- 24 Jul, 2013 30 commits
-
-
Peter Hurley authored
Acquiring the write_wait queue spin lock now accounts for the largest slice of cpu time on the tty write path. Two factors contribute to this situation; a overly-pessimistic line discipline write loop which _always_ sets up a wait loop even if i/o will immediately succeed, and on ptys, a wakeup storm from reads and writes. Writer wakeup does not need to be performed by the pty driver. Firstly, since the actual i/o is performed within the write, the line discipline write loop will continue while space remains in the flip buffers. Secondly, when space becomes avail in the line discipline receive buffer (and thus also in the flip buffers), the pty unthrottle re-wakes the writer (non-flow-controlled line disciplines unconditionally unthrottle the driver when data is received). Thus, existing in-kernel i/o is guaranteed to advance. Finally, writer wakeup occurs at the conclusion of the line discipline write (in tty_write_unlock()). This guarantees that any user-space write waiters are woken to continue additional i/o. Signed-off-by:
Peter Hurley <peter@hurleysoftware.com> Signed-off-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-
Peter Hurley authored
LNEXT processing accounts for ~15% of total cpu time in end-to-end tty i/o; factor the lnext test/clear from the per-char i/o path. Instead, attempt to immediately handle the literal next char if not at the end of this received buffer; otherwise, handle the first char of the next received buffer as the literal next char, then continue with normal i/o. Signed-off-by:
-
Peter Hurley authored
gcc will likely inline these single-use functions anyway; remove inline modifier. Signed-off-by:
-
Peter Hurley authored
Always pre-figure the space available in the read_buf and limit the inbound receive request to that amount. For compatibility reasons with the non-flow-controlled interface, n_tty_receive_buf() will continue filling read_buf until all data has been received or receive_room() returns 0. Signed-off-by:
-
Peter Hurley authored
Handle PARMRK processing on the slow per-char i/o path. Signed-off-by:
-
Peter Hurley authored
Convert to modal receive_buf processing; factor char receive processing for unusual termios settings out of normal per-char i/o path. Signed-off-by:
-
Peter Hurley authored
Factor 'special' per-char processing into standalone fn, n_tty_receive_char_special(), which handles processing for chars marked in the char_map. Signed-off-by:
-
Peter Hurley authored
Relocate the IXANY restart tty test to code paths where the the received char is not START_CHAR, STOP_CHAR, INTR_CHAR, QUIT_CHAR or SUSP_CHAR. Fixes the condition when ISIG if off and one of INTR_CHAR, QUIT_CHAR or SUSP_CHAR does not restart i/o. Signed-off-by:
-
Peter Hurley authored
Simplify __receive_buf() into a dispatch function; perform per-char processing for all other modes not already handled. Signed-off-by:
-
Peter Hurley authored
Commit 20bafb3d 'n_tty: Move buffers into n_tty_data' broke the ppc64 build. Include vmalloc.h for the required function declarations. Reported-by:
Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by:
-
Peter Hurley authored
Convert to modal receive_buf() processing; factor receive char processing when tty->closing into n_tty_receive_buf_closing(). Note that EXTPROC when ISTRIP or IUCLC is set continues to be handled by n_tty_receive_char(). Signed-off-by:
-
Peter Hurley authored
When EXTPROC is set without ISTRIP or IUCLC, processing is identical to raw mode; handle this receiving mode as a special-case of raw mode. Signed-off-by:
-
Peter Hurley authored
Convert to modal receive_buf() processing; factor raw mode per-char i/o into n_tty_receive_buf_raw(). Signed-off-by:
-
Peter Hurley authored
Prepare for modal receive_buf() handling; factor handling for TTY_BREAK, TTY_PARITY, TTY_FRAME and TTY_OVERRUN into n_tty_receive_char_flagged(). Signed-off-by:
-
Peter Hurley authored
Reduce the monolithic n_tty_receive_char() complexity; factor the handling of INTR_CHAR, QUIT_CHAR and SUSP_CHAR into n_tty_receive_signal_char(). Signed-off-by:
-
Peter Hurley authored
Convert to modal receive_buf() processing; factor real_raw receive_buf() into n_tty_receive_buf_real_raw(). Signed-off-by:
-
Peter Hurley authored
Signed-off-by:
-
Peter Hurley authored
Signed-off-by:
-
Peter Hurley authored
Reduce pointer reloading and improve locality-of-reference; allocate read_buf and echo_buf within struct n_tty_data. Signed-off-by:
-
Peter Hurley authored
The char and flag buffer local alias pointers, p and f, are unnecessary; remove them. Signed-off-by:
-
Peter Hurley authored
In canonical mode, an EOF which is not the first character of the line causes read() to complete and return the number of characters read so far (commonly referred to as EOF push). However, if the previous read() returned because the user buffer was full _and_ the next character is an EOF not at the beginning of the line, read() must not return 0, thus mistakenly indicating the end-of-file condition. The TTY_PUSH flag is used to indicate an EOF was received which is not at the beginning of the line. Because the EOF push condition is evaluated by a thread other than the read(), multiple EOF pushes can cause a premature end-of-file to be indicated. Instead, discover the 'EOF push as first read character' condition from the read() thread itself, and restart the i/o loop if detected. Signed-off-by:
-
Peter Hurley authored
Separate the head & commit indices from the tail index to avoid cache-line contention (so called 'false-sharing') between concurrent threads. Signed-off-by:
-
Peter Hurley authored
Since neither echo_commit nor echo_tail can change for the duration of __process_echoes loop, substitute index comparison for the snapshot counter. Signed-off-by:
-
Peter Hurley authored
Don't have the driver flush received echoes if no echoes were actually output. Signed-off-by:
-
Peter Hurley authored
Byte-by-byte echo output is painfully slow, requiring a lock/unlock cycle for every input byte. Instead, perform the echo output in blocks of 256 characters, and at least once per flip buffer receive. Enough space is reserved in the echo buffer to guarantee a full block can be saved without overrunning the echo output. Overrun is prevented by discarding the oldest echoes until enough space exists in the echo buffer to receive at least a full block of new echoes. Signed-off-by:
-
Peter Hurley authored
Use output_lock mutex as a memory barrier when storing echo_commit. Signed-off-by:
-
Peter Hurley authored
Adding data to echo_buf (via add_echo_byte()) is guaranteed to be single-threaded, since all callers are from the n_tty_receive_buf() path. Processing the echo_buf can be called from either the n_tty_receive_buf() path or the n_tty_write() path; however, these callers are already serialized by output_lock. Publish cumulative echo_head changes to echo_commit; process echo_buf from echo_tail to echo_commit; remove echo_lock. On echo_buf overrun, claim output_lock to serialize changes to echo_tail. Signed-off-by:
-
Peter Hurley authored
Prepare for lockless echo_buf handling; compute current byte count of echo_buf from head and tail indices. Signed-off-by:
-
Peter Hurley authored
Instead of using a single index to track the current echo_buf position, use a head index when adding to the buffer and a tail index when consuming from the buffer. Allow these head and tail indices to wrap at max representable value; perform modulo reduction via helper functions when accessing the buffer. Signed-off-by:
-
Peter Hurley authored
The echo_overrun field is only assigned and never tested; remove it. Signed-off-by:
-
- 23 Jul, 2013 10 commits
-
-
Peter Hurley authored
TTY_BUFFER_PAGE is only used within drivers/tty/tty_buffer.c; relocate to that file scope. Signed-off-by:
-
Peter Hurley authored
Convert the tty_buffer_flush() exclusion mechanism to a public interface - tty_buffer_lock/unlock_exclusive() - and use the interface to safely write the paste selection to the line discipline. Signed-off-by:
-
Peter Hurley authored
__tty_flush_buffer() is now only called by tty_flush_buffer(); merge functions. Signed-off-by:
-
Peter Hurley authored
Atomic bit ops are no longer required to indicate a flip buffer flush is pending, as the flush_mutex is sufficient barrier. Remove the unnecessary port .iflags field and localize flip buffer state to struct tty_bufhead. Signed-off-by:
-
Peter Hurley authored
Separate the head and tail ptrs to avoid cache-line contention (so called 'false-sharing') between concurrent threads. Signed-off-by:
-
Peter Hurley authored
Now that dropping the buffer lock is not necessary (as result of converting the spin lock to a mutex), the flip buffer flush no longer needs to be handled by the buffer work. Simply signal a flush is required; the buffer work will exit the i/o loop, which allows tty_buffer_flush() to proceed. Signed-off-by:
-
Peter Hurley authored
The buffer work may race with parallel tty_buffer_flush. Use a mutex to guarantee exclusive modify access to the head flip buffer. Remove the unneeded spin lock. Signed-off-by:
-
Peter Hurley authored
Driver-side flip buffer input is already single-threaded; 'publish' the .next link as the last operation on the tail buffer so the 'consumer' sees the already-completed flip buffer. The commit buffer index is already 'published' by driver-side functions. Signed-off-by:
-
Peter Hurley authored
Lockless flip buffers require atomically updating the bytes-in-use watermark. The pty driver also peeks at the watermark value to limit memory consumption to a much lower value than the default; query the watermark with new fn, tty_buffer_space_avail(). Signed-off-by:
-
Peter Hurley authored
Use a 0-sized sentinel to avoid assigning the head ptr from the driver side thread. This also eliminates testing head/tail for NULL. When the sentinel is first 'consumed' by the buffer work (or by tty_buffer_flush()), it is detached from the list but not freed nor added to the free list. Both buffer work and tty_buffer_flush() continue to preserve at least 1 flip buffer to which head & tail is pointed. Signed-off-by:
-
