On 2017年11月10日 05:21, Pierre-Louis Bossart wrote:

On 11/09/2017 12:17 AM, yan.wang@linux.intel.com wrote:

...

From: Yan Wang yan.wang@linux.intel.com

1. "size" should not be considered twice.
2. Change host_offset to uint32_t type.

Signed-off-by: Yan Wang yan.wang@linux.intel.com

src/include/reef/dma-trace.h | 2 +- src/lib/dma-trace.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/include/reef/dma-trace.h b/src/include/reef/dma-trace.h index 12d0956..641a4dc 100644 --- a/src/include/reef/dma-trace.h +++ b/src/include/reef/dma-trace.h @@ -56,7 +56,7 @@ struct dma_trace_data { struct dma_sg_config config; struct dma_trace_buf dmatb; struct dma_copy dc;

• int32_t host_offset;

• uint32_t host_offset; uint32_t host_size; struct work dmat_work; uint32_t enabled;

diff --git a/src/lib/dma-trace.c b/src/lib/dma-trace.c index 5885037..249e18b 100644 --- a/src/lib/dma-trace.c +++ b/src/lib/dma-trace.c @@ -90,7 +90,7 @@ static uint64_t trace_work(void *data, uint64_t delay) /* update host pointer and check for wrap */ d->host_offset += size;

• if (d->host_offset + size >= d->host_size)

• if (d->host_offset >= d->host_size) d->host_offset = 0;

this code pattern is present multiple times in the DMA handling code. I wonder if this is correct when the > condition is true? That'd mean some overrun happened or that some bytes have been lost. Shouldn't this be an exact == condition in there is a true wrap?

actually we should make it to something like this for sake of safe: d->host_offset -= d->host_size;

but considering that we have require the host_size to be multiple of period size, the '>' case won't happen, so here it is simplified to be d->host_offset = 0;

I agree that we should make it more robust.

Thanks, ~Keyon

...

  /* update local pointer and check for wrap */

---

Sound-open-firmware mailing list Sound-open-firmware@alsa-project.org http://mailman.alsa-project.org/mailman/listinfo/sound-open-firmware

Hi, Peierre and Keyon, Thanks for your comments. Because size = min(hsize, lsize), the updated host_offset should not > host_size. So "==" is enough. I will submit it again after modification. Thanks again.

Yan Wang   On Thu, 2017-11-09 at 20:31 -0600, Pierre-Louis Bossart wrote:

...

...

...

@@ -90,7 +90,7 @@ static uint64_t trace_work(void *data, uint64_t  delay)       /* update host pointer and check for wrap */       d->host_offset += size; -    if (d->host_offset + size >= d->host_size) +    if (d->host_offset >= d->host_size)           d->host_offset = 0;

this code pattern is present multiple times in the DMA handling code. I wonder if this is correct when the > condition is true? That'd mean  some overrun happened or that some bytes have been lost. Shouldn't this be an exact == condition in there is a true wrap?

actually we should make it to something like this for sake of safe:     d->host_offset -= d->host_size;

but considering that we have require the host_size to be multiple of  period size, the '>' case won't happen, so here it is simplified to be     d->host_offset = 0;

I agree that we should make it more robust.

Yes, my point exactly. Either the > condition happens and we have to  handle it, or it doesn't by construction due to other restrictions and  the test can be simplified.