[PATCH 4/6] arm64/io: Provide a WC friendly __iowriteXX_copy()

[David Laight]

From: Jason Gunthorpe
> Sent: 21 February 2024 01:17
> 
> The kernel provides driver support for using write combining IO memory
> through the __iowriteXX_copy() API which is commonly used as an optional
> optimization to generate 16/32/64 byte MemWr TLPs in a PCIe environment.
> 
...
> Implement __iowrite32/64_copy() specifically for ARM64 and use inline
> assembly to build consecutive blocks of STR instructions. Provide direct
> support for 64/32/16 large TLP generation in this manner. Optimize for
> common constant lengths so that the compiler can directly inline the store
> blocks.
...
> +/*
> + * This generates a memcpy that works on a from/to address which is aligned to
> + * bits. Count is in terms of the number of bits sized quantities to copy. It
> + * optimizes to use the STR groupings when possible so that it is WC friendly.
> + */
> +#define memcpy_toio_aligned(to, from, count, bits)                        \
> +	({                                                                \
> +		volatile u##bits __iomem *_to = to;                       \
> +		const u##bits *_from = from;                              \
> +		size_t _count = count;                                    \
> +		const u##bits *_end_from = _from + ALIGN_DOWN(_count, 8); \
> +                                                                          \
> +		for (; _from < _end_from; _from += 8, _to += 8)           \
> +			__const_memcpy_toio_aligned##bits(_to, _from, 8); \
> +		if ((_count % 8) >= 4) {    

If (_count & 4) {
                              \
> +			__const_memcpy_toio_aligned##bits(_to, _from, 4); \
> +			_from += 4;                                       \
> +			_to += 4;                                         \
> +		}                                                         \
> +		if ((_count % 4) >= 2) {                                  \
Ditto
> +			__const_memcpy_toio_aligned##bits(_to, _from, 2); \
> +			_from += 2;                                       \
> +			_to += 2;                                         \
> +		}                                                         \
> +		if (_count % 2)                                           \
and again
> +			__const_memcpy_toio_aligned##bits(_to, _from, 1); \
> +	})

But that looks bit a bit large to be inlined.
Except, perhaps, for small constant lengths.
I'd guess that even with write-combining and posted PCIe writes it
doesn't take much for it to be PCIe limited rather than cpu limited?

Is there a sane way to do the same for reads - they are far worse
than writes.

I solved the problem a few years back on a little ppc by using an on-cpu
DMA controller that could do PCIe master accesses and spinning until
the transfer completed.
But that sort of DMA controller seems uncommon.
We now initiate most of the transfers from the slave (an fpga) - after
writing a suitable/sane dma controller for that end.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)