On Feb 16, 2011, at 4:54 PM, Jim Wise wrote:
> Jim Wise <jwise@...> writes:
>> Jim Wise <jwise@...> writes:
>>> Juho Snellman <jsnell@...> writes:
>>>> Now that sourceforge is finally back, it's time for a long overdue release. I'll do it this weekend, so until then testing good, breaking more stuff bad.
>>> Cool! Solaris support for x86_64 will be broken in this release due to
>>> page size issues (1.0.45 works fine, and is on the download page) -- I'm
>>> working on this, but won't have a fix in for 1.0.46. Solaris x86 works
>>> fine, and I'll upload a build as soon as the release is tagged.
>> Actually, I have a fix for the Solaris x86_64 thing which I'll put on
>> launchpad tomorrow morning (still testing the best way to integrate
>> So, if I can fix the sb-posix readdir thing in time for this weekend,
>> 1.0.46 can include builds for both Solaris x86 and Solaris x86_64.
> So, this boils down to revision 1.5 of src/compiler/x86_64/backend-parms.lisp,
> which changed the compiled-in value of *backend-page-bytes* from 4k to
> 32k, which Solaris x86-64 can't support in the default configuration.
I think if you change os_vm_page_size to be BACKEND_PAGE_BYTES instead of the sysconf call it is right now, it'd work again. There seems to be a misunderstanding in the SBCL source code as to what kinds of "pages" are being talked about, OS VM pages, or GC pages. The constraint in SBCL is that GC pages should be some multiple of OS VM page size, not that they are equal. But not all the os-specific code seems to know that...
Probably os_vm_page_size should just disappear.