[PATCH 0/4] avoid disabling interrupts where it is not required

[PATCH 0/4] avoid disabling interrupts where it is not required

[Sebastian Andrzej Siewior]

This series avoids/limits disabling of interrupts in drivers/ide.
This is from the -RT queue where it was reported initially (ata was not
widespread back then) and it hurts -RT.

While looking at alim15x3, there are some drivers which exist in both
worlds (alim15x3 included was far as I can tell). Is there a reason why
we still keep two drivers around for the same hardware? I know on mips
bigsur and swarm select IDE in their defconfig. I fixed (ages ago) ide
for swarm while I was using ata myself. So *I* think this just a relict.
If Ralf's bigsur is still alive, it is using SATA and not the onboard
IDE.

Sebastian

[PATCH 1/4] alim15x3: move irq-restore before pci_dev_put()

[Sebastian Andrzej Siewior]

init_chipset_ali15x3() initializes the chipset during init with disabled
interrupts. There is no need to keep the interrupts disabled during
pci_dev_put().
Move the irq-restore before pci_dev_put() is invoked.

Side note: The same init is performed in
drivers/ata/pata_ali.c::ali_init_chipset() without disabled interrupts.
It looks that the same hardware is supported in the ATA land. Would it
make sense to remove this driver since it is supported in the other
subsystem?

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 drivers/ide/alim15x3.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/ide/alim15x3.c b/drivers/ide/alim15x3.c
index 36f76e28a0bf..3265970aee34 100644
--- a/drivers/ide/alim15x3.c
+++ b/drivers/ide/alim15x3.c
@@ -323,9 +323,9 @@ static int init_chipset_ali15x3(struct pci_dev *dev)
 
 		pci_write_config_byte(dev, 0x53, tmpbyte);
 	}
+	local_irq_restore(flags);
 	pci_dev_put(north);
 	pci_dev_put(isa_dev);
-	local_irq_restore(flags);
 	return 0;
 }
 
-- 
2.17.0

[PATCH 2/4] ide: Handle irq disabling consistently

[Sebastian Andrzej Siewior]

ide_timer_expiry() disables interrupt at function entry when acquiring
hwif->lock. Before disabling the device interrupt it unlocks hwif->lock,
but interrupts stay disabled. After the call to disable_irq() interrupts
are disabled again, which is a pointless exercise.

After the device irq handler has been invoked with interrupts disabled,
hwif->lock is acquired again with spin_lock_irq() because the device irq
handler might have reenabled interrupts. This is not documented and
confusing for the casual reader.

Remove the redundant local_irq_disable() and add a comment which explains
why hwif->lock has to be reacquired with spin_lock_irq().

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 drivers/ide/ide-io.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/ide/ide-io.c b/drivers/ide/ide-io.c
index 6f25da56a169..a444bad7a2aa 100644
--- a/drivers/ide/ide-io.c
+++ b/drivers/ide/ide-io.c
@@ -659,8 +659,7 @@ void ide_timer_expiry (struct timer_list *t)
 		spin_unlock(&hwif->lock);
 		/* disable_irq_nosync ?? */
 		disable_irq(hwif->irq);
-		/* local CPU only, as if we were handling an interrupt */
-		local_irq_disable();
+
 		if (hwif->polling) {
 			startstop = handler(drive);
 		} else if (drive_is_ready(drive)) {
@@ -679,6 +678,7 @@ void ide_timer_expiry (struct timer_list *t)
 				startstop = ide_error(drive, "irq timeout",
 					hwif->tp_ops->read_status(hwif));
 		}
+		/* Disable interrupts again, `handler' might have enabled it */
 		spin_lock_irq(&hwif->lock);
 		enable_irq(hwif->irq);
 		if (startstop == ide_stopped && hwif->polling == 0) {
-- 
2.17.0

[PATCH 3/4] ide: don't disable interrupts during kmap_atomic()

[Sebastian Andrzej Siewior]

ide_pio_bytes() disables interrupts around kmap_atomic(). This is a
leftover from the old kmap_atomic() implementation which relied on fixed
mapping slots, so the caller had to make sure that the same slot could not
be reused from an interrupting context.

kmap_atomic() was changed to dynamic slots long ago and commit 1ec9c5ddc17a
("include/linux/highmem.h: remove the second argument of k[un]map_atomic()")
removed the slot assignements, but the callers were not checked for now
redundant interrupt disabling.

Remove the conditional interrupt disable.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 drivers/ide/ide-taskfile.c | 8 --------
 1 file changed, 8 deletions(-)

diff --git a/drivers/ide/ide-taskfile.c b/drivers/ide/ide-taskfile.c
index abe0822dd429..49f9b4739779 100644
--- a/drivers/ide/ide-taskfile.c
+++ b/drivers/ide/ide-taskfile.c
@@ -237,7 +237,6 @@ void ide_pio_bytes(ide_drive_t *drive, struct ide_cmd *cmd,
 
 	while (len) {
 		unsigned nr_bytes = min(len, cursg->length - cmd->cursg_ofs);
-		int page_is_high;
 
 		page = sg_page(cursg);
 		offset = cursg->offset + cmd->cursg_ofs;
@@ -248,10 +247,6 @@ void ide_pio_bytes(ide_drive_t *drive, struct ide_cmd *cmd,
 
 		nr_bytes = min_t(unsigned, nr_bytes, (PAGE_SIZE - offset));
 
-		page_is_high = PageHighMem(page);
-		if (page_is_high)
-			local_irq_save(flags);
-
 		buf = kmap_atomic(page) + offset;
 
 		cmd->nleft -= nr_bytes;
@@ -270,9 +265,6 @@ void ide_pio_bytes(ide_drive_t *drive, struct ide_cmd *cmd,
 
 		kunmap_atomic(buf);
 
-		if (page_is_high)
-			local_irq_restore(flags);
-
 		len -= nr_bytes;
 	}
 }
-- 
2.17.0

[PATCH 4/4] ide: don't enable/disable interrupts in force threaded-IRQ mode

[Sebastian Andrzej Siewior]

The interrupts are enabled/disabled so the interrupt handler can run
with enabled interrupts while serving the interrupt and not lose other
interrupts especially the timer tick.
If the system runs with force-threaded interrupts then there is no need
to enable the interrupts.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 drivers/ide/ide-iops.c     | 13 +++++++++----
 drivers/ide/ide-taskfile.c |  2 +-
 kernel/irq/manage.c        |  1 +
 3 files changed, 11 insertions(+), 5 deletions(-)

diff --git a/drivers/ide/ide-iops.c b/drivers/ide/ide-iops.c
index 210a0887dd29..d55e9ebd5628 100644
--- a/drivers/ide/ide-iops.c
+++ b/drivers/ide/ide-iops.c
@@ -108,6 +108,7 @@ int __ide_wait_stat(ide_drive_t *drive, u8 good, u8 bad,
 	ide_hwif_t *hwif = drive->hwif;
 	const struct ide_tp_ops *tp_ops = hwif->tp_ops;
 	unsigned long flags;
+	bool irqs_threaded = force_irqthreads;
 	int i;
 	u8 stat;
 
@@ -115,8 +116,10 @@ int __ide_wait_stat(ide_drive_t *drive, u8 good, u8 bad,
 	stat = tp_ops->read_status(hwif);
 
 	if (stat & ATA_BUSY) {
-		local_save_flags(flags);
-		local_irq_enable_in_hardirq();
+		if (!irqs_threaded) {
+			local_save_flags(flags);
+			local_irq_enable_in_hardirq();
+		}
 		timeout += jiffies;
 		while ((stat = tp_ops->read_status(hwif)) & ATA_BUSY) {
 			if (time_after(jiffies, timeout)) {
@@ -129,12 +132,14 @@ int __ide_wait_stat(ide_drive_t *drive, u8 good, u8 bad,
 				if ((stat & ATA_BUSY) == 0)
 					break;
 
-				local_irq_restore(flags);
+				if (!irqs_threaded)
+					local_irq_restore(flags);
 				*rstat = stat;
 				return -EBUSY;
 			}
 		}
-		local_irq_restore(flags);
+		if (!irqs_threaded)
+			local_irq_restore(flags);
 	}
 	/*
 	 * Allow status to settle, then read it again.
diff --git a/drivers/ide/ide-taskfile.c b/drivers/ide/ide-taskfile.c
index 49f9b4739779..d42feeaebd36 100644
--- a/drivers/ide/ide-taskfile.c
+++ b/drivers/ide/ide-taskfile.c
@@ -405,7 +405,7 @@ static ide_startstop_t pre_task_out_intr(ide_drive_t *drive,
 		return startstop;
 	}
 
-	if ((drive->dev_flags & IDE_DFLAG_UNMASK) == 0)
+	if (!force_irqthreads && (drive->dev_flags & IDE_DFLAG_UNMASK) == 0)
 		local_irq_disable();
 
 	ide_set_handler(drive, &task_pio_intr, WAIT_WORSTCASE);
diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index e3336d904f64..4c2ef8084e32 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -24,6 +24,7 @@
 
 #ifdef CONFIG_IRQ_FORCED_THREADING
 __read_mostly bool force_irqthreads;
+EXPORT_SYMBOL_GPL(force_irqthreads);
 
 static int __init setup_forced_irqthreads(char *arg)
 {
-- 
2.17.0

Re: [PATCH 1/4] alim15x3: move irq-restore before pci_dev_put()

[David Miller]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date: Fri,  4 May 2018 16:24:43 +0200

> init_chipset_ali15x3() initializes the chipset during init with disabled
> interrupts. There is no need to keep the interrupts disabled during
> pci_dev_put().
> Move the irq-restore before pci_dev_put() is invoked.
> 
> Side note: The same init is performed in
> drivers/ata/pata_ali.c::ali_init_chipset() without disabled interrupts.
> It looks that the same hardware is supported in the ATA land. Would it
> make sense to remove this driver since it is supported in the other
> subsystem?
> 
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

Acked-by: David S. Miller <davem@davemloft.net>

Re: [PATCH 2/4] ide: Handle irq disabling consistently

[David Miller]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date: Fri,  4 May 2018 16:24:44 +0200

> ide_timer_expiry() disables interrupt at function entry when acquiring
> hwif->lock. Before disabling the device interrupt it unlocks hwif->lock,
> but interrupts stay disabled. After the call to disable_irq() interrupts
> are disabled again, which is a pointless exercise.
> 
> After the device irq handler has been invoked with interrupts disabled,
> hwif->lock is acquired again with spin_lock_irq() because the device irq
> handler might have reenabled interrupts. This is not documented and
> confusing for the casual reader.
> 
> Remove the redundant local_irq_disable() and add a comment which explains
> why hwif->lock has to be reacquired with spin_lock_irq().
> 
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

Acked-by: David S. Miller <davem@davemloft.net>

Re: [PATCH 3/4] ide: don't disable interrupts during kmap_atomic()

[David Miller]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date: Fri,  4 May 2018 16:24:45 +0200

> ide_pio_bytes() disables interrupts around kmap_atomic(). This is a
> leftover from the old kmap_atomic() implementation which relied on fixed
> mapping slots, so the caller had to make sure that the same slot could not
> be reused from an interrupting context.
> 
> kmap_atomic() was changed to dynamic slots long ago and commit 1ec9c5ddc17a
> ("include/linux/highmem.h: remove the second argument of k[un]map_atomic()")
> removed the slot assignements, but the callers were not checked for now
> redundant interrupt disabling.
> 
> Remove the conditional interrupt disable.
> 
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

Acked-by: David S. Miller <davem@davemloft.net>

Re: [PATCH 4/4] ide: don't enable/disable interrupts in force threaded-IRQ mode

[David Miller]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date: Fri,  4 May 2018 16:24:46 +0200

> The interrupts are enabled/disabled so the interrupt handler can run
> with enabled interrupts while serving the interrupt and not lose other
> interrupts especially the timer tick.
> If the system runs with force-threaded interrupts then there is no need
> to enable the interrupts.
> 
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

Acked-by: David S. Miller <davem@davemloft.net>