Re: [PATCH] trace_seq: Increase the buffer size to almost two pages

[Google]

On Mon, 11 Dec 2023 13:28:37 -0500
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Mon, 11 Dec 2023 21:46:27 +0900
> Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx> wrote:
> 
> > > 
> > > By increasing the trace_seq buffer to almost two pages, it can now print
> > > out the first line.
> > > 
> > > This also subtracts the rest of the trace_seq fields from the buffer, so
> > > that the entire trace_seq is now PAGE_SIZE aligned.  
> > 
> > Ok, but I just a bit concern about the memory consumption.
> > Since this is very specific case, can we make it configurable later?
> 
> I was concerned about this too, but it looks like it's allocated and later
> freed in every location except for a couple of instances.
> 
> One is "tracepoint_print_iter" which is used to pipe tracepoints to printk.
> I think we can possibly make that allocated too.
> 
> The other is in ftrace_dump, which I don't think we can easily allocate
> that. Although, we could have it allocated at boot up if
> ftrace_dump_on_oops() is enabled.

Can we reallocate it when we detect such bigger event entry in the path
of trace_marker write? If any issue happens in the reallocation, we will
not finish (commit) such big event in dumping buffer anyway.

> 
> Another KTODO?

Yes, I think so.

Thanks,

> 
> > 
> > Reviewed-by: Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
> > 
> 
> Thanks!
> 
> -- Steve


-- 
Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>