From: linux@arm.linux.org.uk (Russell King - ARM Linux) Date: Sat, 28 Nov 2009 20:25:23 +0000 Subject: [PATCH] ARM: Introduce HAVE_IOREAD_AND_IOWRITE macro. In-Reply-To: References: <20091128182439.GB497@n2100.arm.linux.org.uk> Message-ID: <20091128202523.GD497@n2100.arm.linux.org.uk> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On Sat, Nov 28, 2009 at 09:12:31PM +0100, Krzysztof Halasa wrote: > Russell King - ARM Linux writes: > > > Missed this. > > > > If you want to override a function, it's preferred to do it the way we're > > doing it. > > > > In otherwords, if you want to override a symbol using an inline function. > > Eg: > > > > #ifndef ndelay > > static inline void ndelay(unsigned long x) > > { > > udelay(DIV_ROUND_UP(x, 1000)); > > } > > #define ndelay(x) ndelay(x) > > #endif > > I fear someone may want to remove this later thinking that > "#define a() a()" is not necessary. > > Also we'd need to test each individual macro in the generic header since > we don't know if having ioread8 also means having other functions. > > #ifndef ioread8 > define ioread8() > #endif > > #ifndef ioread16 > define ioread16() > ... > > etc. > > One self-explaining HAVE_IOREAD_AND_IOWRITE has no such problems. > > But I can convert to these #ifndef ioread{8,16,etc} of course if you > wish. Please let me know if you want only ioread8 tested, or all of > them respectively. > > What is important to me is that the IXP4xx functions are not called > __ixp4xx_something() if they do just the something() (instead of maybe > __raw_something() or __indirect_something()). The point of the #ifndef is so that you _can_ redefine them. As illustrated above, you want: static inline void writeb(u8 value, volatile void __iomem *p) { u32 addr = (u32)p; u32 n, byte_enables, data; if (addr >= VMALLOC_START) { __raw_writeb(value, addr); return; } n = addr % 4; byte_enables = (0xf & ~BIT(n)) << IXP4XX_PCI_NP_CBE_BESL; data = value << (8*n); ixp4xx_pci_write(addr, byte_enables | NP_CMD_MEMWRITE, data); } #define writeb(v, p) writeb(v, p) etc and just be done with it - absolutely no need what so ever to change arch/arm/include/asm/io.h. We've had other platforms cope with this for the last 15 years and it hasn't been a problem. I see no reason to change it.