Forward port
Pull in changes in ckolivas/cgminer (up to commit 8fa5ddf, made on Wed Nov 13 11:57:29 2013 +0100).
This is still useful for all functionality not related to SHA256d/Scrypt mining. Obviously, ckolivas/cgminer developers are few bits more familiar with the code base.
This may be useful in certain scenarios. However, server load from keepalive
is increased 6-fold if code is hard-changed from 30 to 5. So, provide it as
an option instead, and use the previous value as a default (30).
Explanation from
015c064396
Kevin's middlecoin fix, CURL TCP keepalive constants lowered:
CURLOPT_TCP_KEEPIDLE from 45 to 5 and CURLOPT_TCP_KEEPINTVL from 30 to
5. Before it'd trigger a keepalive packet after 45 seconds of connection
idle time and then again every 30 seconds. Now it triggers a keepalive
packet after 5 seconds of connection idle time and then again every 5
seconds.
It makes the client more resilient against coin switching pools or just
pools with connection issues in general. It will however add a tiny bit
pressure to the pool server; but a TCP keepalive probe is only about
60-80 bytes, so I don't think it is an issue.
It is exactly what it says, now you can finally put that Frankenrig configuration in one file.
The parameter follows the same design as gpu-engine and gpu-memclock.
Changed GPU status line from "A:xxxxx R:xxx" to "R:0.0%". It shows (rejected/(accepted+rejected))*100, which is the percentage of rejections for a GPU. Total accepted and rejected in top status is unchanged.
GPU status lines now show both threads and intensity.
It is possible to switch back to full date and time on each log line with "--log-dateformat 1". The short time-only format is default with the value of 0.
Log with short time-only format looks like:
[23:59:59] Accepted 389c47d1 Diff 1.16K/512 GPU 0 pool 0
[00:00:16] Log date is now 2013-12-22
[00:00:16] Stratum from pool 0 detected new block
[00:00:33] Accepted 7a4950a4 Diff 536/512 GPU 0 pool 0
Removing it means existing shell scripts fail unless they are updated
to remove passing the option. Instead, just print a deprecation
warning.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
We were copying the *pointer* to the perf levels structure rather
than the structure itself, and thus restoring the perf levels
did basically nothing.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>