Base.purgeAddressSpace() via madvise() #336
Conversation
| enum MADV_DONTNEED = 4; | ||
| enum MADV_FREE = 8; |
There was a problem hiding this comment.
Put them in each platform, there is no guarantee that values are the same.
| enum Advise { | ||
| Purge = MADV_DONTNEED | ||
| } |
There was a problem hiding this comment.
This does not match definitions in mman.h . If there is a notion of purge for the GC, it belongs in the GC.
| @@ -86,6 +86,18 @@ public: | |||
| return (cast(Base*) &this).reserveAddressSpaceImpl(size, alignment); | |||
| } | |||
|
|
|||
| void purgeAddressSpace(void* address, size_t size) shared { | |||
There was a problem hiding this comment.
Why is this in base? it's using none of the data from the base, so that doesn't seems necessary. In addition, why is this taking the lock?
There was a problem hiding this comment.
It's in base because that's where mmap and all related system calls are found.
Where precisely do you want it? And what's the reasoning for it not needing the lock?
There was a problem hiding this comment.
It is not using anything in base. It's not about where I want it, it's about where it belongs. You need to be finding these thing out or we are condemned to have endless back and forth.
There was a problem hiding this comment.
Would it be appropriate to move the madvise etc. to purgeAddressSpace per se?
There was a problem hiding this comment.
What is purgeAdressSpace for here?
Split from #335 , but does not depend on it.
This feature permits the purging of an arbitrary run of dirty pages, rendering it clean, as requested in #270 .
Uses the "lazy"
madviseoperationMADV_FREEon Linux (where supported) and the equiv. but eagerMADV_DONTNEEDoperation on all other supported platforms.AFAIK not feasible to unit test, as the OS provides no immediate confirmation of the effects of
madvise.Required by #337 .