This page has information about various memory management projects and activities which are of interest to embedded Linux developers.
Areas of Interest
Most of these areas have wider reaching implications, but are of relatively simpler in the embedded case, largely thanks to not having to contend with swap and things of that nature. Simpler memory management as well as vendors not afraid of deviation from mainline for product programs makes for an excellent playground for experimenting with new things in the memory management and virtual memory space.
- This applies to both transparent large page usage as well as the more static usage models, primarily relating to work outside of the hugetlb interface/libhugetlbfs.
- Embedded systems suffer from very small TLBs generally using PAGE_SIZE'd pages (4kB) for coverage. In most cases this places the system under very heavy pressure for any kind of userspace work, and very visibly degrading performance, with most applications taking anywhere from 5-40% of their time on the CPU servicing page faults.
- Preliminary discussion on this subject as well as links to additional information is happening through the wiki here: Huge Pages
Page cache compression
- This relates to using various compression algorithms for performing run-time compression and decompression of page cache pages, specifically aimed at both reducing memory pressure as well as helping performance in certain workloads.
- More information can be found on the wiki here CompressedCaching as well as at the SF Compressed Caching home page.
Reserving (and accessing) the top of memory on startup
A quote from Todd's email on how to use the reserved physical memory in "mem=".
Given that you have a fixed address for your memory, and is already reserved, the easier way to use it is by calling mmap() over the /dev/ mem device, use 0 as the start address, and the physical address of the reserved memory as the offset. The flags could be MAP_WRITE| MAP_READ. That will return you a pointer on user space for your memory mapped by the kernel. For example
If your SDRAM base address is 0x80000000 and your memory is of 64MB, but you use the cmdline mem=60M to reserve 4MB at the end. Then your reserved memory will be at 0x83c00000, so all you need to do is
int fd; char *reserved_memory; fd = open("/dev/mem",O_RDWR); reserved_memory = (char *) mmap(0,4*1024*1024,PROT_READ| PROT_WRITE,MAP_SHARED,fd,0x83c00000);
Enhanced Out-Of-Memory (OOM) handling
Several technologies have been developed and suggested for improving the handling out-of-memory conditions with Linux systems.
See http://linux-mm.org/OOM_Killer for information about the OOM killer in the Linux kernel.
Part of OOM avoidance is for the kernel to have an accurate measure of memory utilization. See Accurate Memory Measurement for information on technology in this area.
Here are some I know about (these need to be researched and documented better):
- mem_notify patches
- This set of patches provided a mechanism to notify user-space when memory is getting low, allowing for application-based handling of the condition. These patches were submitted in January 2008.
- See http://lwn.net/Articles/267013/
- Google per-cgroup OOM handler
- Google posted a Request For Comments (RFC) for OOM handling implemented in a per-cgroup fashion. See http://article.gmane.org/gmane.linux.kernel.mm/28376
- Nokia OOM enhancements
- Maemo application enhancements referenced at: http://lwn.net/Articles/267013/ (search for "killable" in the comments)
User "oak" writes (commenting on the mem_notify patches): Posted Feb 3, 2008 14:02 UTC (Sun) by oak (guest, #2786) [Link] ... I thought the point of the patch is for user-space to be able to do the memory management in *manageable places* in code. As mentioned earlier, a lot of user-space code doesn't handle memory allocation failures. And even if it's supposed to be, it can be hard to verify (test) that the failures are handled in *all* cases properly. If user-space can get a pre-notification of a low-memory situation, it can in suitable place in code free memory so that further allocations will succeed (with higher propability). That also allows doing somehing like what maemo does. If system gets notified about kernel low memory shortage, it kills processes which have notified it that they are in "background-killable" state (saved their UI state, able to restore it and not currently visible to user). I think it also notifies applications (currently) through D-BUS about low memory condition. Applications visible to user or otherwise non-background killable are then supposed to free their caches and/or disable features that could take a lot of additional memory. If the caches are from heap instead of memory mapped, it's less likely to help because of heap fragmentation and it requiring more work/time though.
- Paul Mundt submitted a patch to CELF for the 2.6.12 kernel which provided low-memory notifications to user space. See Accurate_Memory_Measurement#Nokia_out-of-memory_notifier_module for more information.
- This module was based on the Linux Security Module system, which has been removed from recent kernels.