RLBox - wasm sandboxed libraries built with incorrect flags
Categories
(Firefox Build System :: General, defect, P3)
Tracking
(firefox74 fixed)
| Tracking | Status | |
|---|---|---|
| firefox74 | --- | fixed |
People
(Reporter: shravanrn, Assigned: froydnj)
References
()
Details
Attachments
(1 file)
When investigating a different bug, I noticed that the wasm sandboxed library (graphite) compiled in tree is being compiled with incorrect flags. Per the prototype available here, the wasm libraries need to be compiled with flags the specify the memory and guard page size. However, when investigating the binaries produced in the nightly Firefox 73.0a1 (2020-01-03) (64-bit), it looks like the produced binary was compiled without these flags.
We would need to fix this as these flags are particularly important for performance, and in certain scenarios are part of the assumptions made by RLBox.
@Nathan - Any chance you have some time to look at this as you are more familiar with the build system? Feel free to re-assign this to me or anyone else as appropriate.
|
Assignee | |
Comment 1•6 years ago
|
||
Sure, that seems pretty straightforward to fix.
|
|
Assignee | |
Comment 2•6 years ago
|
||
...and pull lucetc options out into a separate variable for readability.
|
Reporter | |
Comment 4•6 years ago
|
||
Just a heads up, this bug is responsible for some of the crashes in 1606977(wikipedia crash).
|
||
Comment 5•6 years ago
|
||
I submitted the crash we were looking at last night, here it is: https://crash-stats.mozilla.org/report/index/cc55f6cb-4832-41fb-9c8f-9196a0200107
|
|
Assignee | |
Comment 6•6 years ago
|
||
(In reply to Deian Stefan from comment #5)
I submitted the crash we were looking at last night, here it is: https://crash-stats.mozilla.org/report/index/cc55f6cb-4832-41fb-9c8f-9196a0200107
That crash says it's due to SIGILL; is lucetc generating instructions that aren't supported on your CPU?
|
|
||
Comment 8•6 years ago
|
||
(In reply to Nathan Froyd [:froydnj] from comment #6)
(In reply to Deian Stefan from comment #5)
I submitted the crash we were looking at last night, here it is: https://crash-stats.mozilla.org/report/index/cc55f6cb-4832-41fb-9c8f-9196a0200107
That crash says it's due to
SIGILL; is lucetc generating instructions that aren't supported on your CPU?
The crash is due to a failed bounds check (which compiles to ud2). I think this is the right behavior, but we probably want to investigate why this is happening in this mode at some point (maybe we make assumptions about not using this compilation method).
|
|
||
Comment 9•6 years ago
|
||
| bugherder | ||
|
|
Reporter | |
Comment 10•6 years ago
•
|
||
but we probably want to investigate why this is happening in this mode at some point (maybe we make assumptions about not using this compilation method).
We do indeed assume a heap of 4GB, so the crash is completely expected.
That crash says it's due to SIGILL;
The crash is due to a failed bounds check (which compiles to ud2). I think this is the right behavior
Yup!
Description
•