Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1750770AbWAXWGq (ORCPT ); Tue, 24 Jan 2006 17:06:46 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1750769AbWAXWGq (ORCPT ); Tue, 24 Jan 2006 17:06:46 -0500 Received: from adsl-70-250-156-241.dsl.austtx.swbell.net ([70.250.156.241]:60344 "EHLO gw.microgate.com") by vger.kernel.org with ESMTP id S1750770AbWAXWGp (ORCPT ); Tue, 24 Jan 2006 17:06:45 -0500 Subject: Re: pppd oopses current linu's git tree on disconnect From: Paul Fulghum To: Diego Calleja Cc: alan@lxorguk.ukuu.org.uk, linux-kernel@vger.kernel.org In-Reply-To: <20060124044846.de6508eb.diegocg@gmail.com> References: <20060119010601.f259bb32.diegocg@gmail.com> <1137692039.3279.1.camel@amdx2.microgate.com> <20060119230746.ea78fcf4.diegocg@gmail.com> <43D01537.40705@microgate.com> <20060123034243.22ba0a8f.diegocg@gmail.com> <20060124044846.de6508eb.diegocg@gmail.com> Content-Type: text/plain; charset=utf-8 Date: Tue, 24 Jan 2006 16:06:31 -0600 Message-Id: <1138140391.3223.15.camel@amdx2.microgate.com> Mime-Version: 1.0 X-Mailer: Evolution 2.2.3 (2.2.3-2.fc4) Content-Transfer-Encoding: 8bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 7542 Lines: 236 On Tue, 2006-01-24 at 04:48 +0100, Diego Calleja wrote: > El Mon, 23 Jan 2006 03:42:43 +0100, > Diego Calleja escribió: > > > Ok - I seem to be able to reproduce it in a unpatched kernel using > > a lengthy session. I will try your patch and report back. > > It doesn't seems to happen today at least, after a lentghty session > similar to the one which triggered the bug. I reasonably sure that your problem is lack of locking for the new tty buffering scheme. There is a question of how to best fix it with minimal code change and impact on performance. Alan: Look at the attached patches against 2.6.16-rc1 1. I added an active flag for each tty buffer 2. I added locking to anything that manipulates the buffer lists or sets/clears an active flag This uses a lock dedicated to the tty buffer lists instead of using the tty->read_lock to reduce lock contention. 3. Calls to tty insertion functions set the active flag for the tail buffer: tty_buffer_request_room() tty_insert_flip_string() tty_insert_flip_string_flags() tty_prepare_flip_string() tty_prepare_flip_string_flags() tty_insert_flip_char() 4. The active flag is cleared when calling tty_schedule_flip() con_schedule_flip() tty_flip_buffer_push() This lets the tail buffer be 'owned' by the driver while filling it. flush_to_ldisc stops when an active buffer is encountered. Calling the scheduling functions in #4 releases the buffer for use by flush_to_ldisc. This scheme centralizes the locking so every driver does not need to be modified, and avoids excessive locking/unlocking for each inserted character. The lock is grabbed once by the driver to activate (get ownership of) the tail buffer. The driver fills the buffer (possibly one char at a time). When the driver is done, the buffer is marked inactive and flush_to_ldisc can process it. I've done some initial testing with success. Let me know what you think. --- linux-2.6.16-rc1/drivers/char/tty_io.c 2006-01-24 15:31:45.000000000 -0600 +++ linux-2.6.16-rc1-mg/drivers/char/tty_io.c 2006-01-24 15:33:28.000000000 -0600 @@ -253,6 +253,7 @@ static void tty_buffer_free_all(struct t static void tty_buffer_init(struct tty_struct *tty) { + spin_lock_init(&tty->buf.lock); tty->buf.head = NULL; tty->buf.tail = NULL; tty->buf.free = NULL; @@ -266,6 +267,7 @@ static struct tty_buffer *tty_buffer_all p->used = 0; p->size = size; p->next = NULL; + p->active = 0; p->char_buf_ptr = (char *)(p->data); p->flag_buf_ptr = (unsigned char *)p->char_buf_ptr + size; /* printk("Flip create %p\n", p); */ @@ -312,25 +314,36 @@ static struct tty_buffer *tty_buffer_fin int tty_buffer_request_room(struct tty_struct *tty, size_t size) { - struct tty_buffer *b = tty->buf.tail, *n; + struct tty_buffer *b, *n; int left = 0; + unsigned long flags; + + spin_lock_irqsave(&tty->buf.lock, flags); /* OPTIMISATION: We could keep a per tty "zero" sized buffer to remove this conditional if its worth it. This would be invisible to the callers */ - if(b != NULL) + if ((b = tty->buf.tail) != NULL) left = b->size - b->used; if(left >= size) - return size; + goto done; + /* This is the slow path - looking for new buffers to use */ n = tty_buffer_find(tty, size); - if(n == NULL) - return left; + if(n == NULL) { + size = left; + goto done; + } if(b != NULL) b->next = n; else tty->buf.head = n; tty->buf.tail = n; + +done: + if (tty->buf.tail != NULL) + tty->buf.tail->active = 1; + spin_unlock_irqrestore(&tty->buf.lock, flags); return size; } @@ -396,10 +409,12 @@ EXPORT_SYMBOL_GPL(tty_insert_flip_string int tty_prepare_flip_string(struct tty_struct *tty, unsigned char **chars, size_t size) { int space = tty_buffer_request_room(tty, size); - struct tty_buffer *tb = tty->buf.tail; - *chars = tb->char_buf_ptr + tb->used; - memset(tb->flag_buf_ptr + tb->used, TTY_NORMAL, space); - tb->used += space; + if (space) { + struct tty_buffer *tb = tty->buf.tail; + *chars = tb->char_buf_ptr + tb->used; + memset(tb->flag_buf_ptr + tb->used, TTY_NORMAL, space); + tb->used += space; + } return space; } @@ -416,10 +431,12 @@ EXPORT_SYMBOL_GPL(tty_prepare_flip_strin int tty_prepare_flip_string_flags(struct tty_struct *tty, unsigned char **chars, char **flags, size_t size) { int space = tty_buffer_request_room(tty, size); - struct tty_buffer *tb = tty->buf.tail; - *chars = tb->char_buf_ptr + tb->used; - *flags = tb->flag_buf_ptr + tb->used; - tb->used += space; + if (space) { + struct tty_buffer *tb = tty->buf.tail; + *chars = tb->char_buf_ptr + tb->used; + *flags = tb->flag_buf_ptr + tb->used; + tb->used += space; + } return space; } @@ -2748,7 +2765,7 @@ static void flush_to_ldisc(void *private goto out; } spin_lock_irqsave(&tty->read_lock, flags); - while((tbuf = tty->buf.head) != NULL) { + while((tbuf = tty->buf.head) != NULL && !tbuf->active) { tty->buf.head = tbuf->next; if (tty->buf.head == NULL) tty->buf.tail = NULL; @@ -2852,6 +2869,12 @@ EXPORT_SYMBOL(tty_get_baud_rate); void tty_flip_buffer_push(struct tty_struct *tty) { + unsigned long flags; + spin_lock_irqsave(&tty->buf.lock, flags); + if (tty->buf.tail != NULL) + tty->buf.tail->active = 0; + spin_unlock_irqrestore(&tty->buf.lock, flags); + if (tty->low_latency) flush_to_ldisc((void *) tty); else --- linux-2.6.16-rc1/include/linux/tty.h 2006-01-17 09:31:30.000000000 -0600 +++ linux-2.6.16-rc1-mg/include/linux/tty.h 2006-01-24 14:12:19.000000000 -0600 @@ -57,6 +57,7 @@ struct tty_buffer { unsigned char *flag_buf_ptr; int used; int size; + int active; /* Data points here */ unsigned long data[0]; }; @@ -64,6 +65,7 @@ struct tty_buffer { struct tty_bufhead { struct work_struct work; struct semaphore pty_sem; + spinlock_t lock; struct tty_buffer *head; /* Queue head */ struct tty_buffer *tail; /* Active buffer */ struct tty_buffer *free; /* Free queue head */ --- linux-2.6.16-rc1/include/linux/tty_flip.h 2006-01-24 12:24:47.000000000 -0600 +++ linux-2.6.16-rc1-mg/include/linux/tty_flip.h 2006-01-24 14:46:59.000000000 -0600 @@ -17,7 +17,7 @@ _INLINE_ int tty_insert_flip_char(struct unsigned char ch, char flag) { struct tty_buffer *tb = tty->buf.tail; - if (tb && tb->used < tb->size) { + if (tb && tb->active && tb->used < tb->size) { tb->flag_buf_ptr[tb->used] = flag; tb->char_buf_ptr[tb->used++] = ch; return 1; @@ -27,6 +27,11 @@ _INLINE_ int tty_insert_flip_char(struct _INLINE_ void tty_schedule_flip(struct tty_struct *tty) { + unsigned long flags; + spin_lock_irqsave(&tty->buf.lock, flags); + if (tty->buf.tail != NULL) + tty->buf.tail->active = 0; + spin_unlock_irqrestore(&tty->buf.lock, flags); schedule_delayed_work(&tty->buf.work, 1); } --- linux-2.6.16-rc1/include/linux/kbd_kern.h 2006-01-17 09:31:29.000000000 -0600 +++ linux-2.6.16-rc1-mg/include/linux/kbd_kern.h 2006-01-24 15:38:19.000000000 -0600 @@ -151,6 +151,11 @@ extern unsigned int keymap_count; static inline void con_schedule_flip(struct tty_struct *t) { + unsigned long flags; + spin_lock_irqsave(&t->buf.lock, flags); + if (t->buf.tail != NULL) + t->buf.tail->active = 0; + spin_unlock_irqrestore(&t->buf.lock, flags); schedule_work(&t->buf.work); } - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/