Hi Richard ! I updated to Petalinux 2019.2 and while it still fails it now fails in a different manner. Previously only the reads failed and it was a full page missing. Now the writes fails and its no longer full pages. 64 bytes at the start of the page has seemingly random data, then correct data for 960 bytes, then another 128bytes of random data followed by correct data. So this feels more like some kind of DMA problem. I'll contact Xilinx to see what they say Thanks for your help ! /Otto On Tue, Nov 26, 2019 at 12:26 PM Richard Weinberger <richard@xxxxxx> wrote: > > Otto, > > ----- Ursprüngliche Mail ----- > > Von: "Otto Blom" <oblom586@xxxxxxxxx> > > An: "richard" <richard@xxxxxx> > > CC: "linux-mtd" <linux-mtd@xxxxxxxxxxxxxxxxxxx> > > Gesendet: Dienstag, 26. November 2019 18:11:09 > > Betreff: Re: mtd-utils io_paral test fails on NAND > > > Hi Richard ! > > > > nandpagetest ran for 12h+ w/o failing, so it looks like a driver > > problem like you suggested. > > > > All the times I've seen if fail (around 10) the failed page came back > > with all zeros.. Which is a bit strange as I would think "random" / > > duplicated page would be more likely. Maybe the page read fails > > silently and the controller returns all zeros > > If I was in your situation I'd try to figure where the 0x00 bytes come from. > Does the driver write zeros? Maybe the page is zero initialized and the controller > completely fails to transfer data and does not signal it? > > Since nandpagetest works, but UBI not it points into the direction of DMA. > On the other hand, UBI uses offsets. Maybe this is broken in the driver... > > I had more than once the case where a driver worked like charm if you asked > it for a single page or a whole block but failed if you asked for N pages > at a given offset. > > Thanks, > //richard ______________________________________________________ Linux MTD discussion mailing list http://lists.infradead.org/mailman/listinfo/linux-mtd/