userlib: remove panics in kipc implementation #1937
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The kipc stubs tend to assert on the rc value received from the kernel. This isn't ideal: we generally have to trust the kernel, and we can't do anything useful if it misbehaves. So the assert is merely costing a bunch of flash for a condition that can never happen. This removes these asserts. The system_restart kipc needs to indicate a "noreturn" send, but of course we have no way of doing that right now. So, it panicked if the send returned. This created _another_ can't-happen panic site in flash. I've replaced it with an infinite loop that compiles to a two-byte branch instruction, and in practice will never even be reached (but the compiler doesn't know that). This further reduces text and stack size in El Jefe.
hawkw
reviewed
Nov 25, 2024
hawkw
approved these changes
Nov 25, 2024
Co-authored-by: Eliza Weisman <eliza@elizas.website>
@hawkw probably, but the process of finding them and right-sizing them is unpleasant and I'm currently not up for it. |
That's entirely reasonable; I wasn't going to say you had to do it now, or anything. And, I think it can be justified by saying that, in the absence of better automated processes for estimating max stack depths, giving the supervisor a bigger stack margin is probably more useful unless we are actively trying to scrounge up a few more bytes of RAM for some specific purpose, which we're not at present! |
cbiffle
added a commit
that referenced
this pull request
Nov 26, 2024
The supervisor task in Hubris is not permitted to panic, since it's responsible for handling panics. Jefe has historically contained a bunch of (static) panics, many of which aren't actually possible at runtime. I've been gradually grinding away at these in my other PRs. As of #1937, it's now possible to build a _minimal_ Jefe (like we use on the G0) that contains no panics. So I've enabled that on donglet, and turned on the userlib/no-panic feature that will statically ensure it remains true. Turning on dump support in Jefe causes a bunch of panics to reappear, because humpty is panic-heavy. That's a task for another day.
|
@hawkw if you like getting stack back, you're going to love my next PR. ;-) |
cbiffle
added a commit
that referenced
this pull request
Nov 26, 2024
The supervisor task in Hubris is not permitted to panic, since it's responsible for handling panics. Jefe has historically contained a bunch of (static) panics, many of which aren't actually possible at runtime. I've been gradually grinding away at these in my other PRs. As of #1937, it's now possible to build a _minimal_ Jefe (like we use on the G0) that contains no panics. So I've enabled that on donglet, and turned on the userlib/no-panic feature that will statically ensure it remains true. Turning on dump support in Jefe causes a bunch of panics to reappear, because humpty is panic-heavy. That's a task for another day. In addition to eliminating all panic sites, this also eliminates the last indirect function calls from the generated binaries -- meaning, the static stack size estimate is very likely exact. So I have taken the opportunity to shrink the stack by a bit, reclaiming 256 bytes of RAM on this RAM-starved part.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The kipc stubs tend to assert on the rc value received from the kernel. This isn't ideal: we generally have to trust the kernel, and we can't do anything useful if it misbehaves. So the assert is merely costing a bunch of flash for a condition that can never happen. This removes these asserts.
The system_restart kipc needs to indicate a "noreturn" send, but of course we have no way of doing that right now. So, it panicked if the send returned. This created another can't-happen panic site in flash. I've replaced it with an infinite loop that compiles to a two-byte branch instruction, and in practice will never even be reached (but the compiler doesn't know that).
This further reduces text and stack size in El Jefe.