On Fri, 31 Oct 2014 09:21:49 -0700 (PDT) Dennis Kerrisk <dennisk8...@gmail.com> wrote:
> My Board is a Rev C boared. > > kernel is 3.4.90+, > > Uboot is: > U-Boot 2014.04-10675-g44b53fd (May 23 2014 - 13:02:55) Allwinner Technology > arm-linux-gnueabihf-gcc (Debian 4.7.1-7) 4.7.1 > GNU ld (GNU Binutils for Debian) 2.22 > > > There is a bad md5 after less than a dozen tries: > bf66d2bd1a88c5720b0dc8d8ece43099 - > bf66d2bd1a88c5720b0dc8d8ece43099 - > bf66d2bd1a88c5720b0dc8d8ece43099 - > bf66d2bd1a88c5720b0dc8d8ece43099 - > bf66d2bd1a88c5720b0dc8d8ece43099 - > 55b5d5b97cc18170891e8f8993bd19bc - > bf66d2bd1a88c5720b0dc8d8ece43099 - > bf66d2bd1a88c5720b0dc8d8ece43099 - > bf66d2bd1a88c5720b0dc8d8ece43099 - > bf66d2bd1a88c5720b0dc8d8ece43099 - OK, thanks a lot for running tests and providing feedback. Your A10-Lime appears to be the third one with this particular problem confirmed. And unlike other cases, it is not Rev A, which means that all A10-Lime boards are potentially affected. Originally I thought that the problem may be only in the early pre-production batch. For the test results from my A10-Lime board, you can check the following picture: http://people.freedesktop.org/~siamashka/files/20140512/sunxi-cpufreq-plot.png There the 'cyan' line shows how much voltage is need for my A10-Lime board at each clock frequency to pass the libjpeg-turbo decoding test without exhibiting problems. The 'blue' line is the showing the cpufreq voltages used by the sunxi-3.4 kernel for Allwinner A10. If the 'cyan' line is above 'blue', then we have clear reliability problems. And if the 'cyan' line exactly matches the 'blue' line, then the problems are not reproducible, but we have no safety headroom (which is still not great). If your board is behaving in a similar way, then you either need to increase the voltage from 1.4V up to 1.45V for 1008MHz. Or reduce the maximum clock frequency to 960MHz (it is using 1.4V voltage). The voltage changes can be applied to the following table in the sunxi-3.4 kernel sources (the 'sun4i_dvfs_table'): https://github.com/linux-sunxi/linux-sunxi/blob/sunxi-v3.4.104-r0/arch/arm/plat-sunxi/cpu-freq/cpu-freq-table.c#L100 But the first simple test would be to reduce the CPU clock frequency, make sure that the libjpeg-turbo based test does not detect any problems, and then verify that your original SD card corruption issue disappears too (if not, then we need to do even more troubleshooting). Please run some more tests, try different solutions and report your findings. Then we can try to see, what can be done to improve reliability for all A10-Lime users. Thanks! -- Best regards, Siarhei Siamashka -- You received this message because you are subscribed to the Google Groups "linux-sunxi" group. To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.