<p><span class="h-card" translate="no"><a href="https://mastodon.social/@whitequark" class="u-url mention">@<span>whitequark</span></a></span> <span class="h-card" translate="no"><a href="https://ioc.exchange/@azonenberg" class="u-url mention">@<span>azonenberg</span></a></span> I&#39;m currently struggling with dumping an ESMT F50-series QSPI NAND flash. flashrom doesn&#39;t support it, loading the spinand.ko module on Raspberry Pi detects the chip correctly but fails to access its contents. My suspicion is the SPI controller, so I thought about synthesizing a Cadence QSPI into a Zynq-7000 and accessing it from Petalinux.<br />I have no idea how much effort it would be to write a SPI-NAND applet for Glasgow.</p>

<p><span class="h-card" translate="no"><a href="https://mastodon.social/@whitequark" class="u-url mention">@<span>whitequark</span></a></span> Also most of the cheap FPGA devkits don&#39;t have a MCU, and either 10/100 ethernet or USB2 as the only PC interface</p><p>Dumptruck has an RGMII PHY direct to the FPGA, plus a STM32H735 connected by a fast memory mapped interface (&gt;500 Mbps throughput) to the FPGA.</p><p>Right now all of the dumping algorithms are software driven, as is the TCP/IP stack. over time as I build out an accelerated processing flow and move more of the datapath to FPGA I expect performance to skyrocket. There&#39;s a lot of round tripping now.</p>

<p><span class="h-card" translate="no"><a href="https://ioc.exchange/@azonenberg" class="u-url mention">@<span>azonenberg</span></a></span> oh yeah that makes sense</p>

<p><span class="h-card" translate="no"><a href="https://mastodon.social/@whitequark" class="u-url mention">@<span>whitequark</span></a></span> It&#39;s also going to serve as a general purpose swiss-army-knife devkit for talking to random embedded stuff with a ton of IO at weird voltage levels.</p><p>The FPGA has four full 7 series IO banks broken out (200 pins, 50 each at 3.3 / 2.5 / 1.8 / 1.2) with no level shifters, just directly to a socket.</p><p>A lot of the low cost entry level devkits only have like a few PMODs, and the high end ones have expensive FMC connectors at only like one voltage level.</p><p>There was a gap for a low cost board with a ton of IO at all different voltages.</p>

<p><span class="h-card" translate="no"><a href="https://mastodon.social/@whitequark" class="u-url mention">@<span>whitequark</span></a></span> Yeah. Dumptruck started out as a result of two consecutive gigs at work running into problems dumping a) ONFI with 2.5V core and 1.2V IO and b) 47-pin parallel async NOR.</p><p>So I decided to build a dumper that was capable enough in hardware that I&#39;d never have to touch the hardware again.</p><p>Now I&#39;m building out the gateware/software stack to make it fast and capable; QSPI is the &quot;crash test dummy&quot; for that phase of the project as it&#39;s simple and I have a ton of them lying around that don&#39;t have client data on them (i.e. I can use them freely without any NDA encumbrances).</p>

<p><span class="h-card" translate="no"><a href="https://ioc.exchange/@azonenberg" class="u-url mention">@<span>azonenberg</span></a></span> ah I see. I reverse-engineered ATF15xx for that but nobody came around to make a usable P&amp;R so it&#39;s kind of in limbo</p>

<p><span class="h-card" translate="no"><a href="https://mastodon.social/@whitequark" class="u-url mention">@<span>whitequark</span></a></span> Well, yeah. I never said this was at all the final performance I was going to get.</p><p>There&#39;s multiple unnecessary pipeline stages, reads to a buffer that I then read separately, etc. It can get waaay faster.</p><p>But the big thing dumptruck was built for wasn&#39;t speed, it was exotic things that need a lot of pins and weird IO voltage levels.</p><p>Most notably parallel NOR flash that needs like 45 pins to interface with it.</p>

<p><span class="h-card" translate="no"><a href="https://ioc.exchange/@azonenberg" class="u-url mention">@<span>azonenberg</span></a></span> oh, i hit this on Glasgow long ago :D</p>

<p>Initial quad read support in DUMPRUCK dumping at 6.3 MB/s (50.4 Mbps). There&#39;s lots of room to make this faster still...</p>