Commit Graph

1 Commits (e3fb645821d20621c02cea66ecd1b6e32fd5dc6c)

Author SHA1 Message Date
Xunlei Pang 34986d538a 99base: add memtrace-ko.sh to debug kernel module large memory consumption
The current method for memory debug is to use "rd.memdebug=[0-3]",
it is not enough for debugging kernel modules. For example, when we
want to find out which kernel module consumes a large amount of memory,
"rd.memdebug=[0-3]" won't help too much.

A better way is needed to achieve this requirement, this is useful for
kdump OOM debugging.

The principle of this patch is to use kernel trace to track slab and
buddy allocation calls during kernel module loading(module_init), thus
we can analyze all the trace data and get the total memory consumption.
As for large slab allocation, it will probably fall into buddy allocation,
thus tracing "mm_page_alloc" alone should be enough for the purpose(this
saves quite some trace buffer memory, also large free is quite unlikey
during module loading, we neglect those memory free events).

The trace events include memory calls under "tracing/events/":
  kmem/mm_page_alloc

We also inpect the following events to detect the module loading:
  module/module_load
  module/module_put

Since we use filters to trace events, the final trace data size won't
be too big. Users can adjust the trace buffer size via "trace_buf_size"
kernel boot command line as needed.

We can get the module name and task pid from "module_load" event which
also mark the beginning of the loading, and module_put called by the
same task pid implies the end of the loading. So the memory events
recorded in between by the same task pid are consumed by this module
during loading(i.e. modprobe or module_init()).

With these information, we can record the rough total memory(the larger,
the more precise the result will be) consumption involved by each kernel
module loading.

Thus we introduce this shell script to find out which kernel module
consumes a large amount of memory during loading. Use "rd.memdebug=4"
as the tigger.

After applying this patch and specifying "rd.memdebug=4", during booting
it will print out something extra like below:
0 pages consumed by "pata_acpi"
0 pages consumed by "ata_generic"
1 pages consumed by "drm"
0 pages consumed by "ttm"
0 pages consumed by "drm_kms_helper"
835 pages consumed by "qxl"
0 pages consumed by "mii"
6 pages consumed by "8139cp"
0 pages consumed by "virtio"
0 pages consumed by "virtio_ring"
9 pages consumed by "virtio_pci"
1 pages consumed by "8139too"
0 pages consumed by "serio_raw"
0 pages consumed by "crc32c_intel"
199 pages consumed by "virtio_console"
0 pages consumed by "libcrc32c"
9 pages consumed by "xfs"

From the print, we see clearly that "qxl" consumed the most memory.

This file will be installed as a separate executable named "tracekomem"
in the following patch.

Signed-off-by: Xunlei Pang <xlpang@redhat.com>
2016-11-27 12:09:10 +08:00