[RFC PATCH] powerpc: Convert ____flush_dcache_icache_phys() to C

[Christophe Leroy]

Resulting code (8xx with 16 bytes per cacheline and 16k pages)

0000016c <__flush_dcache_icache_phys>:
 16c:	54 63 00 22 	rlwinm  r3,r3,0,0,17
 170:	7d 20 00 a6 	mfmsr   r9
 174:	39 40 04 00 	li      r10,1024
 178:	55 28 07 34 	rlwinm  r8,r9,0,28,26
 17c:	7c 67 1b 78 	mr      r7,r3
 180:	7d 49 03 a6 	mtctr   r10
 184:	7d 00 01 24 	mtmsr   r8
 188:	4c 00 01 2c 	isync
 18c:	7c 00 18 6c 	dcbst   0,r3
 190:	38 63 00 10 	addi    r3,r3,16
 194:	42 00 ff f8 	bdnz    18c <__flush_dcache_icache_phys+0x20>
 198:	7c 00 04 ac 	hwsync
 19c:	7d 49 03 a6 	mtctr   r10
 1a0:	7c 00 3f ac 	icbi    0,r7
 1a4:	38 e7 00 10 	addi    r7,r7,16
 1a8:	42 00 ff f8 	bdnz    1a0 <__flush_dcache_icache_phys+0x34>
 1ac:	7c 00 04 ac 	hwsync
 1b0:	7d 20 01 24 	mtmsr   r9
 1b4:	4c 00 01 2c 	isync
 1b8:	4e 80 00 20 	blr

Signed-off-by: Christophe Leroy <christophe.leroy@xxxxxx>
---
 This patch is on top of Alastair's series "powerpc: convert cache asm to C"
 Patch 3 of that series should touch __flush_dcache_icache_phys and this
 patch could come just after patch 3.

 arch/powerpc/include/asm/cacheflush.h |  8 +++++
 arch/powerpc/mm/mem.c                 | 55 ++++++++++++++++++++++++++++-------
 2 files changed, 53 insertions(+), 10 deletions(-)

diff --git a/arch/powerpc/include/asm/cacheflush.h b/arch/powerpc/include/asm/cacheflush.h
index 1826bf2cc137..bf4f2dc4eb76 100644
--- a/arch/powerpc/include/asm/cacheflush.h
+++ b/arch/powerpc/include/asm/cacheflush.h
@@ -47,6 +47,14 @@ void flush_icache_user_range(struct vm_area_struct *vma,
 				    struct page *page, unsigned long addr,
 				    int len);
 void flush_dcache_icache_page(struct page *page);
+#if defined(CONFIG_PPC32) && !defined(CONFIG_BOOKE)
+void __flush_dcache_icache_phys(unsigned long physaddr);
+#else
+static inline void __flush_dcache_icache_phys(unsigned long physaddr)
+{
+	BUG();
+}
+#endif
 
 /**
  * flush_dcache_range(): Write any modified data cache blocks out to memory and invalidate them.
diff --git a/arch/powerpc/mm/mem.c b/arch/powerpc/mm/mem.c
index 43be99de7c9a..43009f9227c4 100644
--- a/arch/powerpc/mm/mem.c
+++ b/arch/powerpc/mm/mem.c
@@ -402,6 +402,50 @@ void flush_dcache_page(struct page *page)
 }
 EXPORT_SYMBOL(flush_dcache_page);
 
+#if defined(CONFIG_PPC32) && !defined(CONFIG_BOOKE)
+void __flush_dcache_icache_phys(unsigned long physaddr)
+{
+	unsigned long bytes = l1_dcache_bytes();
+	unsigned long nb = PAGE_SIZE / bytes;
+	unsigned long addr = physaddr & PAGE_MASK;
+	unsigned long msr, msr0;
+	unsigned long loop1 = addr, loop2 = addr;
+
+	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE)) {
+		/* For a snooping icache, we still need a dummy icbi to purge all the
+		 * prefetched instructions from the ifetch buffers. We also need a sync
+		 * before the icbi to order the the actual stores to memory that might
+		 * have modified instructions with the icbi.
+		 */
+		mb(); /* sync */
+		icbi((void *)addr);
+		mb(); /* sync */
+		isync();
+		return;
+	}
+	msr0 = mfmsr();
+	msr = msr0 & ~MSR_DR;
+	asm volatile(
+	    "	mtctr %2;"
+	    "	mtmsr %3;"
+	    "	isync;"
+	    "0:	dcbst	0, %0;"
+	    "	addi	%0, %0, %4;"
+	    "	bdnz	0b;"
+	    "	sync;"
+	    "	mtctr %2;"
+	    "1:	icbi	0, %1;"
+	    "	addi	%1, %1, %4;"
+	    "	bdnz	1b;"
+	    "	sync;"
+	    "	mtmsr %5;"
+	    "	isync;"
+	    : "+r" (loop1), "+r" (loop2)
+	    : "r" (nb), "r" (msr), "i" (bytes), "r" (msr0)
+	    : "ctr", "memory");
+}
+#endif
+
 void flush_dcache_icache_page(struct page *page)
 {
 #ifdef CONFIG_HUGETLB_PAGE
@@ -419,16 +463,7 @@ void flush_dcache_icache_page(struct page *page)
 		__flush_dcache_icache(start);
 		kunmap_atomic(start);
 	} else {
-		unsigned long msr = mfmsr();
-
-		/* Clear the DR bit so that we operate on physical
-		 * rather than virtual addresses
-		 */
-		mtmsr(msr & ~(MSR_DR));
-
-		__flush_dcache_icache((void *)physaddr);
-
-		mtmsr(msr);
+		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 	}
 #endif
 }
-- 
2.13.3