Unplug by name, not number
Dave Higton (281) 668 posts |
I see a few people have been having problems with different modules unplugged when they upgraded from 5.16 to 1.58. Therefore… Now that we have much more space available in “CMOS” in all BeagleBoard-family computers, let’s unplug modules by their names instead of their numbers. Then the correct modules should be unplugged regardless of their positions in ROM. |
WPB (1391) 352 posts |
Or what about by SWI chunk number? Should be unique per module (with the exception of things like WimpSWIVe, but such modules surely wouldn’t be in ROM), and is probably a bit easier to code for than variable-length module name strings? On the other hand, a human-readable representation would be nice. No doubt there’s a good reason why this isn’t a good idea. I’ll wait for someone to point it out! ;) |
Matthew Phillips (473) 721 posts |
The main problem with that idea is that modules do not have to provide SWIs and do not have to have a SWI chunk allocated. |
WPB (1391) 352 posts |
Good point; well made! ;) |
Rik Griffin (98) 264 posts |
Is there a good reason to want to unplug modules? Should we instead consider deprecating this feature? |
Chris Evans (457) 1614 posts |
Unplugging has its uses: |
Andrew Rawnsley (492) 1445 posts |
To be honest, I’m not convinced that the time needed to alter this is worth it over other, more pressing things (willing to be over-ridden by the likes of Jeffrey, however!). It might actually be worth protecting the core modules from being unplugged, but it does open the door for unexpected ramifications. The Iyo problems could probably have been prevented by a “clear cmos” operation during the update. Is it possible to clear cmos (ie. del-power-on) via a simple swi or two? If so, a suitable BASIC program could be placed on the ROOL site which people could use when updating the OS. This would solve the problem, and ensure a “clean” update. |
Sprow (202) 1158 posts |
It is already forbidden to unplug the first 10 or so modules so that even if you do something silly you at least get a command prompt (which you might not be able to type at if you unplugged the keyboard driver, but then delete-power-on would get that back).
If I’m following the thought process right
The conclusion I draw from this is that the name mapping is only important when going from one release to another (or back again), in which case a simple installer could prompt the user which to keep unplugged or not. In fact such an installer was proposed some time ago, but there are always bigger fish to fry in a time constrained situation. |
Dave Higton (281) 668 posts |
When we unplug modules from the command line, we do it by name. Exclusively. One corollary of this is that the code to do so already exists. The only reason that the record of unplugged modules was kept in CMOS RAM by number was that there simply was not anywhere near enough space to do it by name. That restriction no longer applies. |
Chris Evans (457) 1614 posts |
“not anywhere near enough space to do it by name. That restriction no longer applies.” yes in 2K there is quite a lot of spare space. But modern RTC chips often have no non time related CMOS ram. Our board for the Raspberry Pi does have room for another chip[use embargoed] which just happens to have 256 bytes of CMOS RAM! I’m not convinced fitting an extra 2K EEPROM is worthwhile, when softload CMOS is zero cost, available |
Dave Higton (281) 668 posts |
I’m suggesting unplug by name for all machines that use a CMOS file as functional equivalent of CMOS RAM. I too am not convinced of the value of a hardware EEPROM. I think it’s just a commercial opportunity exploited by sellers of RISC OS related stuff. |
Andrew Rawnsley (492) 1445 posts |
Dave – I can assure you there’s no commercial gain from my perspective – quite the opposite! |