在 2015/8/17 5:10, Doug Anderson 写道:

Heiko,

On Fri, Aug 14, 2015 at 3:13 PM, Heiko Stübner [off-list ref] wrote:

quoted

Hi Shawn,

Am Freitag, 14. August 2015, 16:34:35 schrieb Shawn Lin:

quoted

DesignWare MMC Controller can supports two types of DMA
mode: external dma and internal dma. We get a RK312x platform
integrated dw_mmc and ARM pl330 dma controller. This patch add
edmac ops to support these platforms. I've tested it on RK312x
platform with edmac mode and RK3288 platform with idmac mode.

Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>

judging by your "from", I guess you're running this on some older Rockchip soc
without the idma? Because I tried testing this on a Radxa Rock, but only got
failures, from the start (failed to read card status register). In PIO mode
everything works again.


I guess I overlooked just some tiny detail, but to me the dma channel ids seem
correct after all. Maybe you have any hints what I'm doing wrong?

If I were a guessing man (which I'm not), I'd guess that perhaps
you're running into troubles with our friend the PL330.

There appear to be strange issues with the PL330 on Rockchip SoCs.  I
was only peripherally involved with them, but I know at least about
some of the patches in our tree, like:

Thanks, Doug, that's the root cause. PL330 on Rockchip Socs need your 
patches, and we might need another patch to limit pl330 burst_len to 
16(Hmm...seems another quirks for rockchip but not on your tree);

Hi Heiko,
I just get a Radxa Rock luckily and test my patch based on 
http://radxa.com/Rock/Linux_Mainline. Yes, it can't work.

If you test my patchset on Rockchip platform, pls apply pl330 patch from 
my tree based on kernel 4.2-RC3. AND, temporary hack dw_mmc to limit 
pl330 burst_len to 16.
This is a dmaengine or pl330 problem(I guest it should be upstreamed 
later?), but my patchset is for *generic* dw_mmc to support emdac, so 
other platforms should never need the hack of pl330.

pl330.patch is for pl330 changes
r3xxx.patch is for pl330 quirks

AND pls limit burst_len to 16 for BROKRN pl330 of rockchips.
static int dw_mci_edmac_start_dma(
...
+        u32 burst_limit = 0;
+        u32 mburst;
+        u32 idx, rx_wmark, tx_wmark;

...
         /* Match burst msize with external dma config */
         fifoth_val = mci_readl(host, FIFOTH);
-	cfg.dst_maxburst = mszs[(fifoth_val >> 28) & 0x7];
-	cfg.src_maxburst = cfg.dst_maxburst;
+	/* HACK for BROKEN pl330 */
+       mburst = mszs[(fifoth_val >> 28) & 0x7];
+       burst_limit = 16;
+       if (mburst > burst_limit) {
+               mburst = burst_limit;
+               idx = (ilog2(mburst) > 0) ? (ilog2(mburst) - 1) : 0;
+               rx_wmark = mszs[idx] - 1;
+               tx_wmark = (host->fifo_depth) / 2;
+               fifoth_val = SDMMC_SET_FIFOTH(idx, rx_wmark, tx_wmark);
+               mci_writel(host, FIFOTH, fifoth_val);
+       }
+	cfg.dst_maxburst = mburst;
+       cfg.src_maxburst = cfg.dst_maxburst;

https://chromium-review.googlesource.com/237607
FROMLIST: DMA: pl330: support burst mode for dev-to-mem and mem-to-dev transmit

https://chromium-review.googlesource.com/237393
CHROMIUM: dmaengine: pl330: support quirks for some broken

https://chromium-review.googlesource.com/237396
CHROMIUM: dmaengine: pl330: add quirk for broken no flushp

https://chromium-review.googlesource.com/237394
CHROMIUM: ARM: dts: rockchip: Add broken-no-flushp into rk3288.dtsi

https://chromium-review.googlesource.com/242063
CHROMIUM: ASoC: rockchip_i2s: modify DMA max burst to 1

-- 
Shawn Lin