AcoSail/kernel_optimize_test
8
0
Fork 0
forked from luck/tmp_suning_uos_patched
Code Pull Requests Activity

ARC: [Review] Multi-platform image #6: cpu-to-dma-addr optional

Browse Source 
All the current platforms can work with 0x8000_0000 based dma_addr_t
since the Bus Bridges typically ignore the top bit (the only excpetion
was Angel4 PCI-AHB bridge which we no longer care for).
That way we don't need plat-specific cpu-addr to bus-addr conversion.

Hooks still provided - just in case a platform has an obscure device
which say needs 0 based bus address.

That way <asm/dma_mapping.h> no longer needs to unconditinally include
<plat/dma_addr.h>

Also verfied that on Angel4 board, other peripherals (IDE-disk / EMAC)
work fine with 0x8000_0000 based dma addresses.

Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Acked-by: Arnd Bergmann <arnd@arndb.de>
This commit is contained in:
Vineet Gupta 2013-01-22 16:53:57 +05:30
parent decae9d3e8
commit fc7943d29e
3 changed files with 20 additions and 45 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
arch/arc/Kconfig
View File

@ -119,6 +119,10 @@ config CPU_BIG_ENDIAN
help
Build kernel for Big Endian Mode of ARC CPU
# If a platform can't work with 0x8000_0000 based dma_addr_t
config ARC_PLAT_NEEDS_CPU_TO_DMA
bool
config SMP
bool "Symmetric Multi-Processing (Incomplete)"
default n

16
arch/arc/include/asm/dma-mapping.h
View File

@ -13,7 +13,23 @@
#include <asm-generic/dma-coherent.h>
#include <asm/cacheflush.h>
#ifndef CONFIG_ARC_PLAT_NEEDS_CPU_TO_DMA
/*
* dma_map_* API take cpu addresses, which is kernel logical address in the
* untranslated address space (0x8000_0000) based. The dma address (bus addr)
* ideally needs to be 0x0000_0000 based hence these glue routines.
* However given that intermediate bus bridges can ignore the high bit, we can
* do with these routines being no-ops.
* If a platform/device comes up which sriclty requires 0 based bus addr
* (e.g. AHB-PCI bridge on Angel4 board), then it can provide it's own versions
*/
#define plat_dma_addr_to_kernel(dev, addr) ((unsigned long)(addr))
#define plat_kernel_addr_to_dma(dev, ptr) ((dma_addr_t)(ptr))
#else
#include <plat/dma_addr.h>
#endif
void *dma_alloc_noncoherent(struct device *dev, size_t size,
dma_addr_t *dma_handle, gfp_t gfp);

45
arch/arc/plat-arcfpga/include/plat/dma_addr.h
View File

@ -1,45 +0,0 @@
/*
* Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*
* vineetg: Feb 2009
* -For AA4 board, kernel to DMA address APIs
*/
/*
* kernel addresses are 0x800_000 based, while Bus addr are 0 based
*/
#ifndef __PLAT_DMA_ADDR_H
#define __PLAT_DMA_ADDR_H
#include <linux/device.h>
static inline unsigned long plat_dma_addr_to_kernel(struct device *dev,
dma_addr_t dma_addr)
{
return dma_addr + PAGE_OFFSET;
}
static inline dma_addr_t plat_kernel_addr_to_dma(struct device *dev, void *ptr)
{
unsigned long addr = (unsigned long)ptr;
/*
* To Catch buggy drivers which can call DMA map API with kernel vaddr
* i.e. for buffers alloc via vmalloc or ioremap which are not
* gaurnateed to be PHY contiguous and hence unfit for DMA anyways.
* On ARC kernel virtual address is 0x7000_0000 to 0x7FFF_FFFF, so
* ideally we want to check this range here, but our implementation is
* better as it checks for even worse user virtual address as well.
*/
if (likely(addr >= PAGE_OFFSET))
return addr - PAGE_OFFSET;
BUG();
return addr;
}
#endif