I wrote in a commit yesterday how there was no ROI on doing this
yet somehow I kept on doing it. I blame poor impulse control.
Turns out there was a series of race conditions and bugs that
caused this thing to "mostly work" on the first go at it but
really hard to debug and get very right.
1. Litex UART has a weird undocumented API. You advance the Rx
FIFO by writing `1` to the PENDING bit, and no other way. However,
you determine if there are any more characters to receive by
reading the EMPTY bit and seeing if it is 1. The previous method
wrote `1` to the PENDING bit and then used the PENDING bit to
see if a character was available, under the theory that the
PENDING bit would be set again if there was still a character.
This is somewhat true, except that the PENDING bit is set only
some amount of time later which is about comparable to a few
CPU instructions, so if you had a fast loop path the PENDING
bit would show as cleared and you'd exit, which would cause
no further interrupts to fire because the interrupt was
effecitvely not acknowledged.
This isn't written up anywhere but I figured this out by
just reading the source code for Litex UART and BIOS:
e8be391504/litex/soc/cores/uart.py (L264-L279)ccd3ab17be/litex/soc/software/libbase/uart.c (L38-L44)
2. I had the Key Injector IP block in the `sys` domain, which
would have its clock cut when the CPU went to sleep, which would
cause input characters to disappear when the CPU went into power
saving. moving this to "always on" domain fixes that.
3. I had the UART PHY in the always on block (so you could
send or receive exactly one character), but not the
separate UART block (which contained the FIFOs).
The UART block would also go into power
savings, and this would cause FIFO entries to go missing.
4. Escape sequences get sent as a burst of characters. The Key Injector
originally did not have a FIFO of its own, so it would only
remember the *last* key of the sequence that was written, causing
some keys to be dropped occassionally. Adding a 16-deep FIFO
to the injector fixes this.
5. The code I had to read out the injector's results was designed with
the "strobe" bit (to indicate that the FIFO had characters) in the
same regsiter as the character itself. Thus the loop that 1. checked
if the fifo was empty then 2. read and forwarded the character
would actually perform two reads from the FIFO. However, this "mostly worked"
I think because the FIFO retains the last value that was written
even when it was empty, so if only one character was read at a time
everything was fine. It only didn't work when you stuffed two characters
in there, as you'd drop the first one. Reading the FIFO once and
then interpreting the code with a bitmask fixes this.
6. The escape sequences, in their full glory with no dropped characters,
required a refactored parsing routine, which I have now added.
7. backspace->delete mappings, the bane of computer engineers since 1960.
OK. very bald yak. Here I lay down the shears next to several
mooshi pillows stuffed with the finest Yak hair.
just couldn't let it go. figured out the issue is a race condition
in how the Litex UARTs are constructed...I don't know how Litex
normally handles this kind of situation but they have this funky
construction where reading the fifo doesn't advance it, you have
to...clear the pending bit. I think it was meant to be a "clever"
thing that saves an instruction or something like that since you
always have to clear the pending bit.
Anyways, the previous incarnation of this loop checked to see
if the pending bit was set again and if it was it'd run back
and fetch another character. This turns out to be a bad idea
because the pending bit goes through a multireg and other stuff
before it's readable which means it has a lot of delay on it.
This leads to a race condition.
This version actually checks if the fifo is empty and if it isn't,
it reads the character *and* pumps the pending bit (because
the FIFO doesn't auto-advance when you read it, you have to clear
the pending bit).
This works a lot better, but it's not at all well documented how
this works and it's very hard to tell from reading the source code
because it's written using really fancy migen abstractions that
are also poorly documented and have a lot of cursed constructions.
Anyways. At least the console doesn't die now when you type
a control character, *but* you do have the problem that the
escape codes get mis-interpreted sometimes. If it really really
becomes important later on to fix that it would involve tracing
out the sequence of keys that arrive at InjectKey and figuring
out if things are coming out of order or what is triggering the
bad escape codes.
there is still some mysterious problem involving a race condition
of the raw hardware console UART Rx path deciding it doesn't want
to report new characters anymore to the interrupt handler that
is triggered when multi-byte sequences are shoved down the
pipe. So long as you're typing simple commands it's fine, but
escape sequences tend to lock things up. I don't think there's
enough ROI to continue investigating this facility.
Previously, the memlcd had 22 bytes of dirty bits following a line. This
was reduced to 2 bytes, but the logic checking the dirty bytes had not
been updated.
Reduce the dirty byte count to 2 bytes.
Signed-off-by: Sean Cross <sean@xobs.io>
We run a branch of minifb because we need the `set_key_state`
function, which is not available on the main branch. This
function is necessary to extract raw scan keycodes from the
keyboard so we can respond to e.g. menu and prediction emulation
keys (Home and Fn keys). The mainline minifb crate eats all
of those key events and does not pass them on.
the re-regsitration is a "send" not a "lend" now, because
the imef can't block during the context switch.
Not sure why I made it a lend in the first place, probably just
wasn't thinking...
only cone connection is allowed to the IMEF anyways, because
the GAM does traffic control on who has authorization to access it.
So, no need for multi-callback support. Saves a tiny bit of RAM.
request_focus is "dangerous" because any random server could
request a UX context to pop up. There is a checked version
that requires a GAM token that should be used, but, this leaves
the question of how does the initial app that's supposed to be
focused gets set.
A small mod to the UxRegistration call is made to check for
the name of the default focus app, and when it is seen, a thread
is fired off to request focus for that app. That allows us
to deprecate this otherwise dangerous call.
trust levels now default to a mid-level, with the password
modals explicitly tested for by the Context logic (the idea
being to deny any untrusted canvases the ability to render
inverted text).
ball bounce app also basically works, and context
switching between the bounce app and the main menu
is working too.