John Morrison wrote on Fri, May 28, 2004 at 04:36:26PM -0400: > Hi Martin; > > On Friday 28 May 2004 04:03 pm, you wrote: > > They changed the memory mapping of executables. > > Agh. If it's not broken, change it, eh? > > > There is some unidentified space reserved at 0x2a7e6000 to 0x54f4f000 > > where you can do MAP_FIXED mmap fine but segfault. Gotta love these > > people :-) > > So much for Linux stability.
FreeBSD all the way! Well, theoretically it is for the better because in the new Fedora core 2 memory scheme you can actually exhaust memory my mmap at unspecified addresses. In previous Linux version you had to manually fill the holes below the shared libs. > > I am designing a memory map for ITA which will work on both and will > > upload the stuff when it's ready. > > With the one CMUCL you are building, how much foreign (C++, in case you > care) heap will I be able to have? I understand current CMUCL builds > restrict this to some modest number of MBytes. I don't think this is what is happening. Calling malloc from C works fine for us to exhaust memory that is not taken up by CMUCL spaces. Our build leaves a lot of C space for mmap or malloc (glibc malloc will fall back to mmap when brk fails). You should be fine, but you have to take into account that we "right-size" most spaces for us which will mean if you need more space in one of them you are screwed. Unfortunately I'm running into moderate trouble with fragmentation. If I place the CMUCL spaces so that they don't clash with anything neither under Linux nor Fedora core 2 I give up the large unfragmented block of mmap'able memory about 0x4400. Not sure we want to risk fragmentation and then getting hit by new data files not fitting in one piece. We might prefer to kick Fedora Core 2 to the curb instead. Martin -- %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Martin Cracauer <[EMAIL PROTECTED]> http://www.cons.org/cracauer/ No warranty. This email is probably produced by one of my cats stepping on the keys. No, I don't have an infinite number of cats.
