[PATCH] net: davinci_emac: Replace devm_request_irq with request

Discussion:

[PATCH] net: davinci_emac: Replace devm_request_irq with request_irq

Christian Riesch

2014-03-05 10:25:28 UTC

In commit 6892b41d9701283085b655c6086fb57a5d63fa47

Author: Lad, Prabhakar <prabhakar.csengg at gmail.com>
Date: Tue Jun 25 21:24:51 2013 +0530
net: davinci: emac: Convert to devm_* api

the call of request_irq is replaced by devm_request_irq and the call
of free_irq is removed. But since interrupts are requested in
emac_dev_open, doing ifconfig up/down on the board requests the
interrupts again each time, causing devm_request_irq to fail.

This patch reverts said commit partially: It changes the driver back
to use request_irq instead of devm_request_irq, puts free_irq back in
place, but keeps the remaining changes of the original patch.

Reported-by: Jon Ringle <jon at ringle.org>
Signed-off-by: Christian Riesch <christian.riesch at omicron.at>
Cc: Lad, Prabhakar <prabhakar.csengg at gmail.com>
Cc: <stable at vger.kernel.org>
---
drivers/net/ethernet/ti/davinci_emac.c | 25 +++++++++++++++++++++----
1 file changed, 21 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/ti/davinci_emac.c b/drivers/net/ethernet/ti/davinci_emac.c
index cd9b164..2514304 100644
--- a/drivers/net/ethernet/ti/davinci_emac.c
+++ b/drivers/net/ethernet/ti/davinci_emac.c
@@ -1532,7 +1532,7 @@ static int emac_dev_open(struct net_device *ndev)
struct device *emac_dev = &ndev->dev;
u32 cnt;
struct resource *res;
- int ret;
+ int q, m, ret;
int i = 0;
int k = 0;
struct emac_priv *priv = netdev_priv(ndev);
@@ -1567,8 +1567,7 @@ static int emac_dev_open(struct net_device *ndev)

while ((res = platform_get_resource(priv->pdev, IORESOURCE_IRQ, k))) {
for (i = res->start; i <= res->end; i++) {
- if (devm_request_irq(&priv->pdev->dev, i, emac_irq,
- 0, ndev->name, ndev))
+ if (request_irq(i, emac_irq, 0, ndev->name, ndev))
goto rollback;
}
k++;
@@ -1641,7 +1640,15 @@ static int emac_dev_open(struct net_device *ndev)

rollback:

- dev_err(emac_dev, "DaVinci EMAC: devm_request_irq() failed");
+ dev_err(emac_dev, "DaVinci EMAC: request_irq() failed");
+
+ for (q = k; k >= 0; k--) {
+ for (m = i; m >= res->start; m--)
+ free_irq(m, ndev);
+ res = platform_get_resource(priv->pdev, IORESOURCE_IRQ, k-1);
+ m = res->end;
+ }
+
ret = -EBUSY;
err:
pm_runtime_put(&priv->pdev->dev);
@@ -1659,6 +1666,9 @@ err:
*/
static int emac_dev_stop(struct net_device *ndev)
{
+ struct resource *res;
+ int i = 0;
+ int irq_num;
struct emac_priv *priv = netdev_priv(ndev);
struct device *emac_dev = &ndev->dev;

@@ -1674,6 +1684,13 @@ static int emac_dev_stop(struct net_device *ndev)
if (priv->phydev)
phy_disconnect(priv->phydev);

+ /* Free IRQ */
+ while ((res = platform_get_resource(priv->pdev, IORESOURCE_IRQ, i))) {
+ for (irq_num = res->start; irq_num <= res->end; irq_num++)
+ free_irq(irq_num, priv->ndev);
+ i++;
+ }
+
if (netif_msg_drv(priv))
dev_notice(emac_dev, "DaVinci EMAC: %s stopped\n", ndev->name);

--
1.7.9.5

Prabhakar Lad

2014-03-06 05:38:18 UTC

Permalink

Hi Christian,

Thanks for the patch.

On Wed, Mar 5, 2014 at 3:55 PM, Christian Riesch

Post by Christian Riesch
In commit 6892b41d9701283085b655c6086fb57a5d63fa47
Author: Lad, Prabhakar <prabhakar.csengg at gmail.com>
Date: Tue Jun 25 21:24:51 2013 +0530
net: davinci: emac: Convert to devm_* api
the call of request_irq is replaced by devm_request_irq and the call
of free_irq is removed. But since interrupts are requested in
emac_dev_open, doing ifconfig up/down on the board requests the
interrupts again each time, causing devm_request_irq to fail.
This patch reverts said commit partially: It changes the driver back
to use request_irq instead of devm_request_irq, puts free_irq back in
place, but keeps the remaining changes of the original patch.
Reported-by: Jon Ringle <jon at ringle.org>
Signed-off-by: Christian Riesch <christian.riesch at omicron.at>
Cc: Lad, Prabhakar <prabhakar.csengg at gmail.com>
Cc: <stable at vger.kernel.org>

Acked-by: Lad, Prabhakar <prabhakar.csengg at gmail.com>

Regards,
--Prabhakar lad

Mugunthan V N

2014-03-06 08:27:58 UTC

Permalink

Instead of moving back to request irq. can you move the devm interrupt
request code from open to probe so that the issue is fixed, IRQ will be
requested on module insert and IRQ will be free in module remove and
there will not be any failures in devm request irq while repeating
ifup/ifdown and also during module insert/remove.

Regards
Mugunthan V N

Christian Riesch

2014-03-06 08:51:43 UTC

Permalink

Post by Mugunthan V N

This is exactly what I tried first, please have a look at my first patch
and the discussion [1]. Florian Fainelli asked me to revert Prabhakar's
patch instead to bring back the "usual situation", "just in case some
uncontrolled interrupt bit triggers an interrupt while the interface is
down for instance..."

Florian Fainelli, Mugunthan VN, which solution is better?

Regards, Christian

[1] http://marc.info/?t=139394310200001&r=1&w=2

Post by Mugunthan V N
Regards
Mugunthan V N
_______________________________________________
Davinci-linux-open-source mailing list
Davinci-linux-open-source at linux.davincidsp.com
http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source

Florian Fainelli

2014-03-06 17:16:05 UTC

Permalink

Post by Mugunthan V N

This is exactly what I tried first, please have a look at my first patch and
the discussion [1]. Florian Fainelli asked me to revert Prabhakar's patch
instead to bring back the "usual situation", "just in case some uncontrolled
interrupt bit triggers an interrupt while the interface is down for
instance..."
Florian Fainelli, Mugunthan VN, which solution is better?

I think the solution that does:

- request_irq() in ndo_open()
- free_irq() in ndo_stop()

is the safest option because it ensure that the various interrupt
lines are masked at the interrupt controller level, and not just at
Ethernet MAC level, and as such gives you more control over what the
hardware does. It also mimics what 99% of the Ethernet drivers do,
which will help identifying those common patterns programmatically.

--
Florian

Grygorii Strashko

2014-03-07 13:44:47 UTC

Permalink

Post by Florian Fainelli

Post by Mugunthan V N

This is exactly what I tried first, please have a look at my first patch and
the discussion [1]. Florian Fainelli asked me to revert Prabhakar's patch
instead to bring back the "usual situation", "just in case some uncontrolled
interrupt bit triggers an interrupt while the interface is down for
instance..."
Florian Fainelli, Mugunthan VN, which solution is better?

- request_irq() in ndo_open()
- free_irq() in ndo_stop()
is the safest option because it ensure that the various interrupt
lines are masked at the interrupt controller level, and not just at
Ethernet MAC level, and as such gives you more control over what the
hardware does. It also mimics what 99% of the Ethernet drivers do,
which will help identifying those common patterns programmatically.

Why are you not using enable_irq()/disable_irq()?
Also, IRQ_NOAUTOEN can be used
irq_set_status_flags(IRQ_NOAUTOEN);
before [devm_]request_irq().

The normal way is to request resources from .rpobe(), use them and free on .remove().

Actually, I've found at least one example - drivers/net/ethernet/apple/bmac.c

Regards,
- grygorii

David Miller

2014-03-06 21:57:36 UTC

Permalink

From: Christian Riesch <christian.riesch at omicron.at>
Date: Wed, 5 Mar 2014 11:25:28 +0100

Post by Christian Riesch
@@ -1641,7 +1640,15 @@ static int emac_dev_open(struct net_device *ndev)
- dev_err(emac_dev, "DaVinci EMAC: devm_request_irq() failed");
+ dev_err(emac_dev, "DaVinci EMAC: request_irq() failed");
+
+ for (q = k; k >= 0; k--) {
+ for (m = i; m >= res->start; m--)
+ free_irq(m, ndev);
+ res = platform_get_resource(priv->pdev, IORESOURCE_IRQ, k-1);
+ m = res->end;
+ }

This final assignment in the loop of 'm' is unused?

The inner for() loop always started with "m = i;"

Christian Riesch

2014-03-07 02:28:00 UTC

Permalink

From: Christian Riesch <christian.riesch at omicron.at <javascript:;>>
Date: Wed, 5 Mar 2014 11:25:28 +0100

k-1);

Post by Christian Riesch
+ m = res->end;
+ }

This final assignment in the loop of 'm' is unused?
The inner for() loop always started with "m = i;"

This should probably read

i = res->end;

I just took the old code, without thinking about it. Thanks for catching it.

Christian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://linux.davincidsp.com/pipermail/davinci-linux-open-source/attachments/20140307/f5648133/attachment.html>

Christian Riesch

2014-03-07 06:44:20 UTC

Permalink

[Sent again, sorry for the HTML mail before.]

Post by David Miller
From: Christian Riesch <christian.riesch at omicron.at>
Date: Wed, 5 Mar 2014 11:25:28 +0100

This final assignment in the loop of 'm' is unused?
The inner for() loop always started with "m = i;"

This should probably read?

i = res->end;?

I just took the old code without thinking about it. Thanks for catching it.

Christian
?

Christian Riesch

2014-03-07 11:42:28 UTC

Permalink

Hi,

Post by David Miller
From: Christian Riesch <christian.riesch at omicron.at>
Date: Wed, 5 Mar 2014 11:25:28 +0100

^^^^
So this should definitely read "i = res->end;".

Post by David Miller

Post by Christian Riesch
+ }

This final assignment in the loop of 'm' is unused?
The inner for() loop always started with "m = i;"

But it is more than just that. The entire rollback mechanism of
emac_dev_open is a mess. It just won't work.

1) Freeing the interrupts. Currently, the code tries to do a
platform_get_resource(priv->pdev, IORESOURCE_IRQ, -1) in its last
iteration. To make this work, the code should be something like

for (q = k; q >= 0; q--) {
res = platform_get_resource(priv->pdev, IORESOURCE_IRQ, k-1);
if (q != k)
i = res->end;

for (m = i; m >= res->start; m--)
free_irq(m, ndev);
}

2) Wrong order of err: and rollback: labels. Currently the function does
something like this:

pm_runtime_get
request irq
if requesting irqs fails, goto rollback
setup phy
if phy setup fails, goto err
return 0

rollback:
free irqs
err:
pm_runtime_put

So if PHY initialization fails, the IRQs are never freed. So rollback and
err must be exchanged, and some additonal code must be added to keep the
correct values of k and i for rollback.

3) RX DMA descriptors are not freed in case of an error: Right before
requesting the irqs, the function creates DMA descriptors for the RX
channel. These RX descriptors are never freed when we jump to either
rollback or err. And I think there is also no way to free them as long as
cpdma_ctlr_start() has not been called (which is done between requesting
irqs and phy setup in the code).

Problems 1+2 are easy to solve. Prabhakar, Mugunthan, any ideas for 3? I
think cpsw_ndo_start() in cpsw.c has the same problem. It will also try to
call cpdma_ctlr_stop in error handling before cpdma_ctlr_start was called
in some cases.

Regards, Christian

Christian Riesch

2014-03-07 14:07:37 UTC

Permalink

Signed-off-by: Christian Riesch <christian.riesch at omicron.at>
Cc: Prabhakar Lad <prabhakar.csengg at gmail.com>
Cc: Mugunthan V N <mugunthanvnm at ti.com>
Cc: Florian Fainelli <f.fainelli at gmail.com>
---

Hi again,

How about this solution for fixing the rollback of emac_dev_open()?
Especially the change in davinci_cpdma.c, would this break anything?

The patch applies on top of
[PATCH] net: davinci_emac: Replace devm_request_irq with request_irq

Regards,
Christian

drivers/net/ethernet/ti/davinci_cpdma.c | 4 +--
drivers/net/ethernet/ti/davinci_emac.c | 44 ++++++++++++++++++++-----------
2 files changed, 31 insertions(+), 17 deletions(-)

diff --git a/drivers/net/ethernet/ti/davinci_cpdma.c b/drivers/net/ethernet/ti/davinci_cpdma.c
index 364d0c7..88ef270 100644
--- a/drivers/net/ethernet/ti/davinci_cpdma.c
+++ b/drivers/net/ethernet/ti/davinci_cpdma.c
@@ -355,7 +355,7 @@ int cpdma_ctlr_stop(struct cpdma_ctlr *ctlr)
int i;

spin_lock_irqsave(&ctlr->lock, flags);
- if (ctlr->state != CPDMA_STATE_ACTIVE) {
+ if (ctlr->state == CPDMA_STATE_TEARDOWN) {
spin_unlock_irqrestore(&ctlr->lock, flags);
return -EINVAL;
}
@@ -891,7 +891,7 @@ int cpdma_chan_stop(struct cpdma_chan *chan)
unsigned timeout;

spin_lock_irqsave(&chan->lock, flags);
- if (chan->state != CPDMA_STATE_ACTIVE) {
+ if (chan->state == CPDMA_STATE_TEARDOWN) {
spin_unlock_irqrestore(&chan->lock, flags);
return -EINVAL;
}
diff --git a/drivers/net/ethernet/ti/davinci_emac.c b/drivers/net/ethernet/ti/davinci_emac.c
index 2514304..8f0e69c 100644
--- a/drivers/net/ethernet/ti/davinci_emac.c
+++ b/drivers/net/ethernet/ti/davinci_emac.c
@@ -1533,8 +1533,8 @@ static int emac_dev_open(struct net_device *ndev)
u32 cnt;
struct resource *res;
int q, m, ret;
+ int res_num = 0, irq_num = 0;
int i = 0;
- int k = 0;
struct emac_priv *priv = netdev_priv(ndev);

pm_runtime_get(&priv->pdev->dev);
@@ -1564,14 +1564,24 @@ static int emac_dev_open(struct net_device *ndev)
}

/* Request IRQ */
+ while ((res = platform_get_resource(priv->pdev, IORESOURCE_IRQ,
+ res_num))) {
+ for (irq_num = res->start; irq_num <= res->end; irq_num++) {
+ dev_err(emac_dev, "Request IRQ %d\n", irq_num);
+ if (request_irq(irq_num, emac_irq, 0, ndev->name,
+ ndev)) {
+ dev_err(emac_dev,
+ "DaVinci EMAC: request_irq() failed\n");
+ ret = -EBUSY;

- while ((res = platform_get_resource(priv->pdev, IORESOURCE_IRQ, k))) {
- for (i = res->start; i <= res->end; i++) {
- if (request_irq(i, emac_irq, 0, ndev->name, ndev))
goto rollback;
+ }
}
- k++;
+ res_num++;
}
+ /* prepare counters for rollback in case of an error */
+ res_num--;
+ irq_num--;

/* Start/Enable EMAC hardware */
emac_hw_enable(priv);
@@ -1638,19 +1648,23 @@ static int emac_dev_open(struct net_device *ndev)

return 0;

-rollback:
-
- dev_err(emac_dev, "DaVinci EMAC: request_irq() failed");
+err:
+ emac_int_disable(priv);
+ napi_disable(&priv->napi);

- for (q = k; k >= 0; k--) {
- for (m = i; m >= res->start; m--)
+rollback:
+ for (q = res_num; q >= 0; q--) {
+ res = platform_get_resource(priv->pdev, IORESOURCE_IRQ, q);
+ /* at the first iteration, irq_num is already set to the
+ * right value
+ */
+ if (q != res_num)
+ irq_num = res->end;
+
+ for (m = irq_num; m >= res->start; m--)
free_irq(m, ndev);
- res = platform_get_resource(priv->pdev, IORESOURCE_IRQ, k-1);
- m = res->end;
}
-
- ret = -EBUSY;
-err:
+ cpdma_ctlr_stop(priv->dma);
pm_runtime_put(&priv->pdev->dev);
return ret;
}

--
1.7.9.5

Mugunthan V N

2014-03-07 14:45:00 UTC

Permalink

Post by Christian Riesch
Signed-off-by: Christian Riesch <christian.riesch at omicron.at>
Cc: Prabhakar Lad <prabhakar.csengg at gmail.com>
Cc: Mugunthan V N <mugunthanvnm at ti.com>
Cc: Florian Fainelli <f.fainelli at gmail.com>
---
Hi again,
How about this solution for fixing the rollback of emac_dev_open()?
Especially the change in davinci_cpdma.c, would this break anything?
The patch applies on top of
[PATCH] net: davinci_emac: Replace devm_request_irq with request_irq
Regards,
Christian
drivers/net/ethernet/ti/davinci_cpdma.c | 4 +--
drivers/net/ethernet/ti/davinci_emac.c | 44 ++++++++++++++++++++-----------
2 files changed, 31 insertions(+), 17 deletions(-)
diff --git a/drivers/net/ethernet/ti/davinci_cpdma.c b/drivers/net/ethernet/ti/davinci_cpdma.c
index 364d0c7..88ef270 100644
--- a/drivers/net/ethernet/ti/davinci_cpdma.c
+++ b/drivers/net/ethernet/ti/davinci_cpdma.c
@@ -355,7 +355,7 @@ int cpdma_ctlr_stop(struct cpdma_ctlr *ctlr)
int i;
spin_lock_irqsave(&ctlr->lock, flags);
- if (ctlr->state != CPDMA_STATE_ACTIVE) {
+ if (ctlr->state == CPDMA_STATE_TEARDOWN) {
spin_unlock_irqrestore(&ctlr->lock, flags);
return -EINVAL;
}
@@ -891,7 +891,7 @@ int cpdma_chan_stop(struct cpdma_chan *chan)
unsigned timeout;
spin_lock_irqsave(&chan->lock, flags);
- if (chan->state != CPDMA_STATE_ACTIVE) {
+ if (chan->state == CPDMA_STATE_TEARDOWN) {
spin_unlock_irqrestore(&chan->lock, flags);
return -EINVAL;
}

Even when in idle mode chan stop should return error.

Regards
Mugunthan V N

Christian Riesch

2014-03-10 07:16:14 UTC

Permalink

Post by Mugunthan V N

Post by Christian Riesch
Signed-off-by: Christian Riesch <christian.riesch at omicron.at>
Cc: Prabhakar Lad <prabhakar.csengg at gmail.com>
Cc: Mugunthan V N <mugunthanvnm at ti.com>
Cc: Florian Fainelli <f.fainelli at gmail.com>
---
Hi again,
How about this solution for fixing the rollback of emac_dev_open()?
Especially the change in davinci_cpdma.c, would this break anything?
The patch applies on top of
[PATCH] net: davinci_emac: Replace devm_request_irq with request_irq
Regards,
Christian
drivers/net/ethernet/ti/davinci_cpdma.c | 4 +--
drivers/net/ethernet/ti/davinci_emac.c | 44
++++++++++++++++++++----------- 2 files changed, 31 insertions(+), 17
deletions(-)
diff --git a/drivers/net/ethernet/ti/davinci_cpdma.c
b/drivers/net/ethernet/ti/davinci_cpdma.c index 364d0c7..88ef270 100644
--- a/drivers/net/ethernet/ti/davinci_cpdma.c
+++ b/drivers/net/ethernet/ti/davinci_cpdma.c
@@ -355,7 +355,7 @@ int cpdma_ctlr_stop(struct cpdma_ctlr *ctlr)
int i;
spin_lock_irqsave(&ctlr->lock, flags);
- if (ctlr->state != CPDMA_STATE_ACTIVE) {
+ if (ctlr->state == CPDMA_STATE_TEARDOWN) {
spin_unlock_irqrestore(&ctlr->lock, flags);
return -EINVAL;
}
@@ -891,7 +891,7 @@ int cpdma_chan_stop(struct cpdma_chan *chan)
unsigned timeout;
spin_lock_irqsave(&chan->lock, flags);
- if (chan->state != CPDMA_STATE_ACTIVE) {
+ if (chan->state == CPDMA_STATE_TEARDOWN) {
spin_unlock_irqrestore(&chan->lock, flags);
return -EINVAL;
}

Even when in idle mode chan stop should return error.

Can you please explain? I do not see why.

I must be able to call cpdma_ctlr_stop in idle mode to free the rx
descriptors in case ndo_open in davinci_emac.c fails. Any other ideas how
to solve the problem addressed in the rest of the patch?

Thanks, Christian

Prabhakar Lad

2014-03-13 18:17:50 UTC

Permalink

Hi Mugunthan,

On Mon, Mar 10, 2014 at 12:46 PM, Christian Riesch

Post by Christian Riesch

Post by Mugunthan V N

Post by Christian Riesch
Signed-off-by: Christian Riesch <christian.riesch at omicron.at>
Cc: Prabhakar Lad <prabhakar.csengg at gmail.com>
Cc: Mugunthan V N <mugunthanvnm at ti.com>
Cc: Florian Fainelli <f.fainelli at gmail.com>
---
Hi again,
How about this solution for fixing the rollback of emac_dev_open()?
Especially the change in davinci_cpdma.c, would this break anything?
The patch applies on top of
[PATCH] net: davinci_emac: Replace devm_request_irq with request_irq
Regards,
Christian
drivers/net/ethernet/ti/davinci_cpdma.c | 4 +--
drivers/net/ethernet/ti/davinci_emac.c | 44
++++++++++++++++++++----------- 2 files changed, 31 insertions(+), 17
deletions(-)
diff --git a/drivers/net/ethernet/ti/davinci_cpdma.c
b/drivers/net/ethernet/ti/davinci_cpdma.c index 364d0c7..88ef270 100644
--- a/drivers/net/ethernet/ti/davinci_cpdma.c
+++ b/drivers/net/ethernet/ti/davinci_cpdma.c
@@ -355,7 +355,7 @@ int cpdma_ctlr_stop(struct cpdma_ctlr *ctlr)
int i;
spin_lock_irqsave(&ctlr->lock, flags);
- if (ctlr->state != CPDMA_STATE_ACTIVE) {
+ if (ctlr->state == CPDMA_STATE_TEARDOWN) {
spin_unlock_irqrestore(&ctlr->lock, flags);
return -EINVAL;
}
@@ -891,7 +891,7 @@ int cpdma_chan_stop(struct cpdma_chan *chan)
unsigned timeout;
spin_lock_irqsave(&chan->lock, flags);
- if (chan->state != CPDMA_STATE_ACTIVE) {
+ if (chan->state == CPDMA_STATE_TEARDOWN) {
spin_unlock_irqrestore(&chan->lock, flags);
return -EINVAL;
}

Even when in idle mode chan stop should return error.

Can you please explain? I do not see why.
I must be able to call cpdma_ctlr_stop in idle mode to free the rx
descriptors in case ndo_open in davinci_emac.c fails. Any other ideas how to
solve the problem addressed in the rest of the patch?

Do you have any comments on this patch ? I have tested it on OMAP-L138 evm
and works fine. If you have none comments may be Christian can consolidate this
patch with earlier and repost it.

Thanks,
--Prabhakar Lad