openwrt-xburst/target/linux/brcm-2.4/patches/003-bcm47xx_cache_fixes.patch

diff -urN linux.old/arch/mips/kernel/entry.S linux.dev/arch/mips/kernel/entry.S
--- linux.old/arch/mips/kernel/entry.S	2005-07-05 16:46:49.000000000 +0200
+++ linux.dev/arch/mips/kernel/entry.S	2005-07-06 11:23:55.000000000 +0200
@@ -100,6 +100,10 @@
 		 * and R4400 SC and MC versions.
 		 */
 NESTED(except_vec3_generic, 0, sp)
+#ifdef CONFIG_BCM4710
+		nop
+		nop
+#endif
 #if R5432_CP0_INTERRUPT_WAR
 		mfc0	k0, CP0_INDEX
 #endif
diff -urN linux.old/arch/mips/mm/c-r4k.c linux.dev/arch/mips/mm/c-r4k.c
--- linux.old/arch/mips/mm/c-r4k.c	2005-07-05 16:46:49.000000000 +0200
+++ linux.dev/arch/mips/mm/c-r4k.c	2005-07-06 11:23:55.000000000 +0200
@@ -14,6 +14,12 @@
 #include <linux/mm.h>
 #include <linux/bitops.h>
 
+#ifdef CONFIG_BCM4710
+#include "../bcm947xx/include/typedefs.h"
+#include "../bcm947xx/include/sbconfig.h"
+#include <asm/paccess.h>
+#endif
+
 #include <asm/bcache.h>
 #include <asm/bootinfo.h>
 #include <asm/cacheops.h>
@@ -40,6 +46,8 @@
 	.bc_inv = (void *)no_sc_noop
 };
 
+int bcm4710 = 0;
+EXPORT_SYMBOL(bcm4710);
 struct bcache_ops *bcops = &no_sc_ops;
 
 #define cpu_is_r4600_v1_x()	((read_c0_prid() & 0xfffffff0) == 0x2010)
@@ -64,8 +72,10 @@
 static inline void r4k_blast_dcache_page_setup(void)
 {
 	unsigned long dc_lsize = current_cpu_data.dcache.linesz;
-
-	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;
@@ -77,7 +87,9 @@
 {
 	unsigned long dc_lsize = current_cpu_data.dcache.linesz;
 
-	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;
@@ -89,7 +101,9 @@
 {
 	unsigned long dc_lsize = current_cpu_data.dcache.linesz;
 
-	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;
@@ -266,6 +280,7 @@
 	r4k_blast_dcache();
 	r4k_blast_icache();
 
+	if (!bcm4710)
 	switch (current_cpu_data.cputype) {
 	case CPU_R4000SC:
 	case CPU_R4000MC:
@@ -304,10 +319,10 @@
 	 * Kludge alert.  For obscure reasons R4000SC and R4400SC go nuts if we
 	 * only flush the primary caches but R10000 and R12000 behave sane ...
 	 */
-	if (current_cpu_data.cputype == CPU_R4000SC ||
+	if (!bcm4710 && (current_cpu_data.cputype == CPU_R4000SC ||
 	    current_cpu_data.cputype == CPU_R4000MC ||
 	    current_cpu_data.cputype == CPU_R4400SC ||
-	    current_cpu_data.cputype == CPU_R4400MC)
+	    current_cpu_data.cputype == CPU_R4400MC))
 		r4k_blast_scache();
 }
 
@@ -383,12 +398,15 @@
 	unsigned long ic_lsize = current_cpu_data.icache.linesz;
 	unsigned long addr, aend;
 
+	addr = start & ~(dc_lsize - 1);
+	aend = (end - 1) & ~(dc_lsize - 1);
+
 	if (!cpu_has_ic_fills_f_dc) {
 		if (end - start > dcache_size)
 			r4k_blast_dcache();
 		else {
-			addr = start & ~(dc_lsize - 1);
-			aend = (end - 1) & ~(dc_lsize - 1);
+			BCM4710_PROTECTED_FILL_TLB(addr);
+			BCM4710_PROTECTED_FILL_TLB(aend);
 
 			while (1) {
 				/* Hit_Writeback_Inv_D */
@@ -403,8 +421,6 @@
 	if (end - start > icache_size)
 		r4k_blast_icache();
 	else {
-		addr = start & ~(ic_lsize - 1);
-		aend = (end - 1) & ~(ic_lsize - 1);
 		while (1) {
 			/* Hit_Invalidate_I */
 			protected_flush_icache_line(addr);
@@ -413,6 +429,9 @@
 			addr += ic_lsize;
 		}
 	}
+
+	if (bcm4710)
+		flush_cache_all();
 }
 
 /*
@@ -443,7 +462,8 @@
 	if (cpu_has_subset_pcaches) {
 		unsigned long addr = (unsigned long) page_address(page);
 
-		r4k_blast_scache_page(addr);
+		if (!bcm4710)
+			r4k_blast_scache_page(addr);
 		ClearPageDcacheDirty(page);
 
 		return;
@@ -451,6 +471,7 @@
 
 	if (!cpu_has_ic_fills_f_dc) {
 		unsigned long addr = (unsigned long) page_address(page);
+
 		r4k_blast_dcache_page(addr);
 		ClearPageDcacheDirty(page);
 	}
@@ -477,7 +498,7 @@
 	/* Catch bad driver code */
 	BUG_ON(size == 0);
 
-	if (cpu_has_subset_pcaches) {
+	if (!bcm4710 && cpu_has_subset_pcaches) {
 		unsigned long sc_lsize = current_cpu_data.scache.linesz;
 
 		if (size >= scache_size) {
@@ -509,6 +530,8 @@
 		R4600_HIT_CACHEOP_WAR_IMPL;
 		a = addr & ~(dc_lsize - 1);
 		end = (addr + size - 1) & ~(dc_lsize - 1);
+		BCM4710_FILL_TLB(a);
+		BCM4710_FILL_TLB(end);
 		while (1) {
 			flush_dcache_line(a);	/* Hit_Writeback_Inv_D */
 			if (a == end)
@@ -527,7 +550,7 @@
 	/* Catch bad driver code */
 	BUG_ON(size == 0);
 
-	if (cpu_has_subset_pcaches) {
+	if (!bcm4710 && (cpu_has_subset_pcaches)) {
 		unsigned long sc_lsize = current_cpu_data.scache.linesz;
 
 		if (size >= scache_size) {
@@ -554,6 +577,8 @@
 		R4600_HIT_CACHEOP_WAR_IMPL;
 		a = addr & ~(dc_lsize - 1);
 		end = (addr + size - 1) & ~(dc_lsize - 1);
+		BCM4710_FILL_TLB(a);
+		BCM4710_FILL_TLB(end);
 		while (1) {
 			flush_dcache_line(a);	/* Hit_Writeback_Inv_D */
 			if (a == end)
@@ -577,6 +602,8 @@
 	unsigned long dc_lsize = current_cpu_data.dcache.linesz;
 
 	R4600_HIT_CACHEOP_WAR_IMPL;
+	BCM4710_PROTECTED_FILL_TLB(addr);
+	BCM4710_PROTECTED_FILL_TLB(addr + 4);
 	protected_writeback_dcache_line(addr & ~(dc_lsize - 1));
 	protected_flush_icache_line(addr & ~(ic_lsize - 1));
 	if (MIPS4K_ICACHE_REFILL_WAR) {
@@ -986,10 +1013,12 @@
 	case CPU_R4000MC:
 	case CPU_R4400SC:
 	case CPU_R4400MC:
-		probe_scache_kseg1 = (probe_func_t) (KSEG1ADDR(&probe_scache));
-		sc_present = probe_scache_kseg1(config);
-		if (sc_present)
-			c->options |= MIPS_CPU_CACHE_CDEX_S;
+		if (!bcm4710) {
+			probe_scache_kseg1 = (probe_func_t) (KSEG1ADDR(&probe_scache));
+			sc_present = probe_scache_kseg1(config);
+			if (sc_present)
+				c->options |= MIPS_CPU_CACHE_CDEX_S;
+		}
 		break;
 
 	case CPU_R10000:
@@ -1041,6 +1070,19 @@
 static inline void coherency_setup(void)
 {
 	change_c0_config(CONF_CM_CMASK, CONF_CM_DEFAULT);
+	
+#if defined(CONFIG_BCM4310) || defined(CONFIG_BCM4704) || defined(CONFIG_BCM5365)
+	if (BCM330X(current_cpu_data.processor_id)) {
+		uint32 cm;
+
+		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
@@ -1073,6 +1115,12 @@
 	memcpy((void *)(KSEG0 + 0x100), &except_vec2_generic, 0x80);
 	memcpy((void *)(KSEG1 + 0x100), &except_vec2_generic, 0x80);
 
+	if (current_cpu_data.cputype == CPU_BCM4710 && (current_cpu_data.processor_id & PRID_REV_MASK) == 0) {
+		printk("Enabling BCM4710A0 cache workarounds.\n");
+		bcm4710 = 1;
+	} else
+		bcm4710 = 0;
+	
 	probe_pcache();
 	setup_scache();
 
diff -urN linux.old/arch/mips/mm/tlbex-mips32.S linux.dev/arch/mips/mm/tlbex-mips32.S
--- linux.old/arch/mips/mm/tlbex-mips32.S	2005-07-05 16:46:49.000000000 +0200
+++ linux.dev/arch/mips/mm/tlbex-mips32.S	2005-07-06 11:23:56.000000000 +0200
@@ -90,6 +90,9 @@
 	.set	noat
 	LEAF(except_vec0_r4000)
 	.set	mips3
+#ifdef CONFIG_BCM4704
+	nop
+#endif
 #ifdef CONFIG_SMP
 	mfc0	k1, CP0_CONTEXT
 	la	k0, pgd_current
diff -urN linux.old/include/asm-mips/r4kcache.h linux.dev/include/asm-mips/r4kcache.h
--- linux.old/include/asm-mips/r4kcache.h	2005-07-05 16:46:49.000000000 +0200
+++ linux.dev/include/asm-mips/r4kcache.h	2005-07-06 12:52:57.000000000 +0200
@@ -15,6 +15,18 @@
 #include <asm/asm.h>
 #include <asm/cacheops.h>
 
+#ifdef CONFIG_BCM4710
+#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
+
 #define cache_op(op,addr)						\
 	__asm__ __volatile__(						\
 	"	.set	noreorder				\n"	\
@@ -27,12 +39,25 @@
 
 static inline void flush_icache_line_indexed(unsigned long addr)
 {
-	cache_op(Index_Invalidate_I, addr);
+	unsigned int way;
+	unsigned long ws_inc = 1UL << current_cpu_data.dcache.waybit;
+	
+	for (way = 0; way < current_cpu_data.dcache.ways; way++) {
+		cache_op(Index_Invalidate_I, addr);
+		addr += ws_inc;
+	}
 }
 
 static inline void flush_dcache_line_indexed(unsigned long addr)
 {
-	cache_op(Index_Writeback_Inv_D, addr);
+	unsigned int way;
+	unsigned long ws_inc = 1UL << current_cpu_data.dcache.waybit;
+	
+	for (way = 0; way < current_cpu_data.dcache.ways; way++) {
+		BCM4710_DUMMY_RREG();
+		cache_op(Index_Writeback_Inv_D, addr);
+		addr += ws_inc;
+	}
 }
 
 static inline void flush_scache_line_indexed(unsigned long addr)
@@ -47,6 +72,7 @@
 
 static inline void flush_dcache_line(unsigned long addr)
 {
+	BCM4710_DUMMY_RREG();
 	cache_op(Hit_Writeback_Inv_D, addr);
 }
 
@@ -91,6 +117,7 @@
  */
 static inline void protected_writeback_dcache_line(unsigned long addr)
 {
+	BCM4710_DUMMY_RREG();
 	__asm__ __volatile__(
 		".set noreorder\n\t"
 		".set mips3\n"
@@ -138,6 +165,62 @@
 		: "r" (base),						\
 		  "i" (op));
 
+#define cache_unroll(base,op)                   \
+	__asm__ __volatile__("                  \
+		.set noreorder;                 \
+		.set mips3;                     \
+		cache %1, (%0);                 \
+		.set mips0;                     \
+		.set reorder"                   \
+		:                               \
+		: "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);
+
+	while(start < end) {
+		BCM4710_DUMMY_RREG();
+		cache_unroll(start,Index_Writeback_Inv_D);
+		start += current_cpu_data.dcache.linesz;
+	}
+}
+
+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_unroll(start,Hit_Writeback_Inv_D);
+		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_unroll(addr,Index_Writeback_Inv_D);
+		}
+	}	
+}
+
 static inline void blast_dcache16(void)
 {
 	unsigned long start = KSEG0;
@@ -148,8 +231,9 @@
 	unsigned long ws, addr;
 
 	for (ws = 0; ws < ws_end; ws += ws_inc) 
-		for (addr = start; addr < end; addr += 0x200)
+		for (addr = start; addr < end; addr += 0x200) {
 			cache16_unroll32(addr|ws,Index_Writeback_Inv_D);
+		}
 }
 
 static inline void blast_dcache16_page(unsigned long page)
@@ -173,8 +257,9 @@
 	unsigned long ws, addr;
 
 	for (ws = 0; ws < ws_end; ws += ws_inc) 
-		for (addr = start; addr < end; addr += 0x200) 
+		for (addr = start; addr < end; addr += 0x200) {
 			cache16_unroll32(addr|ws,Index_Writeback_Inv_D);
+		}
 }
 
 static inline void blast_icache16(void)
@@ -196,6 +281,7 @@
 	unsigned long start = page;
 	unsigned long end = start + PAGE_SIZE;
 
+	BCM4710_FILL_TLB(start);
 	do {
 		cache16_unroll32(start,Hit_Invalidate_I);
 		start += 0x200;
@@ -281,6 +367,7 @@
 		: "r" (base),						\
 		  "i" (op));
 
+
 static inline void blast_dcache32(void)
 {
 	unsigned long start = KSEG0;
@@ -291,8 +378,9 @@
 	unsigned long ws, addr;
 
 	for (ws = 0; ws < ws_end; ws += ws_inc) 
-		for (addr = start; addr < end; addr += 0x400) 
+		for (addr = start; addr < end; addr += 0x400) {
 			cache32_unroll32(addr|ws,Index_Writeback_Inv_D);
+		}
 }
 
 static inline void blast_dcache32_page(unsigned long page)
@@ -316,8 +404,9 @@
 	unsigned long ws, addr;
 
 	for (ws = 0; ws < ws_end; ws += ws_inc) 
-		for (addr = start; addr < end; addr += 0x400) 
+		for (addr = start; addr < end; addr += 0x400) {
 			cache32_unroll32(addr|ws,Index_Writeback_Inv_D);
+		}
 }
 
 static inline void blast_icache32(void)
@@ -339,6 +428,7 @@
 	unsigned long start = page;
 	unsigned long end = start + PAGE_SIZE;
 
+	BCM4710_FILL_TLB(start);
 	do {
 		cache32_unroll32(start,Hit_Invalidate_I);
 		start += 0x400;
@@ -443,6 +533,7 @@
 	unsigned long start = page;
 	unsigned long end = start + PAGE_SIZE;
 
+	BCM4710_FILL_TLB(start);
 	do {
 		cache64_unroll32(start,Hit_Invalidate_I);
 		start += 0x800;
diff -urN linux.old/include/asm-mips/stackframe.h linux.dev/include/asm-mips/stackframe.h
--- linux.old/include/asm-mips/stackframe.h	2005-07-05 16:46:49.000000000 +0200
+++ linux.dev/include/asm-mips/stackframe.h	2005-07-06 11:23:56.000000000 +0200
@@ -209,6 +209,20 @@
 
 #endif
 
+#if defined(CONFIG_BCM4710) || defined(CONFIG_BCM4704)
+
+#undef RESTORE_SP_AND_RET
+#define RESTORE_SP_AND_RET                               \
+		lw      sp,  PT_R29(sp);                 \
+		.set    mips3;                           \
+		nop;                                     \
+		nop;                                     \
+		eret;                                    \
+		.set    mips0
+
+#endif
+
+
 #define RESTORE_SP                                       \
 		lw	sp,  PT_R29(sp);                 \
 
diff -urN linux.old/mm/memory.c linux.dev/mm/memory.c
--- linux.old/mm/memory.c	2005-04-04 03:42:20.000000000 +0200
+++ linux.dev/mm/memory.c	2005-07-06 11:23:56.000000000 +0200
@@ -925,6 +925,7 @@
 	flush_page_to_ram(new_page);
 	flush_cache_page(vma, address);
 	establish_pte(vma, address, page_table, pte_mkwrite(pte_mkdirty(mk_pte(new_page, vma->vm_page_prot))));
+	flush_icache_page(vma, new_page);
 }
 
 /*