No even weak pull up on GPIO during reset on BeagleBoneBlack?

Not really a starterware question ... but: this is not extreme, it is caused by design of Sitara. Independent from what resistor you use, a pull-up is always a pull-up and therefore a logical HIGH for the processor. And when you are doing this on one of the boot-selection pins, you tell the CPU to use a different boot-device (which will not exist for most cases).

So your pull-up is not only some kind of resistor, it is an logical state which is used as information for the CPU.

I knew some GPIO pins have these functions, so I tried to avoid them.

For example, if we connect a default RS232 (TTL) signal (logic 1) to BBB's serial port, it will not boot, but I don't find any info about such connection will make BBB enter special boot mode. This practice has NO effect on older hardware of BBB, specifically, rev A5 works fine, but not C 

Are you referring to the console port? I'm not sure what's going on with that thing but I've noticed it's prone to cause u-boot to see a spurious char and as a result halt on the "Press any key to interrupt boot" part. The console port doesn't connect directly to the processor but goes via a buffer IC powered by vdd_3v3b, whose power sequencing has changed slightly in rev A6, although I'm not sure why that would have any relevant effect.

On the other hand I've also noticed very strange things sometimes. We had an LCD display panel connected via an FPD-Link ("LVDS") framer to the beaglebone, and if the backlight of the panel wasn't turned off, the BBB was incapable of powering off: doing so resulted in immediate power on again. Really bizarre.