Re: [PATCH V2 1/2] ARM: mmc: bcm281xx SDHCI driver

From: Christian Daudt
Date: Tue May 21 2013 - 04:22:34 EST

Hi Arnd,
Thanks for the review. See below for comments.

On 13-05-16 03:09 PM, Arnd Bergmann wrote:
On Friday 10 May 2013, Christian Daudt wrote:
+struct sdhci_bcm_kona_cfg {
+ unsigned int max_freq;
+ int is_8bit;
+ int irq;
+ int cd_gpio;
+ int wp_gpio;
+ int non_removable;
I see no use for this structure to be separate: a lot of the fields are
duplicated in the sdhci_host, or should just get merged into
ok. Will do
+struct sdhci_bcm_kona_dev {
+ struct sdhci_bcm_kona_cfg *cfg;
+ struct device *dev;
+ struct sdhci_host *host;
+ struct clk *peri_clk;
+ struct clk *sleep_clk;
The *dev and *host members in this structure are redundant, just
allocate it together with sdhci_host and use use container_of()
to get from the sdhci_host back it it.
+static void sdhci_bcm_kona_sd_init(struct sdhci_host *host)
+ unsigned int val;
+ /* enable the interrupt from the IP core */
+ val = sdhci_readl(host, KONA_SDHOST_COREIMR);
+ val |= KONA_SDHOST_IP;
+ sdhci_writel(host, val, KONA_SDHOST_COREIMR);
+ /* Enable the AHB clock gating module to the host */
+ val = sdhci_readl(host, KONA_SDHOST_CORECTRL);
+ val |= KONA_SDHOST_EN;
+ /*
+ * Back-to-Back register write needs a delay of 1ms at bootup (min 10uS)
+ * Back-to-Back writes to same register needs delay when SD bus clock
+ * is very low w.r.t AHB clock, mainly during boot-time and during card
+ * insert-removal.
+ */
+ mdelay(1);
+ sdhci_writel(host, val, KONA_SDHOST_CORECTRL);
Why not use msleep() instead of mdelay() here?
I don't think that there's any reason. will replace.

+static int sdhci_bcm_kona_sd_card_emulate(struct sdhci_host *host, int insert)
+ struct sdhci_pltfm_host *pltfm_priv = sdhci_priv(host);
+ struct sdhci_bcm_kona_dev *kona_dev = pltfm_priv->priv;
+ u32 val;
+ unsigned long flags;
+ /* this function can be called from various contexts including ISR */
+ spin_lock_irqsave(&host->lock, flags);
+ /* Ensure SD bus scanning to detect media change */
+ host->mmc->rescan_disable = 0;
+ /*
+ * Back-to-Back register write needs a delay of min 10uS.
+ * Back-to-Back writes to same register needs delay when SD bus clock
+ * is very low w.r.t AHB clock, mainly during boot-time and during card
+ * insert-removal.
+ * We keep 20uS
+ */
+ udelay(20);
+ val = sdhci_readl(host, KONA_SDHOST_CORESTAT);
Does the delay have to be done with interrupts disabled? That is not particularly

I hope the hardware designers have been appropriately punished for the creating
such crap.
I had some internal discussions on this one, and the code was originally written for non-threaded irqs. Now that it is only called as threaded_irq thread_fn, it is safe to replace the spinlock that includes the delay with a mutex instead.

+static void sdhci_bcm_kona_init_74_clocks(struct sdhci_host *host,
+ u8 power_mode)
+ if (power_mode == MMC_POWER_OFF)
+ return;
+ else
+ mdelay(10);
This requires at the minimum a comment about why the mdelay is needed.
Maybe we can change the set_ios function so we never need to call it
in atomic context.
I'll look into this one.
+static struct sdhci_bcm_kona_cfg * __init sdhci_bcm_kona_parse_dt(
+ struct platform_device *pdev)
+ struct sdhci_bcm_kona_cfg *cfg;
+ struct device_node *np = pdev->dev.of_node;
+ u32 temp;
fold this function into probe()
+ if (!np)
+ return NULL;
+ cfg = devm_kzalloc(&pdev->dev, sizeof(*cfg), GFP_KERNEL);
+ if (!cfg) {
+ dev_err(&pdev->dev, "Can't allocate platform cfg\n");
+ return NULL;
+ }
Not needed
what is not needed ?
+static int __init sdhci_bcm_kona_probe(struct platform_device *pdev)
+ const struct of_device_id *match;
constant, so not needed.
I'll remove it.
+ struct sdhci_bcm_kona_cfg *kona_cfg = NULL;
No need to initialize this.
+ const struct sdhci_pltfm_data *plat_data;
make it global.
why make this global ?

+ struct sdhci_bcm_kona_dev *kona_dev = NULL;
No need to initialize this.
+ kona_dev = devm_kzalloc(dev, sizeof(*kona_dev), GFP_KERNEL);
+ if (!kona_dev) {
+ dev_err(dev, "Can't allocate kona_dev\n");
+ ret = -ENOMEM;
+ goto err_pltfm_free;
+ }
It is rather silly to have the base sdhci code allocate extra
memory for the platform drivers but then require an extra allocation.
Better change the sdhci_pltfm_init function to let you pass the extra
allocation size.
ok, I'll look into this.
No person?

A collective effort :)


To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at