n_tty: Fix PARMRK over-throttling
If PARMRK is enabled, the available read buffer space computation is overly-pessimistic, which results in severely throttled i/o, even in the absence of parity errors. For example, if the 4k read buffer contains 1k processed data, the input worker will compute available space of 333 bytes, despite 3k being available. At 1365 chars of processed data, 0 space available is computed. *Divide remaining space* by 3, truncating down (if left == 2, left = 0). Reported-by: Christian Riesch <christian.riesch@omicron.at> Conflicts: drivers/tty/n_tty.c Signed-off-by: Peter Hurley <peter@hurleysoftware.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This commit is contained in:
parent
70aca71f92
commit
06c49f9fa3
1 changed files with 4 additions and 8 deletions
|
@ -1669,9 +1669,8 @@ n_tty_receive_buf_common(struct tty_struct *tty, const unsigned char *cp,
|
|||
|
||||
while (1) {
|
||||
/*
|
||||
* When PARMRK is set, multiply read_cnt by 3, since each byte
|
||||
* might take up to three times as many spaces (depending on
|
||||
* its flags, e.g. parity error). [This calculation is wrong.]
|
||||
* When PARMRK is set, each input char may take up to 3 chars
|
||||
* in the read buf; reduce the buffer space avail by 3x
|
||||
*
|
||||
* If we are doing input canonicalization, and there are no
|
||||
* pending newlines, let characters through without limit, so
|
||||
|
@ -1683,13 +1682,10 @@ n_tty_receive_buf_common(struct tty_struct *tty, const unsigned char *cp,
|
|||
* read_tail (so this producer will not overwrite unread data)
|
||||
*/
|
||||
size_t tail = smp_load_acquire(&ldata->read_tail);
|
||||
size_t head = ldata->read_head;
|
||||
|
||||
room = N_TTY_BUF_SIZE - (ldata->read_head - tail) - 1;
|
||||
if (I_PARMRK(tty))
|
||||
room = N_TTY_BUF_SIZE - (head - tail) * 3 - 1;
|
||||
else
|
||||
room = N_TTY_BUF_SIZE - (head - tail) - 1;
|
||||
|
||||
room /= 3;
|
||||
if (room <= 0)
|
||||
room = ldata->icanon && ldata->canon_head == tail;
|
||||
|
||||
|
|
Loading…
Reference in a new issue