Re: [PATCH 1/3] fpga manager: Add cyclonespi driver for Altera fpgas

From: Joshua Clayton <stillcompiling@gmail.com>
To: atull <atull@opensource.altera.com>,
	Moritz Fischer <moritz.fischer@ettus.com>
Cc: Rob Herring <robh+dt@kernel.org>,
	Mark Rutland <mark.rutland@arm.com>,
	Shawn Guo <shawnguo@kernel.org>,
	Sascha Hauer <kernel@pengutronix.de>,
	Fabio Estevam <fabio.estevam@nxp.com>,
	Russell King <linux@armlinux.org.uk>,
	Devicetree List <devicetree@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	linux-arm-kernel <linux-arm-kernel@lists.infradead.org>,
	julia@ni.com
Subject: Re: [PATCH 1/3] fpga manager: Add cyclonespi driver for Altera fpgas
Date: Fri, 7 Oct 2016 13:54:02 -0700	[thread overview]
Message-ID: <46d5a356-f2c1-3b9f-45f3-4910c97695dc@gmail.com> (raw)
In-Reply-To: <alpine.DEB.2.02.1610071248320.32627@linuxheads99>


On 10/07/2016 11:21 AM, atull wrote:
> On Fri, 7 Oct 2016, Moritz Fischer wrote:
>
>>> +static inline u32 revbit8x4(u32 n)
>>> +{
>>> +       n = ((n & 0xF0F0F0F0UL) >> 4) | ((n & 0x0F0F0F0FUL) << 4);
>>> +       n = ((n & 0xCCCCCCCCUL) >> 2) | ((n & 0x33333333UL) << 2);
>>> +       n = ((n & 0xAAAAAAAAUL) >> 1) | ((n & 0x55555555UL) << 1);
>>> +       return n;
>>> +}
The real issue is that The FPGA wants lsb first, and my SPI driver
doesn't support it.
What I really wanted to do here was to get generic support for lsb-first
SPI into the SPI subsystem.
>> During the Zynq FPGA manager reviews we decided that manipulating the bitstream
>> to be consumable by the driver is userland's job.
> Moritz, Can you remind me what that issue was there (or point me to
> that email, I can't find it)?  I don't think I had a problem with that
> in your case.  In general I think if these drivers can take the
> bitstream that comes from the manufacturer's tools and stuff it into
> the FPGA, then we are accomplishing what we want.  So I am OK with
> this here.  The intent of the driver is to load a standard rbf, same
> as the other Altera FPGA drivers.
> There is a problem here though it will be easy to fix.  This call to
> revbit8x4 should happen in cyclonespi_write(), not in
> cyclonespi_write_init(). The reason for that is that write_init() may
> just get the first chunk of the image (the header) and that write()
> will be called multiple times for the remaining chunks.  The current
> FPGA manager API won't show this problem since you have to give
> fpga_mgr_buf_load the whole image buffer at once.  But it is easy to
> imagine that some time in the future we may want to expand the FPGA
> manager API to support streaming where we don't have the whole buffer.
OK.
If generic lsb first support for SPI is too high a bar (which it may be),
I will move the bit reversing code into the write function.
> Thanks for submitting, Joshua.  Will be looking at this over the
> next several days.
>
> Alan
Thanks for the quick response!
I'll be looking forward to your review,

Joshua Clayton