On 24 August 2018 at 03:50, Masahiro Yamada
[off-list ref] wrote:

2018-08-23 19:43 GMT+09:00 Ulf Hansson [off-list ref]:

quoted

[...]

quoted

+static void uniphier_sd_external_dma_request(struct tmio_mmc_host *host,
+                                            struct tmio_mmc_data *pdata)
+{
+       struct uniphier_sd_priv *priv = uniphier_sd_priv(host);
+       struct dma_chan *chan;
+
+       chan = dma_request_chan(mmc_dev(host->mmc), "rx-tx");
+       if (IS_ERR(chan)) {
+               dev_warn(mmc_dev(host->mmc),
+                        "failed to request DMA channel. falling back to PIO\n");
+               return; /* just use PIO even for -EPROBE_DEFER */
+       }
+
+       /* this driver uses a single channel for both RX an TX */
+       priv->chan = chan;
+       host->chan_rx = chan;
+       host->chan_tx = chan;
+
+       tasklet_init(&host->dma_issue, uniphier_sd_external_dma_issue,
+                    (unsigned long)host);

Isn't it time to get rid of tasklets for tmio?

No worries, I don't require that to be changed prior $subject patch,
however using threaded IRQs is a justified modernization for the tmio
variants.

Yeah, I also thought this.

But, the irq handler is in the TMIO common code,
so it would affect Renesas code.

Yep.

tmio-mmc has lots to be improved.
Tasklet is not the only problem.

Sure.

Before investing more effort,
I want working code base for my platform.

Totally acceptable!

[...]

Kind regards
Uffe