On 12/6/2021 3:31 PM, Arnd Bergmann wrote:
On Mon, Dec 6, 2021 at 10:52 AM Sai Prakash Ranjan
<quic_saipraka@xxxxxxxxxxx> wrote:
Yes just the trace after read/write won't serve our usecase where we
expect crashes/hangs on accessing
these registers but internally we did have a log_post_read_mmio() as
well, if it is useful then I can add it.
Are there any downsides to tracing both before and after, besides another growth
in binary size? Aside from the 'value', that would also allow
measuring the time it
takes to complete a readl(), which may be valuable for other users as these
can be significant.
Ah yes, that would be useful. No downsides as far as I know other than
the size
but that should be fine given this depends on ftrace.
Not sure how to best do that that, we could return a timestamp from the 'before'
tracepoint and pass it into the 'after' tracepoint in order to log the
difference, or just
rely on calculating the differences in user space based on the log.
For trace events, timing information is already logged by ftrace
infrastructure. Most of the users do
use these for timing information based on post processing these logs
looking at these timestamps,
so we should be good using that as well.
For the 'write' style accessors, the timing data would be less interesting, at
least for posted PCI transactions, but it may be helpful to do the same for
symmetry reasons.
Ok, I will add these post read/write logging in the next version.
Thanks,
Sai
