Blackfin: split optimization settings more

We need to place icache flush funcs into L1 inst sram to work around a
hardware anomaly.  But this currently breaks SMP support as the L1 inst
sram is per-core and cannot be called directly.  So in preparation for
making that work, split the two options.

Further, split out the SMP depend so that we can allow some for SMP.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>

arch/blackfin/Kconfig

@@ -690,13 +690,13 @@ endmenu
 
 
 menu "Blackfin Kernel Optimizations"
-	depends on !SMP
 
 comment "Memory Optimizations"
 
 config I_ENTRY_L1
 	bool "Locate interrupt entry code in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, interrupt entry code (STORE/RESTORE CONTEXT) is linked
 	  into L1 instruction memory. (less latency)
@@ -704,6 +704,7 @@ config I_ENTRY_L1
 config EXCPT_IRQ_SYSC_L1
 	bool "Locate entire ASM lowlevel exception / interrupt - Syscall and CPLB handler code in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the entire ASM lowlevel exception and interrupt entry code
 	  (STORE/RESTORE CONTEXT) is linked into L1 instruction memory.
@@ -712,6 +713,7 @@ config EXCPT_IRQ_SYSC_L1
 config DO_IRQ_L1
 	bool "Locate frequently called do_irq dispatcher function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the frequently called do_irq dispatcher function is linked
 	  into L1 instruction memory. (less latency)
@@ -719,6 +721,7 @@ config DO_IRQ_L1
 config CORE_TIMER_IRQ_L1
 	bool "Locate frequently called timer_interrupt() function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the frequently called timer_interrupt() function is linked
 	  into L1 instruction memory. (less latency)
@@ -726,6 +729,7 @@ config CORE_TIMER_IRQ_L1
 config IDLE_L1
 	bool "Locate frequently idle function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the frequently called idle function is linked
 	  into L1 instruction memory. (less latency)
@@ -733,6 +737,7 @@ config IDLE_L1
 config SCHEDULE_L1
 	bool "Locate kernel schedule function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the frequently called kernel schedule is linked
 	  into L1 instruction memory. (less latency)
@@ -740,6 +745,7 @@ config SCHEDULE_L1
 config ARITHMETIC_OPS_L1
 	bool "Locate kernel owned arithmetic functions in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, arithmetic functions are linked
 	  into L1 instruction memory. (less latency)
@@ -747,6 +753,7 @@ config ARITHMETIC_OPS_L1
 config ACCESS_OK_L1
 	bool "Locate access_ok function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the access_ok function is linked
 	  into L1 instruction memory. (less latency)
@@ -754,6 +761,7 @@ config ACCESS_OK_L1
 config MEMSET_L1
 	bool "Locate memset function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the memset function is linked
 	  into L1 instruction memory. (less latency)
@@ -761,6 +769,7 @@ config MEMSET_L1
 config MEMCPY_L1
 	bool "Locate memcpy function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the memcpy function is linked
 	  into L1 instruction memory. (less latency)
@@ -768,6 +777,7 @@ config MEMCPY_L1
 config STRCMP_L1
 	bool "locate strcmp function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the strcmp function is linked
 	  into L1 instruction memory (less latency).
@@ -775,6 +785,7 @@ config STRCMP_L1
 config STRNCMP_L1
 	bool "locate strncmp function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the strncmp function is linked
 	  into L1 instruction memory (less latency).
@@ -782,6 +793,7 @@ config STRNCMP_L1
 config STRCPY_L1
 	bool "locate strcpy function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the strcpy function is linked
 	  into L1 instruction memory (less latency).
@@ -789,6 +801,7 @@ config STRCPY_L1
 config STRNCPY_L1
 	bool "locate strncpy function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, the strncpy function is linked
 	  into L1 instruction memory (less latency).
@@ -796,6 +809,7 @@ config STRNCPY_L1
 config SYS_BFIN_SPINLOCK_L1
 	bool "Locate sys_bfin_spinlock function in L1 Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled, sys_bfin_spinlock function is linked
 	  into L1 instruction memory. (less latency)
@@ -803,6 +817,7 @@ config SYS_BFIN_SPINLOCK_L1
 config IP_CHECKSUM_L1
 	bool "Locate IP Checksum function in L1 Memory"
 	default n
+	depends on !SMP
 	help
 	  If enabled, the IP Checksum function is linked
 	  into L1 instruction memory. (less latency)
@@ -811,7 +826,7 @@ config CACHELINE_ALIGNED_L1
 	bool "Locate cacheline_aligned data to L1 Data Memory"
 	default y if !BF54x
 	default n if BF54x
-	depends on !BF531
+	depends on !SMP && !BF531
 	help
 	  If enabled, cacheline_aligned data is linked
 	  into L1 data memory. (less latency)
@@ -819,7 +834,7 @@ config CACHELINE_ALIGNED_L1
 config SYSCALL_TAB_L1
 	bool "Locate Syscall Table L1 Data Memory"
 	default n
-	depends on !BF531
+	depends on !SMP && !BF531
 	help
 	  If enabled, the Syscall LUT is linked
 	  into L1 data memory. (less latency)
@@ -827,16 +842,17 @@ config SYSCALL_TAB_L1
 config CPLB_SWITCH_TAB_L1
 	bool "Locate CPLB Switch Tables L1 Data Memory"
 	default n
-	depends on !BF531
+	depends on !SMP && !BF531
 	help
 	  If enabled, the CPLB Switch Tables are linked
 	  into L1 data memory. (less latency)
 
-config CACHE_FLUSH_L1
-	bool "Locate cache flush funcs in L1 Inst Memory"
+config ICACHE_FLUSH_L1
+	bool "Locate icache flush funcs in L1 Inst Memory"
 	default y
+	depends on !SMP
 	help
-	  If enabled, the Blackfin cache flushing functions are linked
+	  If enabled, the Blackfin icache flushing functions are linked
 	  into L1 instruction memory.
 
 	  Note that this might be required to address anomalies, but
@@ -844,9 +860,18 @@ config CACHE_FLUSH_L1
 	  If you are using a processor affected by an anomaly, the build
 	  system will double check for you and prevent it.
 
+config DCACHE_FLUSH_L1
+	bool "Locate dcache flush funcs in L1 Inst Memory"
+	default y
+	depends on !SMP
+	help
+	  If enabled, the Blackfin dcache flushing functions are linked
+	  into L1 instruction memory.
+
 config APP_STACK_L1
 	bool "Support locating application stack in L1 Scratch Memory"
 	default y
+	depends on !SMP
 	help
 	  If enabled the application stack can be located in L1
 	  scratch memory (less latency).
@@ -856,7 +881,7 @@ config APP_STACK_L1
 config EXCEPTION_L1_SCRATCH
 	bool "Locate exception stack in L1 Scratch Memory"
 	default n
-	depends on !APP_STACK_L1
+	depends on !SMP && !APP_STACK_L1
 	help
 	  Whenever an exception occurs, use the L1 Scratch memory for
 	  stack storage.  You cannot place the stacks of FLAT binaries
@@ -868,6 +893,7 @@ comment "Speed Optimizations"
 config BFIN_INS_LOWOVERHEAD
 	bool "ins[bwl] low overhead, higher interrupt latency"
 	default y
+	depends on !SMP
 	help
 	  Reads on the Blackfin are speculative. In Blackfin terms, this means
 	  they can be interrupted at any time (even after they have been issued

arch/blackfin/mach-common/arch_checks.c

@@ -61,6 +61,6 @@
 # error "Anomaly 05000220 does not allow you to use Write Back cache with L2 or External Memory"
 #endif
 
-#if ANOMALY_05000491 && !defined(CONFIG_CACHE_FLUSH_L1)
+#if ANOMALY_05000491 && !defined(CONFIG_ICACHE_FLUSH_L1)
 # error You need IFLUSH in L1 inst while Anomaly 05000491 applies
 #endif

arch/blackfin/mach-common/cache.S

@@ -11,12 +11,6 @@
 #include <asm/cache.h>
 #include <asm/page.h>
 
-#ifdef CONFIG_CACHE_FLUSH_L1
-.section .l1.text
-#else
-.text
-#endif
-
 /* 05000443 - IFLUSH cannot be last instruction in hardware loop */
 #if ANOMALY_05000443
 # define BROK_FLUSH_INST "IFLUSH"
@@ -68,11 +62,23 @@
 	RTS;
 .endm
 
+#ifdef CONFIG_ICACHE_FLUSH_L1
+.section .l1.text
+#else
+.text
+#endif
+
 /* Invalidate all instruction cache lines assocoiated with this memory area */
 ENTRY(_blackfin_icache_flush_range)
 	do_flush IFLUSH
 ENDPROC(_blackfin_icache_flush_range)
 
+#ifdef CONFIG_DCACHE_FLUSH_L1
+.section .l1.text
+#else
+.text
+#endif
+
 /* Throw away all D-cached data in specified region without any obligation to
  * write them back.  Since the Blackfin ISA does not have an "invalidate"
  * instruction, we use flush/invalidate.  Perhaps as a speed optimization we