Re: [PATCH-v2] Set irq thread to RT priority on creation
* Ivo Sieben | 2013-06-03 12:12:02 [+0200]: this is in -tip so I take this for v3.8-rt Sebastian -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH-v2] Set irq thread to RT priority on creation
* Ivo Sieben | 2013-06-03 12:12:02 [+0200]: this is in -tip so I take this for v3.8-rt Sebastian -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH-v2] Set irq thread to RT priority on creation
Hi Thomas, 2013/6/3 Thomas Gleixner : > > The question is why there is data present in the UART when the UART > driver has not initialized the UART. Up to the point where the UART is > opened and the interrupt handler is installed the receiver should be > disabled. Also there is the question why a flush does not kill > everything including the pending data in the UART itself. > > And I don't think, that your issue is solved completely. Assume the > following: > > Open UART > Flush Buffers (including UART fifo) > > ---> UART receives data > ---> Interrupt > Data is available in tty buffer > > Write data to UART > > Receive data from UART > > You still get data which got into the buffer before you sent stuff > out. So your application should not be surpriced at all by random data > on the receive line when it starts up. > > Thanks, > > tglx You are absolutely right, the real issue was that my UART device still received data while the device was closed. And yes, the protocol that we use should be robust to unexpected data. I solved & will solve these problems now. So you indeed the uart explanation in my commit log should be removed. The point is that we while we were debugging & tracing this problem, we didn't expect this behavior: in the trace we saw a threaded handler scheduled in on 'normal' priority which surprised us. Also I think there are situations thinkable (altough I cannot come up with a proper example) where it is normal behavior that a IRQ directly kicks in after enabling the interrupts. Regards, Ivo Sieben -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH-v2] Set irq thread to RT priority on creation
Hi Thomas, 2013/6/3 Thomas Gleixner t...@linutronix.de: The question is why there is data present in the UART when the UART driver has not initialized the UART. Up to the point where the UART is opened and the interrupt handler is installed the receiver should be disabled. Also there is the question why a flush does not kill everything including the pending data in the UART itself. And I don't think, that your issue is solved completely. Assume the following: Open UART Flush Buffers (including UART fifo) --- UART receives data --- Interrupt Data is available in tty buffer Write data to UART Receive data from UART You still get data which got into the buffer before you sent stuff out. So your application should not be surpriced at all by random data on the receive line when it starts up. Thanks, tglx You are absolutely right, the real issue was that my UART device still received data while the device was closed. And yes, the protocol that we use should be robust to unexpected data. I solved will solve these problems now. So you indeed the uart explanation in my commit log should be removed. The point is that we while we were debugging tracing this problem, we didn't expect this behavior: in the trace we saw a threaded handler scheduled in on 'normal' priority which surprised us. Also I think there are situations thinkable (altough I cannot come up with a proper example) where it is normal behavior that a IRQ directly kicks in after enabling the interrupts. Regards, Ivo Sieben -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH-v2] Set irq thread to RT priority on creation
On Mon, 3 Jun 2013, Ivo Sieben wrote: > When a threaded irq handler is installed the irq thread is initially created > on normal scheduling priority. Only after the the irq thread is woken up it > immediately sets its priority to RT_FIFO MAX_USER_RT_PRIO/2. > > This means that interrupts that occur directly after the irq handler is > installed will be handled on a normal scheduling priority instead of the > realtime priority that you would expect. Fixed this by setting the RT > priority on creation of the irq_thread. > > This solves the following issue with a UART device driver: > On start of the application there is already data present in the uart RX > fifo buffer. On opening the uart device the hard & threaded interrupt > handlers are installed and the interrupts are enabled. Immediately a hard > irq occurs because there is still data in the RX fifo. Because the threaded > irq handler is still on normal scheduling, my application is not immediatly > interrupted by the threaded handler and continues to run: it tries to flush > the uart input buffer and writes data to the uart device. After this the > threaded handler finally gets scheduled in and fills the buffer with the > "old" received data. When my application reads data from the uart port it > receives the "old" data and gives an error. While I in principle agree with the patch, the issue you are describing is just solved accidentaly. The question is why there is data present in the UART when the UART driver has not initialized the UART. Up to the point where the UART is opened and the interrupt handler is installed the receiver should be disabled. Also there is the question why a flush does not kill everything including the pending data in the UART itself. And I don't think, that your issue is solved completely. Assume the following: Open UART Flush Buffers (including UART fifo) ---> UART receives data ---> Interrupt Data is available in tty buffer Write data to UART Receive data from UART You still get data which got into the buffer before you sent stuff out. So your application should not be surpriced at all by random data on the receive line when it starts up. Thanks, tglx -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[PATCH-v2] Set irq thread to RT priority on creation
When a threaded irq handler is installed the irq thread is initially created on normal scheduling priority. Only after the the irq thread is woken up it immediately sets its priority to RT_FIFO MAX_USER_RT_PRIO/2. This means that interrupts that occur directly after the irq handler is installed will be handled on a normal scheduling priority instead of the realtime priority that you would expect. Fixed this by setting the RT priority on creation of the irq_thread. This solves the following issue with a UART device driver: On start of the application there is already data present in the uart RX fifo buffer. On opening the uart device the hard & threaded interrupt handlers are installed and the interrupts are enabled. Immediately a hard irq occurs because there is still data in the RX fifo. Because the threaded irq handler is still on normal scheduling, my application is not immediatly interrupted by the threaded handler and continues to run: it tries to flush the uart input buffer and writes data to the uart device. After this the threaded handler finally gets scheduled in and fills the buffer with the "old" received data. When my application reads data from the uart port it receives the "old" data and gives an error. Signed-off-by: Ivo Sieben --- v2: * Removed the sched_setscheduler() call in irq_thread() function * Updated commit log why this solves an issue for me with a UART driver kernel/irq/manage.c | 11 ++- 1 file changed, 6 insertions(+), 5 deletions(-) diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c index fa17855..e16caa8 100644 --- a/kernel/irq/manage.c +++ b/kernel/irq/manage.c @@ -840,9 +840,6 @@ static void irq_thread_dtor(struct callback_head *unused) static int irq_thread(void *data) { struct callback_head on_exit_work; - static const struct sched_param param = { - .sched_priority = MAX_USER_RT_PRIO/2, - }; struct irqaction *action = data; struct irq_desc *desc = irq_to_desc(action->irq); irqreturn_t (*handler_fn)(struct irq_desc *desc, @@ -854,8 +851,6 @@ static int irq_thread(void *data) else handler_fn = irq_thread_fn; - sched_setscheduler(current, SCHED_FIFO, ); - init_task_work(_exit_work, irq_thread_dtor); task_work_add(current, _exit_work, false); @@ -950,6 +945,9 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new) */ if (new->thread_fn && !nested) { struct task_struct *t; + static const struct sched_param param = { + .sched_priority = MAX_USER_RT_PRIO/2, + }; t = kthread_create(irq_thread, new, "irq/%d-%s", irq, new->name); @@ -957,6 +955,9 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new) ret = PTR_ERR(t); goto out_mput; } + + sched_setscheduler(t, SCHED_FIFO, ); + /* * We keep the reference to the task struct even if * the thread dies to avoid that the interrupt code -- 1.7.9.5 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[PATCH-v2] Set irq thread to RT priority on creation
When a threaded irq handler is installed the irq thread is initially created on normal scheduling priority. Only after the the irq thread is woken up it immediately sets its priority to RT_FIFO MAX_USER_RT_PRIO/2. This means that interrupts that occur directly after the irq handler is installed will be handled on a normal scheduling priority instead of the realtime priority that you would expect. Fixed this by setting the RT priority on creation of the irq_thread. This solves the following issue with a UART device driver: On start of the application there is already data present in the uart RX fifo buffer. On opening the uart device the hard threaded interrupt handlers are installed and the interrupts are enabled. Immediately a hard irq occurs because there is still data in the RX fifo. Because the threaded irq handler is still on normal scheduling, my application is not immediatly interrupted by the threaded handler and continues to run: it tries to flush the uart input buffer and writes data to the uart device. After this the threaded handler finally gets scheduled in and fills the buffer with the old received data. When my application reads data from the uart port it receives the old data and gives an error. Signed-off-by: Ivo Sieben meltedpiano...@gmail.com --- v2: * Removed the sched_setscheduler() call in irq_thread() function * Updated commit log why this solves an issue for me with a UART driver kernel/irq/manage.c | 11 ++- 1 file changed, 6 insertions(+), 5 deletions(-) diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c index fa17855..e16caa8 100644 --- a/kernel/irq/manage.c +++ b/kernel/irq/manage.c @@ -840,9 +840,6 @@ static void irq_thread_dtor(struct callback_head *unused) static int irq_thread(void *data) { struct callback_head on_exit_work; - static const struct sched_param param = { - .sched_priority = MAX_USER_RT_PRIO/2, - }; struct irqaction *action = data; struct irq_desc *desc = irq_to_desc(action-irq); irqreturn_t (*handler_fn)(struct irq_desc *desc, @@ -854,8 +851,6 @@ static int irq_thread(void *data) else handler_fn = irq_thread_fn; - sched_setscheduler(current, SCHED_FIFO, param); - init_task_work(on_exit_work, irq_thread_dtor); task_work_add(current, on_exit_work, false); @@ -950,6 +945,9 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new) */ if (new-thread_fn !nested) { struct task_struct *t; + static const struct sched_param param = { + .sched_priority = MAX_USER_RT_PRIO/2, + }; t = kthread_create(irq_thread, new, irq/%d-%s, irq, new-name); @@ -957,6 +955,9 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new) ret = PTR_ERR(t); goto out_mput; } + + sched_setscheduler(t, SCHED_FIFO, param); + /* * We keep the reference to the task struct even if * the thread dies to avoid that the interrupt code -- 1.7.9.5 -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH-v2] Set irq thread to RT priority on creation
On Mon, 3 Jun 2013, Ivo Sieben wrote: When a threaded irq handler is installed the irq thread is initially created on normal scheduling priority. Only after the the irq thread is woken up it immediately sets its priority to RT_FIFO MAX_USER_RT_PRIO/2. This means that interrupts that occur directly after the irq handler is installed will be handled on a normal scheduling priority instead of the realtime priority that you would expect. Fixed this by setting the RT priority on creation of the irq_thread. This solves the following issue with a UART device driver: On start of the application there is already data present in the uart RX fifo buffer. On opening the uart device the hard threaded interrupt handlers are installed and the interrupts are enabled. Immediately a hard irq occurs because there is still data in the RX fifo. Because the threaded irq handler is still on normal scheduling, my application is not immediatly interrupted by the threaded handler and continues to run: it tries to flush the uart input buffer and writes data to the uart device. After this the threaded handler finally gets scheduled in and fills the buffer with the old received data. When my application reads data from the uart port it receives the old data and gives an error. While I in principle agree with the patch, the issue you are describing is just solved accidentaly. The question is why there is data present in the UART when the UART driver has not initialized the UART. Up to the point where the UART is opened and the interrupt handler is installed the receiver should be disabled. Also there is the question why a flush does not kill everything including the pending data in the UART itself. And I don't think, that your issue is solved completely. Assume the following: Open UART Flush Buffers (including UART fifo) --- UART receives data --- Interrupt Data is available in tty buffer Write data to UART Receive data from UART You still get data which got into the buffer before you sent stuff out. So your application should not be surpriced at all by random data on the receive line when it starts up. Thanks, tglx -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/