release_console_sem() disables interrupts while it calls upon the
console drivers to dump the pending characters. This disabling of
interrupt is delaying the handling of interrupts leading to weird
latencies and bugs.
The specific scenario is while resuming, the __log_buf has accumulated
around 11,000 bytes of characters and resume_console takes for ever to
send them out. On my system with serial running at 115200 baud it takes
almost a second to flush 11,000 bytes.
I was thinking about limiting the characters to say N bytes and before
the next set of N bytes is sent, we enable interrupts and do the
spin_lock_irqsave again. Basically spin in the for ( ; ; ) loop until
all the outstanding bytes are sent in sets of N bytes.
This issue has been brought up many times earlier -
http://www.webservertalk.com/archive242-2006-6-1537307.html
http://www.webservertalk.com/archive242-2007-11-2129981.html
One reason I gather is that we want to prevent interrupts from printing
unorderly stuff in the __log_buf. But dont interrupts get the
spin_lock(&logbuf_lock) and then update the log_[end/start], con_start
etc? If so we wont see unorderly prints in the __log_buf, the spin_lock
protects those pointers.
Would like ask if limiting the number of characters per loop in
release_console_sem() is a good idea.
Thanks
Abhijeet
Sent by an Employee of the Qualcomm Innovation Center, Inc. The Qualcomm
Innovation Center, Inc is a member of code aurora forum.