openwrtv3/target/linux/brcm-2.6/patches/003-bcm4710_cache_fixes.patch

diff -ur linux.old/arch/mips/kernel/genex.S linux.dev/arch/mips/kernel/genex.S
--- linux.old/arch/mips/kernel/genex.S	2006-10-16 19:09:36.000000000 +0200
+++ linux.dev/arch/mips/kernel/genex.S	2006-10-16 19:06:50.000000000 +0200
@@ -73,6 +73,10 @@
 	.set	push
 	.set	mips3
 	.set	noat
+#ifdef CONFIG_BCM947XX
+	nop
+	nop
+#endif
 	mfc0	k1, CP0_CAUSE
 	li	k0, 31<<2
 	andi	k1, k1, 0x7c
diff -ur linux.old/arch/mips/mm/c-r4k.c linux.dev/arch/mips/mm/c-r4k.c
--- linux.old/arch/mips/mm/c-r4k.c	2006-10-16 19:09:36.000000000 +0200
+++ linux.dev/arch/mips/mm/c-r4k.c	2006-10-16 19:08:46.000000000 +0200
@@ -14,6 +14,15 @@
 #include <linux/mm.h>
 #include <linux/bitops.h>
 
+#ifdef CONFIG_BCM947XX
+#include "../bcm947xx/include/typedefs.h"
+#include "../bcm947xx/include/sbconfig.h"
+#include "../bcm947xx/include/mipsinc.h"
+#undef MTC0
+#undef MFC0
+#undef cache_op
+#include <asm/paccess.h>
+#endif
 #include <asm/bcache.h>
 #include <asm/bootinfo.h>
 #include <asm/cache.h>
@@ -30,6 +39,9 @@
 #include <asm/cacheflush.h> /* for run_uncached() */
 
 
+/* For enabling BCM4710 cache workarounds */
+int bcm4710 = 0;
+
 /*
  * Special Variant of smp_call_function for use by cache functions:
  *
@@ -94,7 +106,9 @@
 {
 	unsigned long  dc_lsize = cpu_dcache_line_size();
 
-	if (dc_lsize == 16)
+	if (bcm4710)
+		r4k_blast_dcache_page = blast_dcache_page;
+	else if (dc_lsize == 16)
 		r4k_blast_dcache_page = blast_dcache16_page;
 	else if (dc_lsize == 32)
 		r4k_blast_dcache_page = r4k_blast_dcache_page_dc32;
@@ -106,7 +120,9 @@
 {
 	unsigned long dc_lsize = cpu_dcache_line_size();
 
-	if (dc_lsize == 16)
+	if (bcm4710)
+		r4k_blast_dcache_page_indexed = blast_dcache_page_indexed;
+	else if (dc_lsize == 16)
 		r4k_blast_dcache_page_indexed = blast_dcache16_page_indexed;
 	else if (dc_lsize == 32)
 		r4k_blast_dcache_page_indexed = blast_dcache32_page_indexed;
@@ -118,7 +134,9 @@
 {
 	unsigned long dc_lsize = cpu_dcache_line_size();
 
-	if (dc_lsize == 16)
+	if (bcm4710)
+		r4k_blast_dcache = blast_dcache;
+	else if (dc_lsize == 16)
 		r4k_blast_dcache = blast_dcache16;
 	else if (dc_lsize == 32)
 		r4k_blast_dcache = blast_dcache32;
@@ -527,6 +545,9 @@
 		r4k_blast_icache();
 	else
 		protected_blast_icache_range(start, end);
+
+	if (bcm4710)
+		r4k_flush_cache_all();
 }
 
 static void r4k_flush_icache_range(unsigned long start, unsigned long end)
@@ -683,6 +704,8 @@
 	unsigned long addr = (unsigned long) arg;
 
 	R4600_HIT_CACHEOP_WAR_IMPL;
+	BCM4710_PROTECTED_FILL_TLB(addr);
+	BCM4710_PROTECTED_FILL_TLB(addr + 4);
 	protected_writeback_dcache_line(addr & ~(dc_lsize - 1));
 	if (!cpu_icache_snoops_remote_store && scache_size)
 		protected_writeback_scache_line(addr & ~(sc_lsize - 1));
@@ -1189,6 +1212,16 @@
 static inline void coherency_setup(void)
 {
 	change_c0_config(CONF_CM_CMASK, CONF_CM_DEFAULT);
+#ifdef CONFIG_BCM947XX
+	if (BCM330X(current_cpu_data.processor_id)) {
+		__u32 cm = read_c0_diag();
+		/* Enable icache */
+		cm |= (1 << 31);
+		/* Enable dcache */
+		cm |= (1 << 30);
+		write_c0_diag(cm);
+	}
+#endif
 
 	/*
 	 * c0_status.cu=0 specifies that updates by the sc instruction use
@@ -1227,6 +1260,15 @@
 
 	/* Default cache error handler for R4000 and R5000 family */
 	set_uncached_handler (0x100, &except_vec2_generic, 0x80);
+	
+	/* Check if special workarounds are required */
+#ifdef CONFIG_BCM947XX
+	if (current_cpu_data.cputype == CPU_BCM4710 && (current_cpu_data.processor_id & 0xff) == 0) {
+		printk("Enabling BCM4710A0 cache workarounds.\n");
+		bcm4710 = 1;
+	} else
+#endif
+		bcm4710 = 0;
 
 	probe_pcache();
 	setup_scache();
diff -ur linux.old/arch/mips/mm/tlbex.c linux.dev/arch/mips/mm/tlbex.c
--- linux.old/arch/mips/mm/tlbex.c	2006-10-16 19:09:36.000000000 +0200
+++ linux.dev/arch/mips/mm/tlbex.c	2006-10-16 19:06:50.000000000 +0200
@@ -38,6 +38,10 @@
 
 /* #define DEBUG_TLB */
 
+#ifdef CONFIG_BCM947XX
+extern int bcm4710;
+#endif
+
 static __init int __attribute__((unused)) r45k_bvahwbug(void)
 {
 	/* XXX: We should probe for the presence of this bug, but we don't. */
@@ -1184,6 +1188,12 @@
 	memset(relocs, 0, sizeof(relocs));
 	memset(final_handler, 0, sizeof(final_handler));
 
+#ifdef CONFIG_BCM947XX
+	if (bcm4710) {
+		i_nop(&p);
+	}
+#endif
+	
 	/*
 	 * create the plain linear handler
 	 */
diff -ur linux.old/include/asm-mips/r4kcache.h linux.dev/include/asm-mips/r4kcache.h
--- linux.old/include/asm-mips/r4kcache.h	2006-10-16 19:09:36.000000000 +0200
+++ linux.dev/include/asm-mips/r4kcache.h	2006-10-16 19:09:11.000000000 +0200
@@ -17,6 +17,18 @@
 #include <asm/cpu-features.h>
 #include <asm/mipsmtregs.h>
 
+#ifdef CONFIG_BCM947XX
+#define BCM4710_DUMMY_RREG() (((sbconfig_t *)(KSEG1ADDR(SB_ENUM_BASE + SBCONFIGOFF)))->sbimstate)
+
+#define BCM4710_FILL_TLB(addr) (*(volatile unsigned long *)(addr))
+#define BCM4710_PROTECTED_FILL_TLB(addr) ({ unsigned long x; get_dbe(x, (volatile unsigned long *)(addr)); })
+#else
+#define BCM4710_DUMMY_RREG()
+
+#define BCM4710_FILL_TLB(addr)
+#define BCM4710_PROTECTED_FILL_TLB(addr)
+#endif
+
 /*
  * This macro return a properly sign-extended address suitable as base address
  * for indexed cache operations.  Two issues here:
@@ -150,6 +162,7 @@
 static inline void flush_dcache_line_indexed(unsigned long addr)
 {
 	__dflush_prologue
+	BCM4710_DUMMY_RREG();
 	cache_op(Index_Writeback_Inv_D, addr);
 	__dflush_epilogue
 }
@@ -169,6 +182,7 @@
 static inline void flush_dcache_line(unsigned long addr)
 {
 	__dflush_prologue
+	BCM4710_DUMMY_RREG();
 	cache_op(Hit_Writeback_Inv_D, addr);
 	__dflush_epilogue
 }
@@ -176,6 +190,7 @@
 static inline void invalidate_dcache_line(unsigned long addr)
 {
 	__dflush_prologue
+	BCM4710_DUMMY_RREG();
 	cache_op(Hit_Invalidate_D, addr);
 	__dflush_epilogue
 }
@@ -208,6 +223,7 @@
  */
 static inline void protected_flush_icache_line(unsigned long addr)
 {
+	BCM4710_DUMMY_RREG();
 	protected_cache_op(Hit_Invalidate_I, addr);
 }
 
@@ -219,6 +235,7 @@
  */
 static inline void protected_writeback_dcache_line(unsigned long addr)
 {
+	BCM4710_DUMMY_RREG();
 	protected_cache_op(Hit_Writeback_Inv_D, addr);
 }
 
@@ -339,8 +356,52 @@
 		: "r" (base),						\
 		  "i" (op));
 
+static inline void blast_dcache(void)
+{
+	unsigned long start = KSEG0;
+	unsigned long dcache_size = current_cpu_data.dcache.waysize * current_cpu_data.dcache.ways;
+	unsigned long end = (start + dcache_size);
+
+	do {
+		BCM4710_DUMMY_RREG();
+		cache_op(Index_Writeback_Inv_D, start);
+		start += current_cpu_data.dcache.linesz;
+	} while(start < end);
+}
+
+static inline void blast_dcache_page(unsigned long page)
+{
+	unsigned long start = page;
+	unsigned long end = start + PAGE_SIZE;
+
+	BCM4710_FILL_TLB(start);
+	do {
+		BCM4710_DUMMY_RREG();
+		cache_op(Hit_Writeback_Inv_D, start);
+		start += current_cpu_data.dcache.linesz;
+	} while(start < end);
+}
+
+static inline void blast_dcache_page_indexed(unsigned long page)
+{
+	unsigned long start = page;
+	unsigned long end = start + PAGE_SIZE;
+	unsigned long ws_inc = 1UL << current_cpu_data.dcache.waybit;
+	unsigned long ws_end = current_cpu_data.dcache.ways <<
+	                       current_cpu_data.dcache.waybit;
+	unsigned long ws, addr;
+	for (ws = 0; ws < ws_end; ws += ws_inc) {
+		start = page + ws;
+		for (addr = start; addr < end; addr += current_cpu_data.dcache.linesz) {
+			BCM4710_DUMMY_RREG();
+			cache_op(Index_Writeback_Inv_D, addr);
+		}
+	}
+}
+
+
 /* build blast_xxx, blast_xxx_page, blast_xxx_page_indexed */
-#define __BUILD_BLAST_CACHE(pfx, desc, indexop, hitop, lsize) \
+#define __BUILD_BLAST_CACHE(pfx, desc, indexop, hitop, lsize, war) \
 static inline void blast_##pfx##cache##lsize(void)			\
 {									\
 	unsigned long start = INDEX_BASE;				\
@@ -352,6 +413,7 @@
 									\
 	__##pfx##flush_prologue						\
 									\
+	war								\
 	for (ws = 0; ws < ws_end; ws += ws_inc)				\
 		for (addr = start; addr < end; addr += lsize * 32)	\
 			cache##lsize##_unroll32(addr|ws,indexop);	\
@@ -366,6 +428,7 @@
 									\
 	__##pfx##flush_prologue						\
 									\
+	war								\
 	do {								\
 		cache##lsize##_unroll32(start,hitop);			\
 		start += lsize * 32;					\
@@ -384,6 +447,8 @@
 	                       current_cpu_data.desc.waybit;		\
 	unsigned long ws, addr;						\
 									\
+	war								\
+									\
 	__##pfx##flush_prologue						\
 									\
 	for (ws = 0; ws < ws_end; ws += ws_inc)				\
@@ -393,28 +458,30 @@
 	__##pfx##flush_epilogue						\
 }
 
-__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 16)
-__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 16)
-__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 16)
-__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 32)
-__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 32)
-__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 32)
-__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 64)
-__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 64)
-__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 128)
+__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 16, )
+__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 16, BCM4710_FILL_TLB(start);)
+__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 16, )
+__BUILD_BLAST_CACHE(d, dcache, Index_Writeback_Inv_D, Hit_Writeback_Inv_D, 32, )
+__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 32, BCM4710_FILL_TLB(start);)
+__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 32, )
+__BUILD_BLAST_CACHE(i, icache, Index_Invalidate_I, Hit_Invalidate_I, 64, BCM4710_FILL_TLB(start);)
+__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 64, )
+__BUILD_BLAST_CACHE(s, scache, Index_Writeback_Inv_SD, Hit_Writeback_Inv_SD, 128, )
 
 /* build blast_xxx_range, protected_blast_xxx_range */
-#define __BUILD_BLAST_CACHE_RANGE(pfx, desc, hitop, prot) \
+#define __BUILD_BLAST_CACHE_RANGE(pfx, desc, hitop, prot, war, war2) \
 static inline void prot##blast_##pfx##cache##_range(unsigned long start, \
 						    unsigned long end)	\
 {									\
 	unsigned long lsize = cpu_##desc##_line_size();			\
 	unsigned long addr = start & ~(lsize - 1);			\
 	unsigned long aend = (end - 1) & ~(lsize - 1);			\
+	war								\
 									\
 	__##pfx##flush_prologue						\
 									\
 	while (1) {							\
+		war2						\
 		prot##cache_op(hitop, addr);				\
 		if (addr == aend)					\
 			break;						\
@@ -424,13 +491,13 @@
 	__##pfx##flush_epilogue						\
 }
 
-__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, protected_)
-__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, protected_)
-__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I, protected_)
-__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, )
-__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, )
+__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, protected_, BCM4710_PROTECTED_FILL_TLB(addr); BCM4710_PROTECTED_FILL_TLB(aend);, BCM4710_DUMMY_RREG();)
+__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, protected_,, )
+__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I, protected_,, )
+__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D,, BCM4710_FILL_TLB(addr); BCM4710_FILL_TLB(aend);, BCM4710_DUMMY_RREG();)
+__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD,,, )
 /* blast_inv_dcache_range */
-__BUILD_BLAST_CACHE_RANGE(inv_d, dcache, Hit_Invalidate_D, )
-__BUILD_BLAST_CACHE_RANGE(inv_s, scache, Hit_Invalidate_SD, )
+__BUILD_BLAST_CACHE_RANGE(inv_d, dcache, Hit_Invalidate_D,,,BCM4710_DUMMY_RREG();)
+__BUILD_BLAST_CACHE_RANGE(inv_s, scache, Hit_Invalidate_SD,,, )
 
 #endif /* _ASM_R4KCACHE_H */
diff -ur linux.old/include/asm-mips/stackframe.h linux.dev/include/asm-mips/stackframe.h
--- linux.old/include/asm-mips/stackframe.h	2006-10-16 19:09:36.000000000 +0200
+++ linux.dev/include/asm-mips/stackframe.h	2006-10-16 19:06:50.000000000 +0200
@@ -361,6 +361,10 @@
 		.macro	RESTORE_SP_AND_RET
 		LONG_L	sp, PT_R29(sp)
 		.set	mips3
+#ifdef CONFIG_BCM947XX
+		nop
+		nop
+#endif
 		eret
 		.set	mips0
 		.endm