdef CONFIG_CPU_BIG_ENDIAN
+			unsigned int pad:7, c:1, num_entries:8, num_cores:8, ver:8;
+#else
+			unsigned int ver:8, num_cores:8, num_entries:8, c:1, pad:7;
+#endif
+		} fields;
+		unsigned int word;
+	} cbcr;
+
+	cbcr.word = read_aux_reg(ARC_BCR_CLUSTER);
+	if (cbcr.fields.c)
+		ioc_exists = 1;
 }
 #endif
 
@@ -191,6 +207,17 @@ void cache_init(void)
 
 #ifdef CONFIG_ISA_ARCV2
 	read_decode_cache_bcr_arcv2();
+
+	if (ioc_exists) {
+		/* IO coherency base - 0x8z */
+		write_aux_reg(ARC_AUX_IO_COH_AP0_BASE, 0x80000);
+		/* IO coherency aperture size - 512Mb: 0x8z-0xAz */
+		write_aux_reg(ARC_AUX_IO_COH_AP0_SIZE, 0x11);
+		/* Enable partial writes */
+		write_aux_reg(ARC_AUX_IO_COH_PARTIAL, 1);
+		/* Enable IO coherency */
+		write_aux_reg(ARC_AUX_IO_COH_ENABLE, 1);
+	}
 #endif
 }
 
@@ -359,18 +386,26 @@ static inline void __dc_line_op(unsigned long paddr, unsigned long sz,
 
 void invalidate_dcache_range(unsigned long start, unsigned long end)
 {
-	__dc_line_op(start, end - start, OP_INV);
 #ifdef CONFIG_ISA_ARCV2
-	if (slc_exists)
+	if (!ioc_exists)
+#endif
+		__dc_line_op(start, end - start, OP_INV);
+
+#ifdef CONFIG_ISA_ARCV2
+	if (slc_exists && !ioc_exists)
 		__slc_line_op(start, end - start, OP_INV);
 #endif
 }
 
 void flush_dcache_range(unsigned long start, unsigned long end)
 {
-	__dc_line_op(start, end - start, OP_FLUSH);
 #ifdef CONFIG_ISA_ARCV2
-	if (slc_exists)
+	if (!ioc_exists)
+#endif
+		__dc_line_op(start, end - start, OP_FLUSH);
+
+#ifdef CONFIG_ISA_ARCV2
+	if (slc_exists && !ioc_exists)
 		__slc_line_op(start, end - start, OP_FLUSH);
 #endif
 }
@@ -382,18 +417,26 @@ void flush_cache(unsigned long start, unsigned long size)
 
 void invalidate_dcache_all(void)
 {
-	__dc_entire_op(OP_INV);
 #ifdef CONFIG_ISA_ARCV2
-	if (slc_exists)
+	if (!ioc_exists)
+#endif
+		__dc_entire_op(OP_INV);
+
+#ifdef CONFIG_ISA_ARCV2
+	if (slc_exists && !ioc_exists)
 		__slc_entire_op(OP_INV);
 #endif
 }
 
 void flush_dcache_all(void)
 {
-	__dc_entire_op(OP_FLUSH);
 #ifdef CONFIG_ISA_ARCV2
-	if (slc_exists)
+	if (!ioc_exists)
+#endif
+		__dc_entire_op(OP_FLUSH);
+
+#ifdef CONFIG_ISA_ARCV2
+	if (slc_exists && !ioc_exists)
 		__slc_entire_op(OP_FLUSH);
 #endif
 }