tty_port use for isicom
Subsystems (4)
cover tty_port use for isicom
Pending patch proposal plus the isicom fix on the end. Could do with a good review Jiri.
---
Alan Cox (4):
isicom: fix deadlock on shutdown
isicom: sort out the board init logic
isicom: switch to the new tty_port_open helper
tty_port: add "tty_port_open" helper
--
Two fish in a tank: One says to the other "Can you drive this ?"
PATCH 1/4 tty_port: add "tty_port_open" helper
For the moment this just moves the USB logic over and fixes the 'what if we open and hangup at the same time' race noticed by Oliver Neukum.
---
drivers/char/tty_port.c | 36 ++++++++++++++++++++++++++++- drivers/usb/serial/usb-serial.c | 49 ++++++++++++++++----------------------- include/linux/tty.h | 10 +++++++- 3 files changed, 64 insertions(+), 31 deletions(-)
drivers/char/tty_port.c +35 -1
@@ -99,10 +99,11 @@ EXPORT_SYMBOL(tty_port_tty_set);static void tty_port_shutdown (struct tty_port *port) { +mutex_lock (&port->mutex);if (port->ops->shutdown &&test_and_clear_bit (ASYNCB_INITIALIZED , &port->flags)) port->ops->shutdown (port); - +mutex_unlock (&port->mutex); }/**
@@ -375,3 +376,36 @@ void tty_port_close(struct tty_port *port, struct tty_struct *tty,tty_port_tty_set (port,NULL ); }EXPORT_SYMBOL (tty_port_close); + +int tty_port_open (struct tty_port *port,struct tty_struct *tty, +struct file *filp) +{ +spin_lock_irq (&port->lock); +if (!tty_hung_up_p (filp)) + ++port->count; +spin_unlock_irq (&port->lock); +tty_port_tty_set (port, tty); + +/* +* Do the device-specific open only if the hardware isn't +* already initialized. Serialize open and shutdown using the +* port mutex. +*/ + +mutex_lock (&port->mutex); + +if (!test_bit (ASYNCB_INITIALIZED , &port->flags)) { +if (port->ops->activate) { +int retval = port->ops->activate (port, tty); +if (retval) { +mutex_unlock (&port->mutex); +return retval; + } + } +set_bit (ASYNCB_INITIALIZED , &port->flags); + } +mutex_unlock (&port->mutex); +return tty_port_block_til_ready (port, tty, filp); +} + +EXPORT_SYMBOL (tty_port_open);
drivers/usb/serial/usb-serial.c +20 -29
67 lines not on this page · raw message
include/linux/tty.h +9 -1
@@ -191,9 +191,15 @@ struct tty_port_operations {/* Control the DTR line */ void (*dtr_rts)(struct tty_port *port,int raise);/* Called when the last close completes or a hangup finishes -IFF the port was initialized. Do not use to free resources */ +IFF the port was initialized. Do not use to free resources. Called +under the port mutex to serialize against activate/shutdowns */ void (*shutdown)(struct tty_port *port);void (*drop)(struct tty_port *port); +/* Called under the port mutex from tty_port_open, serialized using +the port mutex */ +/* FIXME: long term getting the tty argument *out* of this would be +good for consoles */ +int (*activate)(struct tty_port *port,struct tty_struct *tty); };struct tty_port {
@@ -468,6 +474,8 @@ extern int tty_port_close_start(struct tty_port *port,extern void tty_port_close_end (struct tty_port *port,struct tty_struct *tty);extern void tty_port_close (struct tty_port *port,struct tty_struct *tty,struct file *filp); +extern int tty_port_open (struct tty_port *port, +struct tty_struct *tty,struct file *filp);extern inline int tty_port_users (struct tty_port *port) {return port->count + port->blocked_open;
PATCH 2/4 isicom: switch to the new tty_port_open helper
In reply to Alan CoxTrivial conversion in this case so might as well do it while testing the port_open design is right
---
drivers/char/isicom.c | 88 ++++++++++++------------------------------------- 1 files changed, 21 insertions(+), 67 deletions(-)
drivers/char/isicom.c +21 -67
@@ -804,24 +804,21 @@ static inline void isicom_setup_board(struct isi_board *bp) bp->status |=BOARD_ACTIVE ;for (channel =0 ; channel < bp->port_count; channel++, port++)drop_dtr_rts (port); + bp->count++;spin_unlock_irqrestore (&bp->card_lock, flags); } -static int isicom_setup_port (struct tty_struct *tty) +static int isicom_activate (struct tty_port *tport,struct tty_struct *tty) { -struct isi_port *port = tty->driver_data; + struct isi_port *port = container_of(tport, struct isi_port, port);struct isi_board *card = port->card;unsigned long flags; -if (port->port.flags &ASYNC_INITIALIZED ) -return 0 ; -if (tty_port_alloc_xmit_buf (&port->port) <0 ) +if (tty_port_alloc_xmit_buf (tport) <0 )return -ENOMEM ;spin_lock_irqsave (&card->card_lock, flags); -clear_bit (TTY_IO_ERROR , &tty->flags); -if (port->port.count ==1 ) - card->count++; +isicom_setup_board (card); port->xmit_cnt = port->xmit_head = port->xmit_tail =0 ;
@@ -832,9 +829,7 @@ static int isicom_setup_port(struct tty_struct *tty)outw (((ISICOM_KILLTX |ISICOM_KILLRX ) <<8 ) |0x06 , card->base);InterruptTheCard (card->base); } -isicom_config_port (tty); - port->port.flags |=ASYNC_INITIALIZED ;spin_unlock_irqrestore (&card->card_lock, flags);return 0 ;
@@ -871,31 +866,20 @@ static struct tty_port *isicom_find_port(struct tty_struct *tty)return &port->port; } - +static int isicom_open (struct tty_struct *tty,struct file *filp) {struct isi_port *port;struct isi_board *card;struct tty_port *tport; -int error =0 ; tport =isicom_find_port (tty);if (tport ==NULL )return -ENODEV ; port = container_of(tport, struct isi_port, port); card = &isi_card[BOARD (tty->index)]; -isicom_setup_board (card); -/* FIXME: locking on port.count etc */ - port->port.count++; - tty->driver_data = port; -tty_port_tty_set (&port->port, tty); -/* FIXME: Locking on Initialized flag */ -if (!test_bit (ASYNCB_INITIALIZED , &tport->flags)) - error =isicom_setup_port (tty); -if (error ==0 ) - error =tty_port_block_til_ready (&port->port, tty, filp); -return error; +return tty_port_open (tport, tty, filp); }/* close et all */
@@ -914,40 +898,21 @@ static void isicom_shutdown_port(struct isi_port *port) tty =tty_port_tty_get (&port->port); -if (!(port->port.flags &ASYNC_INITIALIZED )) { -tty_kref_put (tty); -return ; - } -tty_port_free_xmit_buf (&port->port); - port->port.flags &= ~ASYNC_INITIALIZED ; -/* 3rd October 2000 : Vinayak P Risbud */ -tty_port_tty_set (&port->port,NULL ); - -/*Fix done by Anil .S on 30-04-2001 -remote login through isi port has dtr toggle problem -due to which the carrier drops before the password prompt -appears on the remote end. Now we drop the dtr only if the -HUPCL(Hangup on close) flag is set for the tty*/ - -if (C_HUPCL (tty)) -/* drop dtr on this port */ -drop_dtr (port); - -/* any other port uninits */ -if (tty) -set_bit (TTY_IO_ERROR , &tty->flags); -if (--card->count <0 ) {pr_dbg ("isicom_shutdown_port: bad board(0x%lx) count %d.\n" , card->base, card->count); card->count =0 ; } -/* last port was closed, shutdown that boad too */ -if (C_HUPCL (tty)) { -if (!card->count) -isicom_shutdown_board (card); +/* last port was closed, shutdown that board too */ +if (tty &&C_HUPCL (tty)) { +/* FIXME: this logic is bogus - it's the old logic that was +bogus before but it still wants fixing */ +if (!card->count) { +if (card->status &BOARD_ACTIVE ) + card->status &= ~BOARD_ACTIVE ; + } }tty_kref_put (tty); }
@@ -968,7 +933,7 @@ static void isicom_flush_buffer(struct tty_struct *tty)tty_wakeup (tty); } -static void isicom_close_port (struct tty_port *port) +static void isicom_shutdown (struct tty_port *port) { struct isi_port *ip = container_of(port, struct isi_port, port);struct isi_board *card = ip->card;
@@ -977,10 +942,8 @@ static void isicom_close_port(struct tty_port *port)/* indicate to the card that no more data can be received on this port */ spin_lock_irqsave (&card->card_lock, flags); -if (port->flags &ASYNC_INITIALIZED ) { - card->port_status &= ~(1 << ip->channel); -outw (card->port_status, card->base +0x02 ); - } + card->port_status &= ~(1 << ip->channel); +outw (card->port_status, card->base +0x02 );isicom_shutdown_port (ip);spin_unlock_irqrestore (&card->card_lock, flags); }
@@ -991,12 +954,7 @@ static void isicom_close(struct tty_struct *tty, struct file *filp)struct tty_port *port = &ip->port;if (isicom_paranoia_check (ip, tty->name,"isicom_close" ))return ; - -if (tty_port_close_start (port, tty, filp) ==0 ) -return ; -isicom_close_port (port); -isicom_flush_buffer (tty); -tty_port_close_end (port, tty); +tty_port_close (port, tty, filp); }/* write et all */
@@ -1326,15 +1284,9 @@ static void isicom_start(struct tty_struct *tty)static void isicom_hangup (struct tty_struct *tty) {struct isi_port *port = tty->driver_data; -unsigned long flags;if (isicom_paranoia_check (port, tty->name,"isicom_hangup" ))return ; - -spin_lock_irqsave (&port->card->card_lock, flags); -isicom_shutdown_port (port); -spin_unlock_irqrestore (&port->card->card_lock, flags); -tty_port_hangup (&port->port); }
@@ -1367,6 +1319,8 @@ static const struct tty_operations isicom_ops = {static const struct tty_port_operations isicom_port_ops = { .carrier_raised = isicom_carrier_raised, .dtr_rts = isicom_dtr_rts, + .activate = isicom_activate, + .shutdown = isicom_shutdown, }; static int __devinit reset_card(struct pci_dev *pdev,
PATCH 3/4 isicom: sort out the board init logic
In reply to Alan CoxSplit this into two flags - INIT meaning the board is set up and ACTIVE meaning the board has ports open. Remove the broken HUPCL casing and push the counts somewhere sensible.
---
drivers/char/isicom.c | 41 +++++++++++------------------------------ include/linux/isicom.h | 1 + 2 files changed, 12 insertions(+), 30 deletions(-)
drivers/char/isicom.c +11 -30
@@ -793,21 +793,19 @@ static inline void isicom_setup_board(struct isi_board *bp) {int channel;struct isi_port *port; -unsigned long flags; -spin_lock_irqsave (&bp->card_lock, flags); -if (bp->status &BOARD_ACTIVE ) { -spin_unlock_irqrestore (&bp->card_lock, flags); -return ; - } - port = bp->ports; - bp->status |=BOARD_ACTIVE ; -for (channel =0 ; channel < bp->port_count; channel++, port++) -drop_dtr_rts (port); bp->count++; -spin_unlock_irqrestore (&bp->card_lock, flags); +if (!(bp->status &BOARD_INIT )) { + port = bp->ports; +for (channel =0 ; channel < bp->port_count; channel++, port++) +drop_dtr_rts (port); + } + bp->status |=BOARD_ACTIVE |BOARD_INIT ; } +/* Activate and thus setup board are protected from races against shutdown +by the tty_port mutex */ +static int isicom_activate (struct tty_port *tport,struct tty_struct *tty) { struct isi_port *port = container_of(tport, struct isi_port, port);
@@ -884,19 +882,10 @@ static int isicom_open(struct tty_struct *tty, struct file *filp)/* close et all */ -static inline void isicom_shutdown_board (struct isi_board *bp) -{ -if (bp->status &BOARD_ACTIVE ) - bp->status &= ~BOARD_ACTIVE ; -} -/* card->lock HAS to be held */ static void isicom_shutdown_port (struct isi_port *port) {struct isi_board *card = port->card; -struct tty_struct *tty; - - tty =tty_port_tty_get (&port->port);tty_port_free_xmit_buf (&port->port);if (--card->count <0 ) {
@@ -904,17 +893,9 @@ static void isicom_shutdown_port(struct isi_port *port) card->base, card->count); card->count =0 ; } -/* last port was closed, shutdown that board too */ -if (tty &&C_HUPCL (tty)) { -/* FIXME: this logic is bogus - it's the old logic that was -bogus before but it still wants fixing */ -if (!card->count) { -if (card->status &BOARD_ACTIVE ) - card->status &= ~BOARD_ACTIVE ; - } - } -tty_kref_put (tty); +if (!card->count) + card->status &=BOARD_ACTIVE ; }static void isicom_flush_buffer (struct tty_struct *tty)
include/linux/isicom.h +1 -0
@@ -67,6 +67,7 @@#define FIRMWARE_LOADED 0x0001#define BOARD_ACTIVE 0x0002 +#define BOARD_INIT 0x0004/* isi_port status bitmap */
PATCH 4/4 isicom: fix deadlock on shutdown
In reply to Alan CoxAlexander Strakh <[email protected]> reported
KERNEL_VERSION: 2.6.31 DESCRIBE:
Driver drivers/char/isicom.c might sleep in atomic context, because it callstty_port_xmit_buf under spin_lock.
./drivers/char/isicom.c:
1307 static void isicom_hangup(struct tty_struct *tty) 1308 {
...
1315 spin_lock_irqsave(&port->card->card_lock, flags);
1316 isicom_shutdown_port(port);
...
Path to might_sleep macro from isicom_hangup:
1. isicom_hangup calls spin_lock_irqsave (drivers/char/isicom.c:1315) and then
calls isicom_shutdown_port.
2. isiscom_shutdown_port calls tty_port_free_xmit_buf at drivers/char/isicom.c:906
3. tty_port_free_xmit_buf calls mutex_lock at drivers/char/tty_port:48
Found by Linux Driver Verification Project.
---
drivers/char/isicom.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/char/isicom.c +1 -1
@@ -887,7 +887,6 @@ static void isicom_shutdown_port(struct isi_port *port) {struct isi_board *card = port->card; -tty_port_free_xmit_buf (&port->port);if (--card->count <0 ) {pr_dbg ("isicom_shutdown_port: bad board(0x%lx) count %d.\n" , card->base, card->count);
@@ -927,6 +926,7 @@ static void isicom_shutdown(struct tty_port *port)outw (card->port_status, card->base +0x02 );isicom_shutdown_port (ip);spin_unlock_irqrestore (&card->card_lock, flags); +tty_port_free_xmit_buf (port); }static void isicom_close (struct tty_struct *tty,struct file *filp)
Re: [PATCH 2/4] isicom: switch to the new tty_port_open helper
In reply to Alan Coxdrivers/char/isicom.c static inline void isicom_setup_board(struct isi_board *bp) · line 817
Quoted 4 lines · Alan Cox
drivers/char/isicom.c +1
-3-if (port->port.flags &ASYNC_INITIALIZED ) -return 0 ; -if (tty_port_alloc_xmit_buf (&port->port) <0 ) +if (tty_port_alloc_xmit_buf (tport) <0 )
I think this won't work well with 1/4. ->activate is called with port->port.mutex (the name in this context), but tty_port_alloc_xmit_buf wants it again.
Re: [PATCH 2/4] isicom: switch to the new tty_port_open helper
Quoted 9 lines · On Wed, 07 Oct 2009 22:11:01 +0200 Jiri Slaby <[email protected]> wrote:
drivers/char/isicom.c static inline void isicom_setup_board(struct isi_board *bp) · line 817
Quoted 4 lines · Alan Cox
drivers/char/isicom.c +1
-3-if (port->port.flags &ASYNC_INITIALIZED ) -return 0 ; -if (tty_port_alloc_xmit_buf (&port->port) <0 ) +if (tty_port_alloc_xmit_buf (tport) <0 )I think this won't work well with 1/4. ->activate is called with
port->port.mutex(the name in this context), buttty_port_alloc_xmit_bufwants it again.
Indeed - so the port buffer allocation stuff wants revisiting. I'll do that tomorrow.
Re: [PATCH 2/4] isicom: switch to the new tty_port_open helper
In reply to Jiri SlabyQuoted 9 lines · On Wed, 07 Oct 2009 22:11:01 +0200 Jiri Slaby <[email protected]> wrote:
drivers/char/isicom.c static inline void isicom_setup_board(struct isi_board *bp) · line 817
Quoted 4 lines · Alan Cox
drivers/char/isicom.c +1
-3-if (port->port.flags &ASYNC_INITIALIZED ) -return 0 ; -if (tty_port_alloc_xmit_buf (&port->port) <0 ) +if (tty_port_alloc_xmit_buf (tport) <0 )I think this won't work well with 1/4. ->activate is called with
port->port.mutex(the name in this context), buttty_port_alloc_xmit_bufwants it again.
Ok fixed that the simple way the buffer allocator now has its own private mutex. Any other problems strike you ?
Re: [PATCH 2/4] isicom: switch to the new tty_port_open helper
Quoted 2 lines · Alan Cox
Ok fixed that the simple way the buffer allocator now has its own private mutex. Any other problems strike you ?
I don't see any. It looks good. Thanks.