Javascript:Hazard Builds: Difference between revisions

(terrence gone.)
(→‎Static Analysis for Rooting Hazards: stuff about heap writes)
Line 1: Line 1:
== Static Analysis for Rooting Hazards ==
== Static Analysis for Rooting and Heap Write Hazards ==


Treeherder can run two static analysis builds: the full browser (linux64-haz), just the JS shell (linux64-shell-haz). They show up on treeherder as '''H''' and '''SM(H)'''.
Treeherder can run two static analysis builds: the full browser (linux64-haz), just the JS shell (linux64-shell-haz). They show up on treeherder as '''H''' and '''SM(H)'''.


=== Diagnosing a hazard failure ===
=== Diagnosing a hazard failure ===
The first step is to look at what sort of hazard is being reported. There are two types that cause the job to fail: stack rooting hazards for garbage collection, and heap write thread safety hazards for stylo.
The summary output will include either the string "<N> rooting hazards detected" or "<N> heap write hazards detected out of <M> allowed". See the appropriate section below for each.
=== Diagnosing a rooting hazards failure ===


Click on the '''H''' build link, select the "Job details" pane on the bottom right, follow the "Inspect Task" link, and download the "public/build/hazards.txt.gz" file.
Click on the '''H''' build link, select the "Job details" pane on the bottom right, follow the "Inspect Task" link, and download the "public/build/hazards.txt.gz" file.
Line 44: Line 50:
* during the resulting garbage collection, the object pointed to by ed.obj is moved to a different location. All pointers stored in the JS heap are updated automatically, as are all rooted pointers. ed.obj is not, because the GC doesn't know about it.
* during the resulting garbage collection, the object pointed to by ed.obj is moved to a different location. All pointers stored in the JS heap are updated automatically, as are all rooted pointers. ed.obj is not, because the GC doesn't know about it.
* after decompilePC returns, something accesses ed.obj. This is now a stale pointer, and may refer to just about anything -- the wrong object, an invalid object, or whatever. Badness 10000, as TeX would say.
* after decompilePC returns, something accesses ed.obj. This is now a stale pointer, and may refer to just about anything -- the wrong object, an invalid object, or whatever. Badness 10000, as TeX would say.
=== Diagnosing a heap write hazard failure ===
For the thread unsafe heap write analysis, a hazard means that some Gecko_* function calls, directly or indirectly, code that writes to something on the heap, or calls an unknown function that *might* write to something on the heap. The analysis requires quite a few annotations to describe things that are actually safe. This section will be expanded as we gain more experience with the analysis, but here are some common issues:
* Adding a new Gecko_* function: often, you will need to annotate any outparams or owned (thread-local) parameters in the <code>treatAsSafeArgument</code> function in <code>js/src/devtools/rootAnalysis/analyzeHeapWrites.js</code>.
* Calling some libc function: if you add a call to some random libc function (eg sin() or floor() or ceil(), though the latter two are already annotated), the analysis will report an "External Function". Add it to <code>checkExternalFunction</code>, assuming it *doesn't* have the possibility of writing to shared heap memory.
* If you call some non-returning (crashing) function that the analysis doesn't know about, you'll need to add it to <code>ignoreContents</code>.
On the other hand, you might have a real thread safety issue on your hands. Shared caches are common problems. Fix it.


=== Analysis implementation ===
=== Analysis implementation ===
Line 81: Line 97:
The most common way to fix a hazard is to change the variable to be a Rooted type, as described in http://dxr.mozilla.org/mozilla-central/source/js/public/RootingAPI.h#l21
The most common way to fix a hazard is to change the variable to be a Rooted type, as described in http://dxr.mozilla.org/mozilla-central/source/js/public/RootingAPI.h#l21


For more complicated cases, ask on #jsapi. If you don't get a response, ping sfink or jonco.
For more complicated cases, ask on #jsapi. If you don't get a response, ping sfink or jonco for rooting hazards, bholley or sfink for heap write hazards. Or if it's a deeper issue with the analysis logic, try bhackett (the author of both analyses.)
Confirmed users
328

edits