IBM Mainframe Discussion List <[email protected]> wrote on 06/12/2008 10:40:47 PM:
> On Thu, 12 Jun 2008 13:55:31 -0500, Wayne Driscoll wrote: > > >My guess as to why the OC is used is because fork will use a Copy on Write > >(COW) protocol for the new address space. They run through the JPA and do > >an OC on each page in order to get the modules copied over. This is > >probably since they can't just re-load all the modules, since source library > >info isn't saved. The sad part of this logic, is that if the program does a > >FORK->EXEC (which is pretty standard), the EXEC will get rid of most of the > >loaded modules. Garth Godfrey or Erin Farr would have the definitive answer > >to this. > > > I'm mystified. What's the purpose of this? Doesn't touching > every page frustrate the very purpose of COW, which is to defer > the copy until the page is actually modified? Storing, yes. Fetching, no. The comments in the code say that the intent was to reference each page (I would think to drive the deferred loading and relocation of adcons, etc), in which case the use of an OC which does a fetch-and-store was simply a mistake which can be correct by using a fetch-only instruction. > And if the module > is properly refreshable as marked, it should never be modified, > and need never be copied -- the pages can remain shared indefinitely. > Once the deferred pages are actually loaded and relocation is done. The relocation is done using real addresses, bypassing DAT, so it is not affected by the page protection used by COW and REFRPROT. >A few years ago, there was some discussion here about the > complexities of the deferred fetch capabilities of Content > Supervision. Is this going on because Content Supervision is > incapable of dealing with modules shared between address > spaces. That discussion had to do with the problems (or to use the euphemism, "challenges") of deferred loading from LNKLST vs. dynamic LNKLST updating. >If so, the Right Fix is a massive undertaking. Well, lets hope not. The designers of deferred loading retired or moved on, and weren't replaced by any masses to undertake anything massive. Jim Mulder z/OS System Test IBM Corp. Poughkeepsie, NY ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
