traced_probes crashes in cpu_reader.cc:308 Failed to parse ftrace page (errno: 0, Success) about perfetto HOT 4 CLOSED

This seems like more a bug in Asahi Linux than a bug in traced_probes as it looks like ftrace is passing an "all-zero" page to the parser which should never happen.

Given we don't see this on any non-Mac platforms and the general immaturity of Linux on Apple Silicon in the first place, I don't think there is anything to do on our side: it's probably better to file this against Asahi Linux kernel developers.

from perfetto.

Could it be related to them using 16K pages by default? I'll try to compile with 4K and see what happens. The kernel shouldn't be different from upstream in this regard. But I will double check and report back if there's anything interesting or file a bug upstream if it's a Linux issue.

from perfetto.

I forgot M1 is 16K: it's reasonably likely that might be the issue. There are various places in the Perfetto stack where we "hard-code" the notion of 4K pages. Architecturally, there is nothing stopping us from actually using 16K but it requires making changes at some very core parts of Perfetto which is a somewhat scary task and requires a lot of reasoning whether the 4K page size is because we are interfacing with Linux ftrace or we just happened to choose 4K for some other reason and reused the constant.

from perfetto.

For Googlers, there is an internal bug about 16K pages at b/269635302

from perfetto.