[Linux] virtual_memory() fails when /proc/meminfo contains ShadowCallStack >= 10MB

[hanson-openai]

Summary

• OS: Linux
• Architecture: ARM
• Psutil version: 7.1.2
• Python version: 3.12.9
• Type: core

Description

On an arm64 Linux machine, psutil.virtual_memory() fails when /proc/meminfo contains a line like:

ShadowCallStack:10373888 kB

Because the parsing in _pslinux.py breaks when no space is present after the metric name:

https://github.com/giampaolo/psutil/blob/a59dae7/psutil/_pslinux.py#L258-L259

            fields = line.split()
            mems[fields[0]] = int(fields[1]) * 1024

The upstream cause is actually due to a quirk in the Linux kernel, where for some reason ShadowCallStack is printed with padding up to 8 digits (none of the other meminfo fields are like this) but without a guaranteed space:

https://github.com/torvalds/linux/blob/8356a5a3b078ca89c526dd6d71e9a76fec571c37/fs/proc/meminfo.c#L113-L116

	seq_printf(m, "ShadowCallStack:%8lu kB\n",

So whenever ShadowCallStack >= 10MB this issue occurs! Suggested fix below.

[hanson-openai]

suggested alternative - happy to submit a quick PR!

# Note: in linux/fs/proc/meminfo.c, some metrics may not have a space after the colon
key, value = line.split(b':', 1)
mems[key + b':'] = int(value.split()[0]) * 1024

@giampaolo

[giampaolo]

Yup! Go for it.

[hanson-openai]

done! (oops just realized the issue name should say GB not MB.)