On 18/05/17 16:14, Robin Murphy wrote:
Before making any DMA API calls, the ETR driver should really be setting its masks to ensure that DMA is possible. Especially since it can address more than the 32-bit default mask set by the AMBA bus code.Signed-off-by: Robin Murphy <robin.mur...@arm.com> --- drivers/hwtracing/coresight/coresight-tmc.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/drivers/hwtracing/coresight/coresight-tmc.c b/drivers/hwtracing/coresight/coresight-tmc.c index d8517d2a968c..864488793f09 100644 --- a/drivers/hwtracing/coresight/coresight-tmc.c +++ b/drivers/hwtracing/coresight/coresight-tmc.c @@ -362,6 +362,13 @@ static int tmc_probe(struct amba_device *adev, const struct amba_id *id) desc.type = CORESIGHT_DEV_TYPE_SINK; desc.subtype.sink_subtype = CORESIGHT_DEV_SUBTYPE_SINK_BUFFER; desc.ops = &tmc_etr_cs_ops; + /* + * ETR configuration uses a 40-bit AXI master in place of + * the embedded SRAM of ETB/ETF. + */ + ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(40)); + if (ret) + goto out; } else { desc.type = CORESIGHT_DEV_TYPE_LINKSINK; desc.subtype.link_subtype = CORESIGHT_DEV_SUBTYPE_LINK_FIFO;
Tested this patch on Juno, with the SMMU enabled for the etr. I was able to allocate 32M using (the CMA was only 16M). FWIW : Tested-by: Suzuki K Poulose <suzuki.poul...@arm.com>
