Bug 197783
Summary: | System is using much more swap than all processes' VmSwap combined | ||
---|---|---|---|
Product: | Memory Management | Reporter: | Matt Whitlock (kernel) |
Component: | Other | Assignee: | Andrew Morton (akpm) |
Status: | RESOLVED INVALID | ||
Severity: | normal | CC: | nate |
Priority: | P1 | ||
Hardware: | Intel | ||
OS: | Linux | ||
Kernel Version: | 4.13.9-gentoo | Subsystem: | |
Regression: | No | Bisected commit-id: |
Description
Matt Whitlock
2017-11-06 07:12:24 UTC
A possibly related observation: if I run "find / -xdev" in one terminal and "free -s1" in another terminal, I can watch the available memory steadily decline as the find spiders the file system. That's the _available_ memory, not the _free_ memory. Correct me if I'm wrong, but page cache and dentry cache shouldn't be affecting the _available_ memory, as cache entries are supposed to be reclaimable. My file system is XFS. And can anyone explain this? (I turned off my swap after killing enough processes for that to become possible.) # free total used free shared buff/cache available Mem: 8165636 2749000 4902576 26232 514060 4964520 Swap: 0 0 0 # echo 3 > /proc/sys/vm/drop_caches # free total used free shared buff/cache available Mem: 8165636 638988 7328140 26232 198508 7177208 Swap: 0 0 0 Merely dropping the caches increases available memory by more than 2 GB. Again, shouldn't available memory exclude the caches? It's as though something in the page cache or dentry cache is holding giant amounts of unaccountable kernel memory that is only released when the cache objects are destroyed. Any ideas? Note: This report may be related to https://bugs.kde.org/show_bug.cgi?id=368838#c31 Nearly positive that this is due to some kind of leak in plasmashell. The kernel appears to free the memory when plasmashell is terminated. I'm closing this report as invalid, at least until and unless we discover that plasmashell is running into a kernel bug. |