I wrote: > Pulling-up WE# doesn't seem to help :-( Well, perhaps it does. But only a little.
I let the tests run for about a day and got this: http://downloads.qi-hardware.com/people/werner/m1/nor/d2/dist.png Ten samples is a bit low to be sure there's a real difference, but it does seem as if NOR corruption had become slightly less frequent. Before the rework, the frequency hovered around 1/500 while this time it stayed closer to 1/600. Here's where corruptions happened: 00000 ____________________ | 11110111 10101100 | d2/1388-corrupt.bin | 00100000 00000000 | d2/3489-corrupt.bin | 00000000 00000000 | d2/5963-corrupt.bin 00022 ______________1___1_ | 0___00__ 0____00_ | d2/4271-corrupt.bin 1/1 00082 ____________1_____1_ | _1__10__ 1_____11 | d2/1388-corrupt.bin 1/1 0008c ____________1___11__ | 00_11___ _1___000 | d2/1916-corrupt.bin 1/1 00280 __________1_1_______ | ________ _0__0___ | d2/319-corrupt.bin 1/1 00800 ________1___________ | ________ ____0___ | d2/5963-corrupt.bin 1/1 00802 ________1_________1_ | 00__00__ __0__00_ | d2/3865-corrupt.bin 1/1 00820 ________1_____1_____ | ________ _0__1___ | d2/2033-corrupt.bin 1/1 00a92 ________1_1_1__1__1_ | 01__01__ __0__00_ | d2/3489-corrupt.bin 1/1 05514 _____1_1_1_1___1_1__ | 00000000 01010100 | d2/4300-corrupt.bin 1/1 1fffc ___111111111111111__ | __0_____ 00___0__ | d2/6143-corrupt.bin 1/1 The one with all ones is funny. Never saw such a pattern before. This test ran for only about 6160 cycles, after which it was interrupted by an X failure. I'm running more tests now. My goal is to accumulate about 15000 cycles (~8500 to go). This will permit a more reliable estimate of the corruption frequency. Plus, it will shed some light onto the effectiveness of using locking to prevent NOR corruption. As the means to reduce or even eliminate NOR corruption improve, I'll also lose the ability to test the locking (without reverting all the fixes). So it's better to get this done now. - Werner _______________________________________________ http://lists.milkymist.org/listinfo.cgi/devel-milkymist.org IRC: #milkymist@Freenode
