[v2] gpu: host1x: Ignore clients initialization failure

Message ID 20180809231211.26182-1-digetx@gmail.com
State New
Headers show
Series
  • [v2] gpu: host1x: Ignore clients initialization failure
Related show

Commit Message

Dmitry Osipenko Aug. 9, 2018, 11:12 p.m.
From time to time new bugs are popping up, causing some host1x client to
fail its initialization. Currently a single clients initialization failure
causes whole host1x device registration to fail, as a result a single DRM
sub-device initialization failure makes whole DRM initialization to fail.
Let's ignore clients initialization failure, as a result display panel
lights up even if some DRM clients (say GR2D or VIC) fail to initialize.
Actually VIC fails if initramfs misses the firmware file, so with this
change a serial console isn't needed anymore to figure out why display
isn't working.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---

Changelog:

v2: Added WARN_ON() that should get more attention than just a error message.
    Made clients_lock to lock around the lists modification, that is in line
    with the rest of the code.

 drivers/gpu/host1x/bus.c | 20 ++++++++------------
 1 file changed, 8 insertions(+), 12 deletions(-)

Comments

Dmitry Osipenko Aug. 13, 2018, 5:48 p.m. | #1
On Friday, 10 August 2018 02:12:11 MSK Dmitry Osipenko wrote:
> From time to time new bugs are popping up, causing some host1x client to
> fail its initialization. Currently a single clients initialization failure
> causes whole host1x device registration to fail, as a result a single DRM
> sub-device initialization failure makes whole DRM initialization to fail.
> Let's ignore clients initialization failure, as a result display panel
> lights up even if some DRM clients (say GR2D or VIC) fail to initialize.
> Actually VIC fails if initramfs misses the firmware file, so with this
> change a serial console isn't needed anymore to figure out why display
> isn't working.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
> 
> Changelog:
> 
> v2: Added WARN_ON() that should get more attention than just a error
> message. Made clients_lock to lock around the lists modification, that is
> in line with the rest of the code.
> 
>  drivers/gpu/host1x/bus.c | 20 ++++++++------------
>  1 file changed, 8 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
> index 815bdb42e3f0..07bb6b6c1260 100644
> --- a/drivers/gpu/host1x/bus.c
> +++ b/drivers/gpu/host1x/bus.c
> @@ -199,19 +199,23 @@ static void host1x_subdev_unregister(struct
> host1x_device *device, */
>  int host1x_device_init(struct host1x_device *device)
>  {
> -	struct host1x_client *client;
> +	struct host1x_client *client, *cl;
>  	int err;
> 
>  	mutex_lock(&device->clients_lock);
> 
> -	list_for_each_entry(client, &device->clients, list) {
> +	list_for_each_entry_safe(client, cl, &device->clients, list) {
>  		if (client->ops && client->ops->init) {
>  			err = client->ops->init(client);
> -			if (err < 0) {
> +			if (WARN_ON(err < 0)) {
>  				dev_err(&device->dev,
>  					"failed to initialize %s: %d\n",
>  					dev_name(client->dev), err);
> -				goto teardown;
> +
> +				/* add the client to the list of idle clients */
> +				mutex_lock(&clients_lock);
> +				list_add_tail(&client->list, &clients);

This shall be list_move_tail(), v3 is coming.

> +				mutex_unlock(&clients_lock);
Dmitry Osipenko Aug. 16, 2018, 5:20 p.m. | #2
On Monday, 13 August 2018 20:48:25 MSK Dmitry Osipenko wrote:
> On Friday, 10 August 2018 02:12:11 MSK Dmitry Osipenko wrote:
> > From time to time new bugs are popping up, causing some host1x client to
> > fail its initialization. Currently a single clients initialization failure
> > causes whole host1x device registration to fail, as a result a single DRM
> > sub-device initialization failure makes whole DRM initialization to fail.
> > Let's ignore clients initialization failure, as a result display panel
> > lights up even if some DRM clients (say GR2D or VIC) fail to initialize.
> > Actually VIC fails if initramfs misses the firmware file, so with this
> > change a serial console isn't needed anymore to figure out why display
> > isn't working.
> > 
> > Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> > ---
> > 
> > Changelog:
> > 
> > v2: Added WARN_ON() that should get more attention than just a error
> > message. Made clients_lock to lock around the lists modification, that is
> > in line with the rest of the code.
> > 
> >  drivers/gpu/host1x/bus.c | 20 ++++++++------------
> >  1 file changed, 8 insertions(+), 12 deletions(-)
> > 
> > diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
> > index 815bdb42e3f0..07bb6b6c1260 100644
> > --- a/drivers/gpu/host1x/bus.c
> > +++ b/drivers/gpu/host1x/bus.c
> > @@ -199,19 +199,23 @@ static void host1x_subdev_unregister(struct
> > host1x_device *device, */
> > 
> >  int host1x_device_init(struct host1x_device *device)
> >  {
> > 
> > -	struct host1x_client *client;
> > +	struct host1x_client *client, *cl;
> > 
> >  	int err;
> >  	
> >  	mutex_lock(&device->clients_lock);
> > 
> > -	list_for_each_entry(client, &device->clients, list) {
> > +	list_for_each_entry_safe(client, cl, &device->clients, list) {
> > 
> >  		if (client->ops && client->ops->init) {
> >  		
> >  			err = client->ops->init(client);
> > 
> > -			if (err < 0) {
> > +			if (WARN_ON(err < 0)) {
> > 
> >  				dev_err(&device->dev,
> >  				
> >  					"failed to initialize %s: %d\n",
> >  					dev_name(client->dev), err);
> > 
> > -				goto teardown;
> > +
> > +				/* add the client to the list of idle clients */
> > +				mutex_lock(&clients_lock);
> > +				list_add_tail(&client->list, &clients);
> 
> This shall be list_move_tail(), v3 is coming.
> 
> > +				mutex_unlock(&clients_lock);

I just realized that DRM won't load if any of host1x drivers fail to probe. I 
think the behaviour should be consistent in a case of driver-probe and client-
init failures. It looks like changing drivers probe error handling will be a 
much bigger change, hence let's keep the old behaviour at least for now.

Patch

diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
index 815bdb42e3f0..07bb6b6c1260 100644
--- a/drivers/gpu/host1x/bus.c
+++ b/drivers/gpu/host1x/bus.c
@@ -199,19 +199,23 @@  static void host1x_subdev_unregister(struct host1x_device *device,
  */
 int host1x_device_init(struct host1x_device *device)
 {
-	struct host1x_client *client;
+	struct host1x_client *client, *cl;
 	int err;
 
 	mutex_lock(&device->clients_lock);
 
-	list_for_each_entry(client, &device->clients, list) {
+	list_for_each_entry_safe(client, cl, &device->clients, list) {
 		if (client->ops && client->ops->init) {
 			err = client->ops->init(client);
-			if (err < 0) {
+			if (WARN_ON(err < 0)) {
 				dev_err(&device->dev,
 					"failed to initialize %s: %d\n",
 					dev_name(client->dev), err);
-				goto teardown;
+
+				/* add the client to the list of idle clients */
+				mutex_lock(&clients_lock);
+				list_add_tail(&client->list, &clients);
+				mutex_unlock(&clients_lock);
 			}
 		}
 	}
@@ -219,14 +223,6 @@  int host1x_device_init(struct host1x_device *device)
 	mutex_unlock(&device->clients_lock);
 
 	return 0;
-
-teardown:
-	list_for_each_entry_continue_reverse(client, &device->clients, list)
-		if (client->ops->exit)
-			client->ops->exit(client);
-
-	mutex_unlock(&device->clients_lock);
-	return err;
 }
 EXPORT_SYMBOL(host1x_device_init);