powerpc/mpc85xx: Rewrite spin table to comply with ePAPR v1.1

Move spin table to cached memory to comply with ePAPR v1.1.
Load R3 with 64-bit value if CONFIG_SYS_PPC64 is defined.

'M' bit is set for DDR TLB to maintain cache coherence.

See details in doc/README.mpc85xx-spin-table.

Signed-off-by: York Sun <yorksun@freescale.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>

1 files changed, 104 insertions, 75 deletions

diff --git a/arch/powerpc/cpu/mpc85xx/release.S b/arch/powerpc/cpu/mpc85xx/release.S
index d08b1f3..4ba44a9 100644
--- a/arch/powerpc/cpu/mpc85xx/release.S
+++ b/arch/powerpc/cpu/mpc85xx/release.S
@@ -150,10 +150,14 @@ __secondary_start_page:
 #define toreset(x) (x - __secondary_start_page + 0xfffff000)
 
 	/* get our PIR to figure out our table entry */
-	lis	r3,toreset(__spin_table)@h
-	ori	r3,r3,toreset(__spin_table)@l
+	lis	r3,toreset(__spin_table_addr)@h
+	ori	r3,r3,toreset(__spin_table_addr)@l
+	lwz	r3,0(r3)
 
-	/* r10 has the base address for the entry */
+	/*
+	 * r10 has the base address for the entry.
+	 * we cannot access it yet before setting up a new TLB
+	 */
 	mfspr	r0,SPRN_PIR
 #if	defined(CONFIG_E6500)
 /*
@@ -180,7 +184,7 @@ __secondary_start_page:
 #else
 	mr	r4,r0
 #endif
-	slwi	r8,r4,5
+	slwi	r8,r4,6	/* spin table is padded to 64 byte */
 	add	r10,r3,r8
 
 #ifdef CONFIG_E6500
@@ -277,73 +281,111 @@ __secondary_start_page:
 	beq	2b
 #endif
 3:
-
-#define EPAPR_MAGIC		(0x45504150)
-#define ENTRY_ADDR_UPPER	0
-#define ENTRY_ADDR_LOWER	4
-#define ENTRY_R3_UPPER		8
-#define ENTRY_R3_LOWER		12
-#define ENTRY_RESV		16
-#define ENTRY_PIR		20
-#define ENTRY_R6_UPPER		24
-#define ENTRY_R6_LOWER		28
-#define ENTRY_SIZE		32
-
-	/* setup the entry */
-	li	r3,0
-	li	r8,1
-	stw	r4,ENTRY_PIR(r10)
-	stw	r3,ENTRY_ADDR_UPPER(r10)
-	stw	r8,ENTRY_ADDR_LOWER(r10)
-	stw	r3,ENTRY_R3_UPPER(r10)
-	stw	r4,ENTRY_R3_LOWER(r10)
-	stw	r3,ENTRY_R6_UPPER(r10)
-	stw	r3,ENTRY_R6_LOWER(r10)
-
-	/* load r13 with the address of the 'bootpg' in SDRAM */
-	lis	r13,toreset(__bootpg_addr)@h
-	ori	r13,r13,toreset(__bootpg_addr)@l
+	/* setup mapping for the spin table, WIMGE=0b00100 */
+	lis	r13,toreset(__spin_table_addr)@h
+	ori	r13,r13,toreset(__spin_table_addr)@l
 	lwz	r13,0(r13)
+	/* mask by 4K */
+	rlwinm	r13,r13,0,0,19
 
-	/* setup mapping for AS = 1, and jump there */
 	lis	r11,(MAS0_TLBSEL(1)|MAS0_ESEL(1))@h
 	mtspr	SPRN_MAS0,r11
 	lis	r11,(MAS1_VALID|MAS1_IPROT)@h
 	ori	r11,r11,(MAS1_TS|MAS1_TSIZE(BOOKE_PAGESZ_4K))@l
 	mtspr	SPRN_MAS1,r11
-	oris	r11,r13,(MAS2_I|MAS2_G)@h
-	ori	r11,r13,(MAS2_I|MAS2_G)@l
+	oris	r11,r13,(MAS2_M|MAS2_G)@h
+	ori	r11,r13,(MAS2_M|MAS2_G)@l
 	mtspr	SPRN_MAS2,r11
 	oris	r11,r13,(MAS3_SX|MAS3_SW|MAS3_SR)@h
 	ori	r11,r13,(MAS3_SX|MAS3_SW|MAS3_SR)@l
 	mtspr	SPRN_MAS3,r11
+	li	r11,0
+	mtspr	SPRN_MAS7,r11
 	tlbwe
 
-	bl	1f
-1:	mflr	r11
 	/*
-	 * OR in 0xfff to create a mask of the bootpg SDRAM address.  We use
-	 * this mask to fixup the cpu spin table and the address that we want
-	 * to jump to, eg change them from 0xfffffxxx to 0x7ffffxxx if the
-	 * bootpg is at 0x7ffff000 in SDRAM.
+	 * __bootpg_addr has the address of __second_half_boot_page
+	 * jump there in AS=1 space with cache enabled
 	 */
-	ori	r13,r13,0xfff
-	and	r11, r11, r13
-	and	r10, r10, r13
-
-	addi	r11,r11,(2f-1b)
+	lis	r13,toreset(__bootpg_addr)@h
+	ori	r13,r13,toreset(__bootpg_addr)@l
+	lwz	r11,0(r13)
+	mtspr	SPRN_SRR0,r11
 	mfmsr	r13
 	ori	r12,r13,MSR_IS|MSR_DS@l
-
-	mtspr	SPRN_SRR0,r11
 	mtspr	SPRN_SRR1,r12
 	rfi
 
+	/*
+	 * Allocate some space for the SDRAM address of the bootpg.
+	 * This variable has to be in the boot page so that it can
+	 * be accessed by secondary cores when they come out of reset.
+	 */
+	.align L1_CACHE_SHIFT
+	.globl __bootpg_addr
+__bootpg_addr:
+	.long	0
+
+	.global __spin_table_addr
+__spin_table_addr:
+	.long	0
+
+	/*
+	 * This variable is set by cpu_init_r() after parsing hwconfig
+	 * to enable workaround for erratum NMG_CPU_A011.
+	 */
+	.align L1_CACHE_SHIFT
+	.global enable_cpu_a011_workaround
+enable_cpu_a011_workaround:
+	.long	1
+
+	/* Fill in the empty space.  The actual reset vector is
+	 * the last word of the page */
+__secondary_start_code_end:
+	.space 4092 - (__secondary_start_code_end - __secondary_start_page)
+__secondary_reset_vector:
+	b	__secondary_start_page
+
+
+/* this is a separated page for the spin table and cacheable boot code */
+	.align L1_CACHE_SHIFT
+	.global __second_half_boot_page
+__second_half_boot_page:
+#define EPAPR_MAGIC		0x45504150
+#define ENTRY_ADDR_UPPER	0
+#define ENTRY_ADDR_LOWER	4
+#define ENTRY_R3_UPPER		8
+#define ENTRY_R3_LOWER		12
+#define ENTRY_RESV		16
+#define ENTRY_PIR		20
+#define ENTRY_SIZE		64
+	/*
+	 * setup the entry
+	 * r10 has the base address of the spin table.
+	 * spin table is defined as
+	 * struct {
+	 *	uint64_t entry_addr;
+	 *	uint64_t r3;
+	 *	uint32_t rsvd1;
+	 *	uint32_t pir;
+	 * };
+	 * we pad this struct to 64 bytes so each entry is in its own cacheline
+	 */
+	li	r3,0
+	li	r8,1
+	mfspr	r4,SPRN_PIR
+	stw	r3,ENTRY_ADDR_UPPER(r10)
+	stw	r3,ENTRY_R3_UPPER(r10)
+	stw	r4,ENTRY_R3_LOWER(r10)
+	stw	r3,ENTRY_RESV(r10)
+	stw	r4,ENTRY_PIR(r10)
+	msync
+	stw	r8,ENTRY_ADDR_LOWER(r10)
+
 	/* spin waiting for addr */
-2:
-	lwz	r4,ENTRY_ADDR_LOWER(r10)
+3:	lwz	r4,ENTRY_ADDR_LOWER(r10)
 	andi.	r11,r4,1
-	bne	2b
+	bne	3b
 	isync
 
 	/* setup IVORs to match fixed offsets */
@@ -362,8 +404,17 @@ __secondary_start_page:
 	/* mask by ~64M to setup our tlb we will jump to */
 	rlwinm	r12,r4,0,0,5
 
-	/* setup r3, r4, r5, r6, r7, r8, r9 */
+	/*
+	 * setup r3, r4, r5, r6, r7, r8, r9
+	 * r3 contains the value to put in the r3 register at secondary cpu
+	 * entry. The high 32-bits are ignored on 32-bit chip implementations.
+	 * 64-bit chip implementations however shall load all 64-bits
+	 */
+#ifdef CONFIG_SYS_PPC64
+	ld	r3,ENTRY_R3_UPPER(r10)
+#else
 	lwz	r3,ENTRY_R3_LOWER(r10)
+#endif
 	li	r4,0
 	li	r5,0
 	li	r6,0
@@ -404,32 +455,10 @@ __secondary_start_page:
 	mtspr	SPRN_SRR1,r13
 	rfi
 
-	/*
-	 * Allocate some space for the SDRAM address of the bootpg.
-	 * This variable has to be in the boot page so that it can
-	 * be accessed by secondary cores when they come out of reset.
-	 */
-	.globl __bootpg_addr
-__bootpg_addr:
-	.long	0
 
-	.align L1_CACHE_SHIFT
+	.align 6
 	.globl __spin_table
 __spin_table:
 	.space CONFIG_MAX_CPUS*ENTRY_SIZE
-
-	/*
-	 * This variable is set by cpu_init_r() after parsing hwconfig
-	 * to enable workaround for erratum NMG_CPU_A011.
-	 */
-	.align L1_CACHE_SHIFT
-	.global enable_cpu_a011_workaround
-enable_cpu_a011_workaround:
-	.long	1
-
-	/* Fill in the empty space.  The actual reset vector is
-	 * the last word of the page */
-__secondary_start_code_end:
-	.space 4092 - (__secondary_start_code_end - __secondary_start_page)
-__secondary_reset_vector:
-	b	__secondary_start_page
+__spin_table_end:
+	.space 4096 - (__spin_table_end - __spin_table)