mmc: fix incorrect interpretation of card type bits

From: Adrian Hunter <[email protected]>
Date: Fri, 19 Mar 2010 08:06:54 +0200
Subject: [PATCH] mmc: fix incorrect interpretation of card type bits

In the extended CSD register the CARD_TYPE is an 8-bit value of which the upper 6 bits were reserved in JEDEC specifications prior to version 4.4. In version 4.4 two of the reserved bits were designated for identifying support for the newly added High-Speed Dual Data Rate. Unfortunately the mmc_read_ext_csd() function required that the reserved bits be zero instead of ignoring them as it should.

This patch makes mmc_read_ext_csd() ignore the CARD_TYPE bits that are reserved or not yet supported. It also stops the function jumping to the end as though an error occurred, when it is only warns that the CARD_TYPE bits (that it does interpret) are invalid.

---

drivers/mmc/core/mmc.c  |    3 +--
include/linux/mmc/mmc.h |    1 +
2 files changed, 2 insertions(+), 2 deletions(-)

index 0eac6c8..e041c00 100644
@@ -225,7 +225,7 @@ static int mmc_read_ext_csd(struct mmc_card *card)

			mmc_card_set_blockaddr(card);
	}

- switch (ext_csd[EXT_CSD_CARD_TYPE]) {

+	switch (ext_csd[EXT_CSD_CARD_TYPE] & EXT_CSD_CARD_TYPE_MASK) {
	case EXT_CSD_CARD_TYPE_52 | EXT_CSD_CARD_TYPE_26:
		card->ext_csd.hs_max_dtr = 52000000;
		break;

@@ -237,7 +237,6 @@ static int mmc_read_ext_csd(struct mmc_card *card)
printk(KERN_WARNING "%s: card is mmc v4 but doesn't "
"support any high-speed modes.\n", mmc_hostname(card->host));

-		goto out;
	}

	if (card->ext_csd.rev >= 3) {

index c02c8db..8a49cbf 100644
@@ -268,6 +268,7 @@ struct _mmc_csd {
 

#define EXT_CSD_CARD_TYPE_26 (1<<0) /* Card can run at 26MHz */
#define EXT_CSD_CARD_TYPE_52 (1<<1) /* Card can run at 52MHz */
+#define EXT_CSD_CARD_TYPE_MASK 0x3 /* Mask out reserved and DDR bits */

#define EXT_CSD_BUS_WIDTH_1 0 /* Card is in 1 bit mode */
#define EXT_CSD_BUS_WIDTH_4 1 /* Card is in 4 bit mode */

1.6.0.4

I need to whine about the changelog again.

I'm guessing that the effect here is that if someone puts one of these new cards into an old kernel, this card will be incorrectly handled. Perhaps in ways which render the whole thing useless.

If that was a good guess then this patch would be needed in earlier kernels too.

But it was only a guess, and I'd prefer not to have to guess. So please do remember to explain things like this in the changelog?

This is for eMMC so that will not happen.

OK. "Backport not necessary."