| Message ID | 09960ba7389058e026b73409018fe1ee61d6dc57.1427469791.git.nicolas.ferre@atmel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Nicolas, There is a little mistake in this patch, which will prevent the GEM from working properly: Le 27/03/2015 16:34, Nicolas Ferre a écrit : > When merging at91_ether and macb driver during 421d9df0628b (net/macb: merge > at91_ether driver into macb driver) the probe function has been split. The code > dealing with initialization of queues is now moved in macb_init() which needs > information computed in the parent macb_probe() function. > So, add the queue_mask information to the private structure and use it when > needed in macb_init(). > > Signed-off-by: Nicolas Ferre <nicolas.ferre@atmel.com> > Cc: Cyrille Pitchen <cyrille.pitchen@atmel.com> > Cc: Boris Brezillon <boris.brezillon@free-electrons.com> > --- > drivers/net/ethernet/cadence/macb.c | 9 ++++----- > drivers/net/ethernet/cadence/macb.h | 1 + > 2 files changed, 5 insertions(+), 5 deletions(-) > > diff --git a/drivers/net/ethernet/cadence/macb.c b/drivers/net/ethernet/cadence/macb.c > index a0a04b3638e6..b710768172d9 100644 > --- a/drivers/net/ethernet/cadence/macb.c > +++ b/drivers/net/ethernet/cadence/macb.c > @@ -2180,7 +2180,7 @@ static void macb_probe_queues(void __iomem *mem, > static int macb_init(struct platform_device *pdev) > { > struct net_device *dev = platform_get_drvdata(pdev); > - unsigned int hw_q, queue_mask, q, num_queues; > + unsigned int hw_q, q; > struct macb *bp = netdev_priv(dev); > struct macb_queue *queue; > int err; > @@ -2226,10 +2226,8 @@ static int macb_init(struct platform_device *pdev) > * register mapping but we don't want to test the queue index then > * compute the corresponding register offset at run time. > */ > - macb_probe_queues(bp->regs, &queue_mask, &num_queues); > - > - for (hw_q = 0, q = 0; hw_q < MACB_MAX_QUEUES; ++hw_q) { > - if (!(queue_mask & (1 << hw_q))) > + for (hw_q = 0, q = 0; hw_q < bp->num_queues; ++hw_q) { > + if (!(bp->queue_mask & (1 << hw_q))) > continue; The exit condition of the for loop must remain "hw_q < MACB_MAX_QUEUES" and should not be replaced by "hw_q < bp->num_queues". Indeed there can be only 2 queues, for instance queue0 and queue7. So if you change the exit conditon of the loop, queue7 won't be initialized and HRESP errors will occur. > > queue = &bp->queues[q]; > @@ -2715,6 +2713,7 @@ static int macb_probe(struct platform_device *pdev) > bp->dev = dev; > bp->regs = mem; > bp->num_queues = num_queues; > + bp->queue_mask = queue_mask; > spin_lock_init(&bp->lock); > > platform_set_drvdata(pdev, dev); > diff --git a/drivers/net/ethernet/cadence/macb.h b/drivers/net/ethernet/cadence/macb.h > index bc6e35c40822..0b6afac91bfe 100644 > --- a/drivers/net/ethernet/cadence/macb.h > +++ b/drivers/net/ethernet/cadence/macb.h > @@ -785,6 +785,7 @@ struct macb { > size_t rx_buffer_size; > > unsigned int num_queues; > + unsigned int queue_mask; > struct macb_queue queues[MACB_MAX_QUEUES]; > > spinlock_t lock; > Best Regards, Cyrille
Hi Nicolas, On Fri, 27 Mar 2015 16:34:09 +0100 Nicolas Ferre <nicolas.ferre@atmel.com> wrote: > When merging at91_ether and macb driver during 421d9df0628b (net/macb: merge > at91_ether driver into macb driver) the probe function has been split. The code > dealing with initialization of queues is now moved in macb_init() which needs > information computed in the parent macb_probe() function. > So, add the queue_mask information to the private structure and use it when > needed in macb_init(). Just to be sure: this is not a bug fix, right ? It just improves the initialization process by calling macb_probe_queues only once (in macb_probe). Anyway, once you've addressed Cyrille's comment, you can add my Acked-by: Boris Brezillon <boris.brezillon@free-electrons.com> > > Signed-off-by: Nicolas Ferre <nicolas.ferre@atmel.com> > Cc: Cyrille Pitchen <cyrille.pitchen@atmel.com> > Cc: Boris Brezillon <boris.brezillon@free-electrons.com> > --- > drivers/net/ethernet/cadence/macb.c | 9 ++++----- > drivers/net/ethernet/cadence/macb.h | 1 + > 2 files changed, 5 insertions(+), 5 deletions(-) > > diff --git a/drivers/net/ethernet/cadence/macb.c b/drivers/net/ethernet/cadence/macb.c > index a0a04b3638e6..b710768172d9 100644 > --- a/drivers/net/ethernet/cadence/macb.c > +++ b/drivers/net/ethernet/cadence/macb.c > @@ -2180,7 +2180,7 @@ static void macb_probe_queues(void __iomem *mem, > static int macb_init(struct platform_device *pdev) > { > struct net_device *dev = platform_get_drvdata(pdev); > - unsigned int hw_q, queue_mask, q, num_queues; > + unsigned int hw_q, q; > struct macb *bp = netdev_priv(dev); > struct macb_queue *queue; > int err; > @@ -2226,10 +2226,8 @@ static int macb_init(struct platform_device *pdev) > * register mapping but we don't want to test the queue index then > * compute the corresponding register offset at run time. > */ > - macb_probe_queues(bp->regs, &queue_mask, &num_queues); > - > - for (hw_q = 0, q = 0; hw_q < MACB_MAX_QUEUES; ++hw_q) { > - if (!(queue_mask & (1 << hw_q))) > + for (hw_q = 0, q = 0; hw_q < bp->num_queues; ++hw_q) { > + if (!(bp->queue_mask & (1 << hw_q))) > continue; > > queue = &bp->queues[q]; > @@ -2715,6 +2713,7 @@ static int macb_probe(struct platform_device *pdev) > bp->dev = dev; > bp->regs = mem; > bp->num_queues = num_queues; > + bp->queue_mask = queue_mask; > spin_lock_init(&bp->lock); > > platform_set_drvdata(pdev, dev); > diff --git a/drivers/net/ethernet/cadence/macb.h b/drivers/net/ethernet/cadence/macb.h > index bc6e35c40822..0b6afac91bfe 100644 > --- a/drivers/net/ethernet/cadence/macb.h > +++ b/drivers/net/ethernet/cadence/macb.h > @@ -785,6 +785,7 @@ struct macb { > size_t rx_buffer_size; > > unsigned int num_queues; > + unsigned int queue_mask; > struct macb_queue queues[MACB_MAX_QUEUES]; > > spinlock_t lock;
Le 27/03/2015 17:24, Cyrille Pitchen a écrit : > Hi Nicolas, > > There is a little mistake in this patch, which will prevent the GEM from > working properly: > > Le 27/03/2015 16:34, Nicolas Ferre a écrit : >> When merging at91_ether and macb driver during 421d9df0628b (net/macb: merge >> at91_ether driver into macb driver) the probe function has been split. The code >> dealing with initialization of queues is now moved in macb_init() which needs >> information computed in the parent macb_probe() function. >> So, add the queue_mask information to the private structure and use it when >> needed in macb_init(). >> >> Signed-off-by: Nicolas Ferre <nicolas.ferre@atmel.com> >> Cc: Cyrille Pitchen <cyrille.pitchen@atmel.com> >> Cc: Boris Brezillon <boris.brezillon@free-electrons.com> >> --- >> drivers/net/ethernet/cadence/macb.c | 9 ++++----- >> drivers/net/ethernet/cadence/macb.h | 1 + >> 2 files changed, 5 insertions(+), 5 deletions(-) >> >> diff --git a/drivers/net/ethernet/cadence/macb.c b/drivers/net/ethernet/cadence/macb.c >> index a0a04b3638e6..b710768172d9 100644 >> --- a/drivers/net/ethernet/cadence/macb.c >> +++ b/drivers/net/ethernet/cadence/macb.c >> @@ -2180,7 +2180,7 @@ static void macb_probe_queues(void __iomem *mem, >> static int macb_init(struct platform_device *pdev) >> { >> struct net_device *dev = platform_get_drvdata(pdev); >> - unsigned int hw_q, queue_mask, q, num_queues; >> + unsigned int hw_q, q; >> struct macb *bp = netdev_priv(dev); >> struct macb_queue *queue; >> int err; >> @@ -2226,10 +2226,8 @@ static int macb_init(struct platform_device *pdev) >> * register mapping but we don't want to test the queue index then >> * compute the corresponding register offset at run time. >> */ >> - macb_probe_queues(bp->regs, &queue_mask, &num_queues); >> - >> - for (hw_q = 0, q = 0; hw_q < MACB_MAX_QUEUES; ++hw_q) { >> - if (!(queue_mask & (1 << hw_q))) >> + for (hw_q = 0, q = 0; hw_q < bp->num_queues; ++hw_q) { >> + if (!(bp->queue_mask & (1 << hw_q))) >> continue; > > The exit condition of the for loop must remain "hw_q < MACB_MAX_QUEUES" and > should not be replaced by "hw_q < bp->num_queues". > Indeed there can be only 2 queues, for instance queue0 and queue7. So if you > change the exit conditon of the loop, queue7 won't be initialized and HRESP > errors will occur. Absolutely. Sorry to have misunderstood this condition. I'll re-spin a v2 series. Bye, >> >> queue = &bp->queues[q]; >> @@ -2715,6 +2713,7 @@ static int macb_probe(struct platform_device *pdev) >> bp->dev = dev; >> bp->regs = mem; >> bp->num_queues = num_queues; >> + bp->queue_mask = queue_mask; >> spin_lock_init(&bp->lock); >> >> platform_set_drvdata(pdev, dev); >> diff --git a/drivers/net/ethernet/cadence/macb.h b/drivers/net/ethernet/cadence/macb.h >> index bc6e35c40822..0b6afac91bfe 100644 >> --- a/drivers/net/ethernet/cadence/macb.h >> +++ b/drivers/net/ethernet/cadence/macb.h >> @@ -785,6 +785,7 @@ struct macb { >> size_t rx_buffer_size; >> >> unsigned int num_queues; >> + unsigned int queue_mask; >> struct macb_queue queues[MACB_MAX_QUEUES]; >> >> spinlock_t lock; >> > > Best Regards, > > Cyrille >
diff --git a/drivers/net/ethernet/cadence/macb.c b/drivers/net/ethernet/cadence/macb.c index a0a04b3638e6..b710768172d9 100644 --- a/drivers/net/ethernet/cadence/macb.c +++ b/drivers/net/ethernet/cadence/macb.c @@ -2180,7 +2180,7 @@ static void macb_probe_queues(void __iomem *mem, static int macb_init(struct platform_device *pdev) { struct net_device *dev = platform_get_drvdata(pdev); - unsigned int hw_q, queue_mask, q, num_queues; + unsigned int hw_q, q; struct macb *bp = netdev_priv(dev); struct macb_queue *queue; int err; @@ -2226,10 +2226,8 @@ static int macb_init(struct platform_device *pdev) * register mapping but we don't want to test the queue index then * compute the corresponding register offset at run time. */ - macb_probe_queues(bp->regs, &queue_mask, &num_queues); - - for (hw_q = 0, q = 0; hw_q < MACB_MAX_QUEUES; ++hw_q) { - if (!(queue_mask & (1 << hw_q))) + for (hw_q = 0, q = 0; hw_q < bp->num_queues; ++hw_q) { + if (!(bp->queue_mask & (1 << hw_q))) continue; queue = &bp->queues[q]; @@ -2715,6 +2713,7 @@ static int macb_probe(struct platform_device *pdev) bp->dev = dev; bp->regs = mem; bp->num_queues = num_queues; + bp->queue_mask = queue_mask; spin_lock_init(&bp->lock); platform_set_drvdata(pdev, dev); diff --git a/drivers/net/ethernet/cadence/macb.h b/drivers/net/ethernet/cadence/macb.h index bc6e35c40822..0b6afac91bfe 100644 --- a/drivers/net/ethernet/cadence/macb.h +++ b/drivers/net/ethernet/cadence/macb.h @@ -785,6 +785,7 @@ struct macb { size_t rx_buffer_size; unsigned int num_queues; + unsigned int queue_mask; struct macb_queue queues[MACB_MAX_QUEUES]; spinlock_t lock;
When merging at91_ether and macb driver during 421d9df0628b (net/macb: merge at91_ether driver into macb driver) the probe function has been split. The code dealing with initialization of queues is now moved in macb_init() which needs information computed in the parent macb_probe() function. So, add the queue_mask information to the private structure and use it when needed in macb_init(). Signed-off-by: Nicolas Ferre <nicolas.ferre@atmel.com> Cc: Cyrille Pitchen <cyrille.pitchen@atmel.com> Cc: Boris Brezillon <boris.brezillon@free-electrons.com> --- drivers/net/ethernet/cadence/macb.c | 9 ++++----- drivers/net/ethernet/cadence/macb.h | 1 + 2 files changed, 5 insertions(+), 5 deletions(-)