[U-Boot] [RFC PATCH 1/1] Read mmc device memory capacity from EXT_CSD if memory is addressed by sector
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@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
Hi Pantos, hi Tom,
I sent this a couple of months ago to the mailing list, never really received a response. We are testing 2014.04-rc3 right now and the issue is still there. Would you still consider bringing this fix in for the upcoming release?
This is for an eMMC chip with an initial memory size > 2GB whose memory size drops below 2GB when turning enhanced (pseudo-SLC) mode on for the user partition. u-boot would then fail memory size detection and assume memory size if zero. You'd see error messages like:
MMC: block number 0x1 exceeds max(0x0) MMC: block number 0x800 exceeds max(0x0) MMC: block number 0x900 exceeds max(0x0)
Thanks, Frank
On 05/02/14 03:00 PM, 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@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 _______________________________________________ U-Boot mailing list U-Boot@lists.denx.de http://lists.denx.de/mailman/listinfo/u-boot
Hi Frank,
On Apr 11, 2014, at 6:35 PM, Frank Bormann wrote:
Hi Pantos, hi Tom,
I sent this a couple of months ago to the mailing list, never really received a response. We are testing 2014.04-rc3 right now and the issue is still there. Would you still consider bringing this fix in for the upcoming release?
This is for an eMMC chip with an initial memory size > 2GB whose memory size drops below 2GB when turning enhanced (pseudo-SLC) mode on for the user partition. u-boot would then fail memory size detection and assume memory size if zero. You'd see error messages like:
MMC: block number 0x1 exceeds max(0x0) MMC: block number 0x800 exceeds max(0x0) MMC: block number 0x900 exceeds max(0x0)
Thanks, Frank
Your patch is corrupted; can you please resend and make sure it applies using git am?
panto@sles11esa:~/u-boot-mmc.git (master)$ git am -3 U-Boot-RFC-1-1-Read-mmc-device-memory-capacity-from-EXT_CSD-if-memory-is-addressed-by-sector.patch Applying: Read mmc device memory capacity from EXT_CSD if memory is addressed by sector fatal: corrupt patch at line 50 Repository lacks necessary blobs to fall back on 3-way merge. Cannot fall back to three-way merge. Patch failed at 0001 Read mmc device memory capacity from EXT_CSD if memory is addressed by sector The copy of the patch that failed is found in: /home/panto/ti/u-boots/u-boot-mmc.git/.git/rebase-apply/patch When you have resolved this problem, run "git am --resolved". If you prefer to skip this patch, run "git am --skip" instead. To restore the original branch and stop patching, run "git am --abort".
Regards
-- Pantelis
participants (2)
-
Frank Bormann -
Pantelis Antoniou