kernel: process: try to use memory fixed address #1928
Conversation
kernel/src/process.rs
Outdated
| // We ran out of memory. Return the original range since (in | ||
| // this step) we were unable to meet the process's | ||
| // requirements. | ||
| (remaining_app_memory, remaining_app_memory_size) |
There was a problem hiding this comment.
This is confusing. Why would this not return an error here?
There was a problem hiding this comment.
My thought was that this step is very narrowly defined in the scope of what it can do. However, perhaps the MPU step could figure something better out that would support the application.
kernel/src/process.rs
Outdated
| // Address already matches, or is earlier in memory, nothing we | ||
| // have to do (or can do). | ||
| (remaining_app_memory, remaining_app_memory_size) |
There was a problem hiding this comment.
Same question, why not return an error here?
kernel/src/process.rs
Outdated
| // Right now, we only support skipping some RAM and leaving a chunk | ||
| // unused so that the memory region starts where the process needs it | ||
| // to. | ||
| let (new_memory_start, new_memory_size) = |
There was a problem hiding this comment.
| let (new_memory_start, new_memory_size) = | |
| let (requested_memory_start, requested_memory_size) = |
"new" is confusing.., whether fixed or not, I think maybe "requested" better expresses that here it's what the loader is requesting.
Now that there are variations of "memory_start", the variables below should probably also change to be something more specific, maybe:
-let (memory_start, memory_size) = match chip.mpu().allocate_app_memory_region(
+let (actual_memory_start, actual_memory_size) = match chip.mpu().allocate_app_memory_region(
bradjc
added
the
blocked
|
I'll mark this blocked for now, since it really requires the updates to libtock-c to be usable, and there are other, more pressing, issues to work on in process.rs at the moment. |
|
I have updated this for after #2022 is merged. |
bradjc
force-pushed
the
kernel-process-find-memory-address
branch
from
15ff477
to
cd20475
Compare
bradjc
force-pushed
the
kernel-process-find-memory-address
branch
from
cd20475
to
15ff477
Compare
bradjc
removed
the
blocked
bradjc
force-pushed
the
kernel-process-find-memory-address
branch
from
15ff477
to
694021b
Compare
|
Ok! This is ready to go now. |
bradjc
added
the
release-blocker
|
@ppannuto This is the last piece (I think) for basic fixed address support to be reasonably usable. |
Pull Request Overview
This pull request adds logic to process.rs to make it try to find a matching memory address if a specific address was requested by the app. Right now all it tries to do is push the app back in memory (i.e. start the app's memory region at a later address).
This also changes the return values from
create(). This fixes the trace process loading debugging messages so that the correct memory region is printed if an app does not start at the beginning of the memory region.Testing Strategy
Running multiple apps on arty-e21.
TODO or Help Wanted
n/a
Documentation Updated
/docs, or no updates are required.Formatting
make prepush.