Blackfin arch: SMP supporting patchset: Blackfin CPLB related code

Blackfin dual core BF561 processor can support SMP like features.
https://docs.blackfin.uclinux.org/doku.php?id=linux-kernel:smp-like

In this patch, we provide SMP extend to Blackfin CPLB related code

Signed-off-by: Graf Yang <graf.yang@analog.com>
Signed-off-by: Bryan Wu <cooloney@kernel.org>

4 files changed, 86 insertions, 34 deletions

diff --git a/arch/blackfin/include/asm/cplb-mpu.h b/arch/blackfin/include/asm/cplb-mpu.h
index 75c67b99d60..80680ad7a37 100644
--- a/arch/blackfin/include/asm/cplb-mpu.h
+++ b/arch/blackfin/include/asm/cplb-mpu.h
@@ -28,6 +28,7 @@
  */
 #ifndef __ASM_BFIN_CPLB_MPU_H
 #define __ASM_BFIN_CPLB_MPU_H
+#include <linux/threads.h>
 
 struct cplb_entry {
 	unsigned long data, addr;
@@ -39,22 +40,22 @@ struct mem_region {
 	unsigned long icplb_data;
 };
 
-extern struct cplb_entry dcplb_tbl[MAX_CPLBS];
-extern struct cplb_entry icplb_tbl[MAX_CPLBS];
+extern struct cplb_entry dcplb_tbl[NR_CPUS][MAX_CPLBS];
+extern struct cplb_entry icplb_tbl[NR_CPUS][MAX_CPLBS];
 extern int first_switched_icplb;
 extern int first_mask_dcplb;
 extern int first_switched_dcplb;
 
-extern int nr_dcplb_miss, nr_icplb_miss, nr_icplb_supv_miss, nr_dcplb_prot;
-extern int nr_cplb_flush;
+extern int nr_dcplb_miss[], nr_icplb_miss[], nr_icplb_supv_miss[];
+extern int nr_dcplb_prot[], nr_cplb_flush[];
 
 extern int page_mask_order;
 extern int page_mask_nelts;
 
-extern unsigned long *current_rwx_mask;
+extern unsigned long *current_rwx_mask[NR_CPUS];
 
-extern void flush_switched_cplbs(void);
-extern void set_mask_dcplbs(unsigned long *);
+extern void flush_switched_cplbs(unsigned int);
+extern void set_mask_dcplbs(unsigned long *, unsigned int);
 
 extern void __noreturn panic_cplb_error(int seqstat, struct pt_regs *);
 
diff --git a/arch/blackfin/include/asm/cplb.h b/arch/blackfin/include/asm/cplb.h
index 9e8b4035fce..5f7545d0620 100644
--- a/arch/blackfin/include/asm/cplb.h
+++ b/arch/blackfin/include/asm/cplb.h
@@ -30,7 +30,6 @@
 #ifndef _CPLB_H
 #define _CPLB_H
 
-#include <asm/blackfin.h>
 #include <mach/anomaly.h>
 
 #define SDRAM_IGENERIC    (CPLB_L1_CHBL | CPLB_USER_RD | CPLB_VALID | CPLB_PORTPRIO)
@@ -55,13 +54,24 @@
 #endif
 
 #define L1_DMEMORY       (CPLB_LOCK | CPLB_COMMON)
+
+#ifdef CONFIG_SMP
+#define L2_ATTR           (INITIAL_T | I_CPLB | D_CPLB)
+#define L2_IMEMORY         (CPLB_COMMON | CPLB_LOCK)
+#define L2_DMEMORY         (CPLB_COMMON | CPLB_LOCK)
+
+#else
 #ifdef CONFIG_BFIN_L2_CACHEABLE
 #define L2_IMEMORY        (SDRAM_IGENERIC)
 #define L2_DMEMORY        (SDRAM_DGENERIC)
 #else
 #define L2_IMEMORY        (CPLB_COMMON)
 #define L2_DMEMORY        (CPLB_COMMON)
-#endif
+#endif /* CONFIG_BFIN_L2_CACHEABLE */
+
+#define L2_ATTR           (INITIAL_T | SWITCH_T | I_CPLB | D_CPLB)
+#endif /* CONFIG_SMP */
+
 #define SDRAM_DNON_CHBL  (CPLB_COMMON)
 #define SDRAM_EBIU       (CPLB_COMMON)
 #define SDRAM_OOPS       (CPLB_VALID | ANOMALY_05000158_WORKAROUND | CPLB_LOCK | CPLB_DIRTY)
@@ -71,14 +81,7 @@
 #define SIZE_1M 0x00100000      /* 1M */
 #define SIZE_4M 0x00400000      /* 4M */
 
-#ifdef CONFIG_MPU
 #define MAX_CPLBS 16
-#else
-#define MAX_CPLBS (16 * 2)
-#endif
-
-#define ASYNC_MEMORY_CPLB_COVERAGE	((ASYNC_BANK0_SIZE + ASYNC_BANK1_SIZE + \
-				 ASYNC_BANK2_SIZE + ASYNC_BANK3_SIZE) / SIZE_4M)
 
 #define CPLB_ENABLE_ICACHE_P	0
 #define CPLB_ENABLE_DCACHE_P	1
diff --git a/arch/blackfin/include/asm/cplbinit.h b/arch/blackfin/include/asm/cplbinit.h
index f845b41147b..6bfc2578816 100644
--- a/arch/blackfin/include/asm/cplbinit.h
+++ b/arch/blackfin/include/asm/cplbinit.h
@@ -36,6 +36,8 @@
 #ifdef CONFIG_MPU
 
 #include <asm/cplb-mpu.h>
+extern void bfin_icache_init(struct cplb_entry *icplb_tbl);
+extern void bfin_dcache_init(struct cplb_entry *icplb_tbl);
 
 #else
 
@@ -46,8 +48,40 @@
 
 #define IN_KERNEL 1
 
-enum
-{ZERO_P, L1I_MEM, L1D_MEM, SDRAM_KERN , SDRAM_RAM_MTD, SDRAM_DMAZ, RES_MEM, ASYNC_MEM, L2_MEM};
+#define ASYNC_MEMORY_CPLB_COVERAGE  ((ASYNC_BANK0_SIZE + ASYNC_BANK1_SIZE + \
+				ASYNC_BANK2_SIZE + ASYNC_BANK3_SIZE) / SIZE_4M)
+
+#define CPLB_MEM CONFIG_MAX_MEM_SIZE
+
+/*
+* Number of required data CPLB switchtable entries
+* MEMSIZE / 4 (we mostly install 4M page size CPLBs
+* approx 16 for smaller 1MB page size CPLBs for allignment purposes
+* 1 for L1 Data Memory
+* possibly 1 for L2 Data Memory
+* 1 for CONFIG_DEBUG_HUNT_FOR_ZERO
+* 1 for ASYNC Memory
+*/
+#define MAX_SWITCH_D_CPLBS (((CPLB_MEM / 4) + 16 + 1 + 1 + 1 \
+				 + ASYNC_MEMORY_CPLB_COVERAGE) * 2)
+
+/*
+* Number of required instruction CPLB switchtable entries
+* MEMSIZE / 4 (we mostly install 4M page size CPLBs
+* approx 12 for smaller 1MB page size CPLBs for allignment purposes
+* 1 for L1 Instruction Memory
+* possibly 1 for L2 Instruction Memory
+* 1 for CONFIG_DEBUG_HUNT_FOR_ZERO
+*/
+#define MAX_SWITCH_I_CPLBS (((CPLB_MEM / 4) + 12 + 1 + 1 + 1) * 2)
+
+/* Number of CPLB table entries, used for cplb-nompu. */
+#define CPLB_TBL_ENTRIES (16 * 4)
+
+enum {
+	ZERO_P, L1I_MEM, L1D_MEM, L2_MEM, SDRAM_KERN, SDRAM_RAM_MTD, SDRAM_DMAZ,
+	RES_MEM, ASYNC_MEM, OCB_ROM
+};
 
 struct cplb_desc {
 	u32 start; /* start address */
@@ -66,8 +100,8 @@ struct cplb_tab {
 	u16 size;
 };
 
-extern u_long icplb_table[];
-extern u_long dcplb_table[];
+extern u_long icplb_tables[NR_CPUS][CPLB_TBL_ENTRIES+1];
+extern u_long dcplb_tables[NR_CPUS][CPLB_TBL_ENTRIES+1];
 
 /* Till here we are discussing about the static memory management model.
  * However, the operating envoronments commonly define more CPLB
@@ -78,15 +112,18 @@ extern u_long dcplb_table[];
  * This is how Page descriptor Table is implemented in uClinux/Blackfin.
  */
 
-extern u_long ipdt_table[];
-extern u_long dpdt_table[];
+extern u_long ipdt_tables[NR_CPUS][MAX_SWITCH_I_CPLBS+1];
+extern u_long dpdt_tables[NR_CPUS][MAX_SWITCH_D_CPLBS+1];
 #ifdef CONFIG_CPLB_INFO
-extern u_long ipdt_swapcount_table[];
-extern u_long dpdt_swapcount_table[];
+extern u_long ipdt_swapcount_tables[NR_CPUS][MAX_SWITCH_I_CPLBS];
+extern u_long dpdt_swapcount_tables[NR_CPUS][MAX_SWITCH_D_CPLBS];
 #endif
+extern void bfin_icache_init(u_long icplbs[]);
+extern void bfin_dcache_init(u_long dcplbs[]);
 
 #endif /* CONFIG_MPU */
 
-extern void generate_cplb_tables(void);
-
+#if defined(CONFIG_BFIN_DCACHE) || defined(CONFIG_BFIN_ICACHE)
+extern void generate_cplb_tables_cpu(unsigned int cpu);
+#endif
 #endif
diff --git a/arch/blackfin/include/asm/mmu_context.h b/arch/blackfin/include/asm/mmu_context.h
index 35593dda2a4..944e29faae4 100644
--- a/arch/blackfin/include/asm/mmu_context.h
+++ b/arch/blackfin/include/asm/mmu_context.h
@@ -37,6 +37,10 @@
 #include <asm/pgalloc.h>
 #include <asm/cplbinit.h>
 
+/* Note: L1 stacks are CPU-private things, so we bluntly disable this
+   feature in SMP mode, and use the per-CPU scratch SRAM bank only to
+   store the PDA instead. */
+
 extern void *current_l1_stack_save;
 extern int nr_l1stack_tasks;
 extern void *l1_stack_base;
@@ -88,12 +92,15 @@ activate_l1stack(struct mm_struct *mm, unsigned long sp_base)
 static inline void switch_mm(struct mm_struct *prev_mm, struct mm_struct *next_mm,
 			     struct task_struct *tsk)
 {
+#ifdef CONFIG_MPU
+	unsigned int cpu = smp_processor_id();
+#endif
 	if (prev_mm == next_mm)
 		return;
 #ifdef CONFIG_MPU
-	if (prev_mm->context.page_rwx_mask == current_rwx_mask) {
-		flush_switched_cplbs();
-		set_mask_dcplbs(next_mm->context.page_rwx_mask);
+	if (prev_mm->context.page_rwx_mask == current_rwx_mask[cpu]) {
+		flush_switched_cplbs(cpu);
+		set_mask_dcplbs(next_mm->context.page_rwx_mask, cpu);
 	}
 #endif
 
@@ -138,9 +145,10 @@ static inline void protect_page(struct mm_struct *mm, unsigned long addr,
 
 static inline void update_protections(struct mm_struct *mm)
 {
-	if (mm->context.page_rwx_mask == current_rwx_mask) {
-		flush_switched_cplbs();
-		set_mask_dcplbs(mm->context.page_rwx_mask);
+	unsigned int cpu = smp_processor_id();
+	if (mm->context.page_rwx_mask == current_rwx_mask[cpu]) {
+		flush_switched_cplbs(cpu);
+		set_mask_dcplbs(mm->context.page_rwx_mask, cpu);
 	}
 }
 #endif
@@ -165,6 +173,9 @@ init_new_context(struct task_struct *tsk, struct mm_struct *mm)
 static inline void destroy_context(struct mm_struct *mm)
 {
 	struct sram_list_struct *tmp;
+#ifdef CONFIG_MPU
+	unsigned int cpu = smp_processor_id();
+#endif
 
 #ifdef CONFIG_APP_STACK_L1
 	if (current_l1_stack_save == mm->context.l1_stack_save)
@@ -179,8 +190,8 @@ static inline void destroy_context(struct mm_struct *mm)
 		kfree(tmp);
 	}
 #ifdef CONFIG_MPU
-	if (current_rwx_mask == mm->context.page_rwx_mask)
-		current_rwx_mask = NULL;
+	if (current_rwx_mask[cpu] == mm->context.page_rwx_mask)
+		current_rwx_mask[cpu] = NULL;
 	free_pages((unsigned long)mm->context.page_rwx_mask, page_mask_order);
 #endif
 }