On Sun, Feb 10, 2019 at 8:59 PM Chuanhong Guo <[email protected]> wrote: > > Hi! > On Mon, Feb 11, 2019 at 7:49 AM Rosen Penev <[email protected]> wrote: > > [...] > > On the PC1, I can go even higher with the SPI frequency to get better > > performance. However, the later versions of the PC1 switched flash > > chips to Spansion to avoid a restart bug. Those are documented as > > accepting 50MHz for read requests. > > FYI: I haven't check the Spansion datasheet, but when Winbond > datasheets say read instruction accepts SPI clock up to 50MHz it > refers to the normal read instruction (0x03) and for higher clock you > should switch to fast read instruction (0x0B) which can be done by > adding m25p,fast-read to dts. I guess the situation with Spansion > datasheets are the same. Wow, thanks for that fast read hint. According to a datasheet I found online, it accepts 50MHz for normal reads and 133MHz for fast reads. I then increased the speed to 80MHz and ran again:
time dd if=/dev/mtdblock3 of=/dev/null = 6.15s 44.854483 seconds for jffs2 to complete building. I then increased to 104MHz time dd if=/dev/mtdblock3 of=/dev/null = 6.23s 44.312618 seconds for jffs2 to complete building 108MHz: time dd if=/dev/mtdblock3 of=/dev/null = 6.28s 44.501574 seconds for jffs2 to complete building I went back down to 50MHz just to make sure. time dd if=/dev/mtdblock3 of=/dev/null = 8.68s 46.344117 seconds for jffs2 to complete building then to 60MHz: time dd if=/dev/mtdblock3 of=/dev/null = 7.37s 46.172334 seconds for jffs2 to complete building and finally 66MHz: time dd if=/dev/mtdblock3 of=/dev/null 45.501151 seconds for jffs2 to complete building 133MHz and 125MHz did not work. The datasheet mentions 66MHz for mismatching core and I/O voltage and 133MHz for matching. 80MHz is still faster though... _______________________________________________ openwrt-devel mailing list [email protected] https://lists.openwrt.org/mailman/listinfo/openwrt-devel
