On Tue, Jul 11, 2017 at 10:50 AM, Madalin-cristian Bucur <madalin.bu...@nxp.com> wrote:
> Hi Arnd, > > Thanks for looking into this, I've tested your fix, it seems to need more > work: > > [ 0.894968] platform: DMA map failed > [ 0.898627] platform: DMA map failed > [ 0.902288] platform: DMA map failed > [ 0.905947] platform: DMA map failed > [ 0.909606] platform: DMA map failed > [ 0.913265] platform: DMA map failed I see: the assignment ended up after the first use, so ->dev was still NULL here. This should fix the problem you saw here: diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c index f7b0b928cd53..988c0212ce7e 100644 --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c @@ -2650,6 +2650,8 @@ static int dpaa_eth_probe(struct platform_device *pdev) for (i = 0; i < DPAA_BPS_NUM; i++) { int err; + /* DMA operations are done on the platform-provided device */ + dpaa_bps[i]->dev = dev->parent; dpaa_bps[i] = dpaa_bp_alloc(dev); if (IS_ERR(dpaa_bps[i])) return PTR_ERR(dpaa_bps[i]); @@ -2657,8 +2659,6 @@ static int dpaa_eth_probe(struct platform_device *pdev) dpaa_bps[i]->raw_size = bpool_buffer_raw_size(i, DPAA_BPS_NUM); /* avoid runtime computations by keeping the usable size here */ dpaa_bps[i]->size = dpaa_bp_size(dpaa_bps[i]->raw_size); - /* DMA operations are done on the platform-provided device */ - dpaa_bps[i]->dev = dev->parent; err = dpaa_bp_alloc_pool(dpaa_bps[i]); if (err < 0) { > @@ -2806,7 +2799,6 @@ static int dpaa_eth_probe(struct platform_device *pdev) > dpaa_bps_free(priv); > bp_create_failed: > fq_probe_failed: > -dev_mask_failed: > mac_probe_failed: > dev_set_drvdata(dev, NULL); > free_netdev(net_dev); > > I'll try to address your concern about performing the DMA mapping on a > different > device than the one set up by the platform code. Thanks! Arnd