On Tue, Dec 27, 2005 at 04:22:22PM +0200, Shachar Shemesh wrote: > The ultimate upshot of all of the above is that Wine jumps throgh hoops > trying to make sure that the memory image under Linux is the same as the > memory image under Windows for loaded programs. I will not be the least > surprised if the above call has something to do with it.
I think that's pretty much a given, as that's the only thing modifying the LDT is good for :-) (where "memory image" encompasses addressing and protection). > I have seen some patches that told mmap to bloody well start allocating > memory from the lower 1GB if the rest of the memory has been exhuasted, > but I don't know whether they made it into the mainstream kernel. > Anticipating your next question - yes, I have had a case where this > actually mattered. Current 2.6 on i386 will start searching for mmap'able areas from TASK_UNMAPPED_BASE == TASK_SIZE / 3 == 0x40000000 (1GB and higher). When the search wraps around, we will restart from the same place (or so says my quick reading of mm/mmap.c:arch_get_unmapped_area()). In other words, no, I don't think such patches made it into mainline. Do you have a URL to such a patch? do you know if this is purely an implementation issue or whether there's a reason why TASK_UNMAPPED_BASE isn't below 1GB in the virtual address space? Cheers, Muli -- Muli Ben-Yehuda http://www.mulix.org | http://mulix.livejournal.com/ ================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]