mail archive of the barebox mailing list
 help / color / mirror / Atom feed
* [PATCH] net: designware: dma_sync_single_for_device buffer before passing to HW
@ 2024-07-24 15:23 Ahmad Fatoum
  2024-07-30  8:10 ` Sascha Hauer
  0 siblings, 1 reply; 2+ messages in thread
From: Ahmad Fatoum @ 2024-07-24 15:23 UTC (permalink / raw)
  To: barebox; +Cc: Ahmad Fatoum

For cache-incoherent DMA, dma_sync_single_for_cpu with DMA_FROM_DEVICE
invalidates the cache, so no stale lines in the cache hide what the
hardware had written to RAM.

On the same systems, dma_sync_single_for_device with DMA_FROM_DEVICE also
invalidates the cache, so no dirty lines from the CPU are evicted
clobbering what the hardware had written to RAM.

In rx_descs_init, the RX buffers need to be invalidated before transferring
ownership to the network controller, therefore we should use
dma_sync_single_for_device for clarity, even if it doesn't change behavior.

Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
---
 drivers/net/designware.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/designware.c b/drivers/net/designware.c
index 183feba2ea5a..66f28b429dea 100644
--- a/drivers/net/designware.c
+++ b/drivers/net/designware.c
@@ -156,7 +156,7 @@ static void rx_descs_init(struct eth_device *dev)
 		else
 			desc_p->dmamac_cntl |= DESC_RXCTRL_RXCHAIN;
 
-		dma_sync_single_for_cpu(dev->parent, desc_p->dmamac_addr,
+		dma_sync_single_for_device(dev->parent, desc_p->dmamac_addr,
 					CONFIG_ETH_BUFSIZE, DMA_FROM_DEVICE);
 		desc_p->txrx_status = DESC_RXSTS_OWNBYDMA;
 	}
-- 
2.39.2




^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH] net: designware: dma_sync_single_for_device buffer before passing to HW
  2024-07-24 15:23 [PATCH] net: designware: dma_sync_single_for_device buffer before passing to HW Ahmad Fatoum
@ 2024-07-30  8:10 ` Sascha Hauer
  0 siblings, 0 replies; 2+ messages in thread
From: Sascha Hauer @ 2024-07-30  8:10 UTC (permalink / raw)
  To: barebox, Ahmad Fatoum


On Wed, 24 Jul 2024 17:23:27 +0200, Ahmad Fatoum wrote:
> For cache-incoherent DMA, dma_sync_single_for_cpu with DMA_FROM_DEVICE
> invalidates the cache, so no stale lines in the cache hide what the
> hardware had written to RAM.
> 
> On the same systems, dma_sync_single_for_device with DMA_FROM_DEVICE also
> invalidates the cache, so no dirty lines from the CPU are evicted
> clobbering what the hardware had written to RAM.
> 
> [...]

Applied, thanks!

[1/1] net: designware: dma_sync_single_for_device buffer before passing to HW
      https://git.pengutronix.de/cgit/barebox/commit/?id=dcd743414b15 (link may not be stable)

Best regards,
-- 
Sascha Hauer <s.hauer@pengutronix.de>




^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2024-07-30  8:10 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-07-24 15:23 [PATCH] net: designware: dma_sync_single_for_device buffer before passing to HW Ahmad Fatoum
2024-07-30  8:10 ` Sascha Hauer

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox