[U-Boot] [RFC PATCH 1/1] Read mmc device memory capacity from EXT_CSD if memory is addressed by sector

[Tomas Melin]

Hi,

We have successfully been using this patch to fix the problem mentioned in the
original message. Could this be picked up upstream?

BR,
Tomas

On Wed Feb 5 21:00:42 CET 2014, Frank Bormann wrote:
>Hello Everyone,
>
>I believe, there is a bug in the mmc driver code pertaining to how u-boot 
>detects memory size of an mmc device. However, I am not 100% sure, my solution 
>conforms to the JEDEC standard. So I am putting it up for discussion.
>
>Previously, sector count indicated by mmc devices in the EXT_CSD
>would only be used in u-boot if the size indicated is greater than
>2 GB. This seems to be incorrect. I am working with a 4 GB Micron
>eMMC device that after partition configuration and setting the
>user data area to enhanced mode has a remaining capacity of less
>than 2 GB for the user data area. JESD84-B50 explicitly states in
>6.2.4 that for these devices SEC_CNT from the EXT_CSD is still
>valid even if the memory size for that device has dropped below
>2 GB by the partition configuration applied. The access mode bits
>from the OCR register seem to provide a better way to decide
>whether to use the CSD-based C_SIZE & C_MULT or the EXT_CSD-based
>SEC_CNT information when determining the device's capacity.
>
>In particular, this fixes a bug where u-boot SPL would assign 0 to
>mmc->block_dev.lba later on in the mmc_startup() function and
>subsequently fail to load u-boot from that mmc due to the original
>C_SIZE & C_MULT code assigning a 4 TB size to mmc->capacity, that
>incorrect capacity never being overwritten by the SEC_CNT capacity
>calculation (due to its size being less than 2 GB) and then finally
>lldiv(mmc->capacity, mmc->read_bl_len) exceeding the 32-bit result
>data type of mmc->block_dev.lba.
>
>Signed-off-by: Frank Bormann <fbormann at yahoo.com>
>---
>  drivers/mmc/mmc.c |   10 +++++-----
>  include/mmc.h     |    1 +
>  2 files changed, 6 insertions(+), 5 deletions(-)
>
>diff --git a/drivers/mmc/mmc.c b/drivers/mmc/mmc.c
>index c6a1c23..c5d1efc 100644
>--- a/drivers/mmc/mmc.c
>+++ b/drivers/mmc/mmc.c
>@@ -935,19 +935,19 @@ static int mmc_startup(struct mmc *mmc)
>         if (!IS_SD(mmc) && (mmc->version >= MMC_VERSION_4)) {
>                 /* check  ext_csd version and capacity */
>                 err = mmc_send_ext_csd(mmc, ext_csd);
>-               if (!err && (ext_csd[EXT_CSD_REV] >= 2)) {
>+               if (!err && (ext_csd[EXT_CSD_REV] >= 2) &&
>+                       (mmc->ocr & OCR_ACCESS_MODE) == OCR_ACCESS_BY_SECTOR) {
>                         /*
>                          * According to the JEDEC Standard, the value of
>-                        * ext_csd's capacity is valid if the value is more
>-                        * than 2GB
>+                        * ext_csd's capacity is valid if the device addresses
>+                        * its memory by sector
>                          */
>                         capacity = ext_csd[EXT_CSD_SEC_CNT] << 0
>                                         | ext_csd[EXT_CSD_SEC_CNT + 1] << 8
>                                         | ext_csd[EXT_CSD_SEC_CNT + 2] << 16
>                                         | ext_csd[EXT_CSD_SEC_CNT + 3] << 24;
>                         capacity *= MMC_MAX_BLOCK_LEN;
>-                       if ((capacity >> 20) > 2 * 1024)
>-                               mmc->capacity_user = capacity;
>+                       mmc->capacity_user = capacity;
>                 }
>
>                 switch (ext_csd[EXT_CSD_REV]) {
>diff --git a/include/mmc.h b/include/mmc.h
>index e1060b9..816b580 100644
>--- a/include/mmc.h
>+++ b/include/mmc.h
>@@ -104,6 +104,7 @@
>  #define OCR_HCS                        0x40000000
>  #define OCR_VOLTAGE_MASK       0x007FFF80
>  #define OCR_ACCESS_MODE                0x60000000
>+#define OCR_ACCESS_BY_SECTOR   (1 << 30)
>
>  #define SECURE_ERASE           0x80000000