At present we add driver-model MMC devices in the order we find them. The 'alias' order is not honoured.
It is difficult to fix this for the case where we have holes in the sequence. But for the common case where the devices are numbered from 0 without any gaps, we can add the devices to the internal data structures in this order.
Signed-off-by: Simon Glass sjg@chromium.org ---
drivers/mmc/mmc.c | 20 +++++++++++++++----- 1 file changed, 15 insertions(+), 5 deletions(-)
diff --git a/drivers/mmc/mmc.c b/drivers/mmc/mmc.c index 3a34028..ae24b63 100644 --- a/drivers/mmc/mmc.c +++ b/drivers/mmc/mmc.c @@ -1772,18 +1772,28 @@ static int mmc_probe(bd_t *bis) #elif defined(CONFIG_DM_MMC) static int mmc_probe(bd_t *bis) { - int ret; + int ret, i; struct uclass *uc; - struct udevice *m; + struct udevice *dev;
ret = uclass_get(UCLASS_MMC, &uc); if (ret) return ret;
- uclass_foreach_dev(m, uc) { - ret = device_probe(m); + /* + * Try to add them in sequence order. Really with driver model we + * should allow holes, but the current MMC list does not allow that. + * So if we request 0, 1, 3 we will get 0, 1, 2. + */ + for (i = 0; ; i++) { + ret = uclass_get_device_by_seq(UCLASS_MMC, i, &dev); + if (ret == -ENODEV) + break; + } + uclass_foreach_dev(dev, uc) { + ret = device_probe(dev); if (ret) - printf("%s - probe failed: %d\n", m->name, ret); + printf("%s - probe failed: %d\n", dev->name, ret); }
return 0;