Skip to content
Snippets Groups Projects
Commit fdce83c1 authored by Mike Frysinger's avatar Mike Frysinger
Browse files

Blackfin: rewrite cache handling functions


Take the cache flush functions from the kernel as they use hardware loops in
order to get optimal performance.

Signed-off-by: default avatarMike Frysinger <vapier@gentoo.org>
parent 84c5f0dc
No related branches found
No related tags found
No related merge requests found
/* cache.S - low level cache handling routines /*
* Copyright (C) 2003-2007 Analog Devices Inc. * Blackfin cache control code
*
* Copyright 2003-2008 Analog Devices Inc.
*
* Enter bugs at http://blackfin.uclinux.org/
*
* Licensed under the GPL-2 or later. * Licensed under the GPL-2 or later.
*/ */
...@@ -8,54 +13,75 @@ ...@@ -8,54 +13,75 @@
#include <asm/blackfin.h> #include <asm/blackfin.h>
.text .text
.align 2 /* Since all L1 caches work the same way, we use the same method for flushing
ENTRY(_blackfin_icache_flush_range) * them. Only the actual flush instruction differs. We write this in asm as
R2 = -32; * GCC can be hard to coax into writing nice hardware loops.
R2 = R0 & R2; *
P0 = R2; * Also, we assume the following register setup:
P1 = R1; * R0 = start address
CSYNC; * R1 = end address
*/
.macro do_flush flushins:req optflushins optnopins label
R2 = -L1_CACHE_BYTES;
/* start = (start & -L1_CACHE_BYTES) */
R0 = R0 & R2;
/* end = ((end - 1) & -L1_CACHE_BYTES) + L1_CACHE_BYTES; */
R1 += -1;
R1 = R1 & R2;
R1 += L1_CACHE_BYTES;
/* count = (end - start) >> L1_CACHE_SHIFT */
R2 = R1 - R0;
R2 >>= L1_CACHE_SHIFT;
P1 = R2;
.ifnb \label
\label :
.endif
P0 = R0;
LSETUP (1f, 2f) LC1 = P1;
1: 1:
IFLUSH[P0++]; .ifnb \optflushins
CC = P0 < P1(iu); \optflushins [P0];
IF CC JUMP 1b(bp); .endif
IFLUSH[P0]; #if ANOMALY_05000443
SSYNC; .ifb \optnopins
2:
.endif
\flushins [P0++];
.ifnb \optnopins
2: \optnopins;
.endif
#else
2: \flushins [P0++];
#endif
RTS; RTS;
.endm
/* Invalidate all instruction cache lines assocoiated with this memory area */
ENTRY(_blackfin_icache_flush_range)
do_flush IFLUSH, , nop
ENDPROC(_blackfin_icache_flush_range) ENDPROC(_blackfin_icache_flush_range)
ENTRY(_blackfin_dcache_flush_range) /* Flush all cache lines assocoiated with this area of memory. */
R2 = -32; ENTRY(_blackfin_icache_dcache_flush_range)
R2 = R0 & R2; do_flush FLUSH, IFLUSH
P0 = R2; ENDPROC(_blackfin_icache_dcache_flush_range)
P1 = R1;
CSYNC;
1:
FLUSH[P0++];
CC = P0 < P1(iu);
IF CC JUMP 1b(bp);
FLUSH[P0];
SSYNC;
RTS;
ENDPROC(_blackfin_dcache_flush_range)
/* 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
* could bang on the DTEST MMRs ...
*/
ENTRY(_blackfin_dcache_flush_invalidate_range) ENTRY(_blackfin_dcache_flush_invalidate_range)
R2 = -32; do_flush FLUSHINV
R2 = R0 & R2;
P0 = R2;
P1 = R1;
CSYNC;
1:
FLUSHINV[P0++];
CC = P0 < P1(iu);
IF CC JUMP 1b(bp);
/*
* If the data crosses a cache line, then we'll be pointing to
* the last cache line, but won't have flushed/invalidated it yet, so do
* one more.
*/
FLUSHINV[P0];
SSYNC;
RTS;
ENDPROC(_blackfin_dcache_flush_invalidate_range) ENDPROC(_blackfin_dcache_flush_invalidate_range)
/* Flush all data cache lines assocoiated with this memory area */
ENTRY(_blackfin_dcache_flush_range)
do_flush FLUSH, , , .Ldfr
ENDPROC(_blackfin_dcache_flush_range)
...@@ -43,6 +43,9 @@ ...@@ -43,6 +43,9 @@
#define SCLK_TO_MSEC(sclk) ((MSEC_PER_SEC * ((sclk) / USEC_PER_MSEC)) / (BFIN_SCLK / USEC_PER_MSEC)) #define SCLK_TO_MSEC(sclk) ((MSEC_PER_SEC * ((sclk) / USEC_PER_MSEC)) / (BFIN_SCLK / USEC_PER_MSEC))
#define MSEC_TO_SCLK(msec) ((((BFIN_SCLK / USEC_PER_MSEC) * (msec)) / MSEC_PER_SEC) * USEC_PER_MSEC) #define MSEC_TO_SCLK(msec) ((((BFIN_SCLK / USEC_PER_MSEC) * (msec)) / MSEC_PER_SEC) * USEC_PER_MSEC)
#define L1_CACHE_SHIFT 5
#define L1_CACHE_BYTES (1 << L1_CACHE_SHIFT)
#include <asm/linkage.h> #include <asm/linkage.h>
#ifndef __ASSEMBLY__ #ifndef __ASSEMBLY__
...@@ -60,6 +63,7 @@ extern u_long get_sclk(void); ...@@ -60,6 +63,7 @@ extern u_long get_sclk(void);
extern void blackfin_icache_flush_range(const void *, const void *); extern void blackfin_icache_flush_range(const void *, const void *);
extern void blackfin_dcache_flush_range(const void *, const void *); extern void blackfin_dcache_flush_range(const void *, const void *);
extern void blackfin_icache_dcache_flush_range(const void *, const void *);
extern void blackfin_dcache_flush_invalidate_range(const void *, const void *); extern void blackfin_dcache_flush_invalidate_range(const void *, const void *);
/* Use DMA to move data from on chip to external memory. While this is /* Use DMA to move data from on chip to external memory. While this is
......
...@@ -15,15 +15,25 @@ ...@@ -15,15 +15,25 @@
void flush_cache(unsigned long addr, unsigned long size) void flush_cache(unsigned long addr, unsigned long size)
{ {
void *start_addr, *end_addr;
int istatus, dstatus;
/* no need to flush stuff in on chip memory (L1/L2/etc...) */ /* no need to flush stuff in on chip memory (L1/L2/etc...) */
if (addr >= 0xE0000000) if (addr >= 0xE0000000)
return; return;
if (icache_status()) start_addr = (void *)addr;
blackfin_icache_flush_range((void *)addr, (void *)(addr + size)); end_addr = (void *)(addr + size);
istatus = icache_status();
dstatus = dcache_status();
if (dcache_status()) if (istatus) {
blackfin_dcache_flush_range((void *)addr, (void *)(addr + size)); if (dstatus)
blackfin_icache_dcache_flush_range(start_addr, end_addr);
else
blackfin_icache_flush_range(start_addr, end_addr);
} else if (dstatus)
blackfin_dcache_flush_range(start_addr, end_addr);
} }
void icache_enable(void) void icache_enable(void)
......
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment