开云体育

The calibration constants are simply stored with all the other data in the
8K NVRAM chip, there's no hardware write protection against MPU
malfunction. The rest of the RAM is used for the MPUs stack, global
variables and general working memory.
So if the MPU "crashes" or otherwise misbehaves for any reason, the
calibration constants could be corrupted. You were having intermittent
kernel test failures, which presumably means there's something wrong with
the digital logic, at least occasionally. This could lead to the MPU
crashing and corrupting the calibration constants.
I've looked at what the kernel tests do on the 2465 (which is a simpler
scope), and they're quite carefully written. As a case in point, they only
use the MPU register state until the RAM tests have been completed. If the
kernel tests are failing for you, then IMHO you need to see what they're
trying to tell you and fix that first.
Looking at table 6-6 on page 6-13 in the service manual, the kernel test
failure should be telling you one of three things:
1. The RAM is bad.
2. ROM U2160 is bad.
3. ROM U2260 is bad
This should be indicated by the CH1/2/3/4 LEDs off, and one or neither of
the +/- LEDs lit - assuming the scope has no options. If any of the
CH1/2/3/4 LEDs are lit, that's pointing to an option problem.

From the service manual, it appears that there are two levels of checking
the calibration constants, and perhaps the persistent state of the scope at
large (FP settings can be saved):
1. Each constant has a parity bit.
2. There's a checksum across all constants.
It's possible that the firmware fixes the whole-region checksum on any
write to the checksummed constants, which would explain why the code
changes.

The other problems I know of that could reasonably lead to RAM corruption
are power supply problems, and the shutdown sequence. When the power is
going away the power supply signals the MPU with an NMI. From the looks of
it (I didn't investigate this deeply) the NMI service routine flushes any
pending writes to the persistent portion of the RAM and then effectively
halts the MPU waiting for the power to die. If the MPU doesn't get that
NMI, it could fail to rewrite the checksum, or just write away randomly as
the power fades.