Messages in this thread | | | From | (Kanoj Sarcar) | Subject | Re: [RFC] [RFT] Shared /dev/zero mmaping feature | Date | Wed, 1 Mar 2000 09:34:53 -0800 (PST) |
| |
> > Hi Kanoj and Linus, > > kanoj@google.engr.sgi.com (Kanoj Sarcar) writes: > > > [snip] > > > > I do not believe there is any good reason to expose the special > > shared memory segment used as a place holder for all /dev/zero > > mappings to users via ipc* commands. This special segment exists > > only because we want to reduce kernel code duplication, and all the > > zshmid_kernel/ zero_id checks just make sure that regular shared > > memory works pretty much the way it did before. (One thing I am > > unhappy about is that this special segment eats up a shm id, but > > that's probably not too bad). > > The appended proposal reduces code duplication and complexity a > lot. (The diff47 needs your patches against other files added.) >
What you have sent is what I used as a first draft for the implementation. The good part of it is that it reduces code duplication. The _really_ bad part is that it penalizes users in terms of numbers of shared memory segments, max size of /dev/zero mappings, and limitations imposed by shm_ctlmax/shm_ctlall/shm_ctlmni etc. I do not think taking up a shmid for each /dev/zero mapping is a good idea ...
Furthermore, I did not want to change behavior of information returned by ipc* and various procfs commands, as well as swapout behavior, thus the creation of the zmap_list. I decided a few lines of special case checking in a handful of places was a much better option.
If the current /dev/zero stuff hampers any plans you have with shm code (eg page cachification), I would be willing to talk about it ...
Kanoj
> I would vote to apply diff48 to the standard kernel. For me the whole > solution is still a workaround. > > Greetings > Christoph
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |