Re: [patch] fix sdmmc bug affecting strictly sector-addressable eMMC chips
This message is in response to this previous discussion:
http://marc.info/?l=openbsd-tech=144074914604130
>> Since sector mode is supported by the host software, shouldn't we
>> always set the MMC_OCR_SECTOR_MODE bit and check the returned OCR
>> value to see if the card also supports sector mode addressing?
>>
>> Related thread: http://marc.info/?l=linux-mmc=129419687808618
> That is a good point. I'll fix that.
Now, instead of waiting for 50 failures (what was I thinking?) we simply
provide the MMC_OCR_SECTOR_MODE bit in CMD1's argument (given the card
is not an SD card) and set the SMF_SECTOR_MODE flag if the card's
response indicates it is indeed sector-addressed.
Ian
Index: sdmmc_mem.c
===
RCS file: /cvs/src/sys/dev/sdmmc/sdmmc_mem.c,v
retrieving revision 1.21
diff -u -p -r1.21 sdmmc_mem.c
--- sdmmc_mem.c 22 Apr 2015 04:02:06 - 1.21
+++ sdmmc_mem.c 9 Sep 2015 23:17:58 -
@@ -525,11 +525,14 @@ sdmmc_mem_send_op_cond(struct sdmmc_soft
cmd.c_opcode = SD_APP_OP_COND;
error = sdmmc_app_command(sc, );
} else {
+ cmd.c_arg |= MMC_OCR_SECTOR_MODE;
cmd.c_opcode = MMC_SEND_OP_COND;
error = sdmmc_mmc_command(sc, );
}
if (error != 0)
break;
+ if (ISSET(MMC_R3(cmd.c_resp), MMC_OCR_SECTOR_MODE))
+ sc->sc_flags |= SMF_SECTOR_MODE;
if (ISSET(MMC_R3(cmd.c_resp), MMC_OCR_MEM_READY) ||
ocr == 0)
break;
Index: sdmmcreg.h
===
RCS file: /cvs/src/sys/dev/sdmmc/sdmmcreg.h,v
retrieving revision 1.6
diff -u -p -r1.6 sdmmcreg.h
--- sdmmcreg.h 23 Sep 2014 12:08:13 - 1.6
+++ sdmmcreg.h 9 Sep 2015 23:17:58 -
@@ -48,6 +48,7 @@
/* OCR bits */
#define MMC_OCR_MEM_READY (1<<31) /* memory power-up status bit */
+#define MMC_OCR_SECTOR_MODE(1<<30) /* Provided in argument
following CMD1 */
#define MMC_OCR_3_5V_3_6V (1<<23)
#define MMC_OCR_3_4V_3_5V (1<<22)
#define MMC_OCR_3_3V_3_4V (1<<21)
Index: sdmmcvar.h
===
RCS file: /cvs/src/sys/dev/sdmmc/sdmmcvar.h,v
retrieving revision 1.22
diff -u -p -r1.22 sdmmcvar.h
--- sdmmcvar.h 12 Sep 2013 11:54:04 - 1.22
+++ sdmmcvar.h 9 Sep 2015 23:17:58 -
@@ -164,6 +164,7 @@ struct sdmmc_softc {
#define SMF_CARD_ATTACHED 0x0020 /* card driver(s) attached */
#defineSMF_STOP_AFTER_MULTIPLE 0x0040 /* send a stop after a multiple
cmd */
#define SMF_CONFIG_PENDING 0x0080 /* config_pending_incr() called */
+#define SMF_SECTOR_MODE 0x0100 /* card is sector-addressed */
uint32_t sc_caps; /* host capability */
#define SMC_CAPS_AUTO_STOP 0x0001 /* send CMD12 automagically by host */
Re: [patch] fix sdmmc bug affecting strictly sector-addressable eMMC chips
On Fri, Aug 28, 2015 at 02:13:44AM -0400, kremlin wrote: sdmmc gives up configuring the card after 100 failed CMD1 commands. My patch starts asserting the MMC_OCR_SECTOR_MODE bit after 50 failed attempts. This fixes the bug on my BBB with Micron chip: Nice work! I wonder why you chose to be so conservative (or arbitrary) about the after 50 failed attempts condition. :) Since sector mode is supported by the host software, shouldn't we always set the MMC_OCR_SECTOR_MODE bit and check the returned OCR value to see if the card also supports sector mode addressing? Related thread: http://marc.info/?l=linux-mmcm=129419687808618 Cheers, Uwe
[patch] fix sdmmc bug affecting strictly sector-addressable eMMC chips
Hello,
This patch fixes a bug affecting certain 'BeagleBone Black' devices
which prevents the onboard eMMC chip from initializing. It has been
discussed previously but was misattributed to manufacturers bumping its
size from 2GB - 4GB:
http://marc.info/?l=openbsd-techm=141122894126274w=2
Some Beaglebones (of any revision) are manufactured with Kingston eMMC
chips while others are manufactured with Micron eMMC chips. The Micron
ones fail during the card identification stage as the sdmmc driver
currently does not assert its ability to handle sector-addressed cards.
The Kingston ones don't.
Quoting the JEDEC eMMC standard, page 44:
If there is no indication by a host to a memory that the host is
capable of handling sector type of addressing the higher than 2GB of
density of memory will change its state to Inactive (similarly to a sit-
uation in which there is no common voltage range to work with)
Watch this occur with -DSDHC_DBUG -DSDMMC_DEBUG:
ommmc1: start cmd 1 arg=0x6 data=0x0 dlen=0 flags=0x1030
proc=sdmmc1
ommmc1: wait_state 1 0 1f7(state=1f7_CLEV)
ommmc1: cmd=0x102 blksize=0 blkcount=0
ommmc1: interrupt status=1_CC
ommmc1: intr status 0x1 error 0
resp[0] 0x00ff8080
ommmc1: cmd 1 done (flags=0x1030 error=0)
sdmmc1: can't send memory OCR
ommmc1: software reset reg=0x100
sdmmc1: can't enable card
sdmmc gives up configuring the card after 100 failed CMD1 commands. My
patch starts asserting the MMC_OCR_SECTOR_MODE bit after 50 failed
attempts. This fixes the bug on my BBB with Micron chip:
ommmc1: start cmd 1 arg=0x4006 data=0x0 dlen=0 flags=0x1030
proc=sdmmc1
ommmc1: wait_state 1 0 1f7(state=1f7_CLEV)
ommmc1: cmd=0x102 blksize=0 blkcount=0
ommmc1: interrupt status=1_CC
ommmc1: intr status 0x1 error 0
resp[0] 0xc0ff8080
ommmc1: cmd 1 done (flags=0x1030 error=0)
ommmc1: start cmd 2 arg=0 data=0x0 dlen=0 flags=0x1630 proc=sdmmc1
The card goes on to be interpreted correctly in the SCSI midlayer:
scsibus1 at sdmmc1: 2 targets, initiator 0
sd1 at scsibus1 targ 1 lun 0: SD/MMC, Drive #01, SCSI2 0/direct fixed
sd1: 3744MB, 512 bytes/sector, 7667712 sectors
I have tried mounting, reading, writing to it and everything seems to
work. The sector size is already read from the CSD register considered
during accesses as to prevent misalign errors so no additional work is
needed to support sector mode.
Ian
Index: sdmmc_mem.c
===
RCS file: /cvs/src/sys/dev/sdmmc/sdmmc_mem.c,v
retrieving revision 1.21
diff -u -p -r1.21 sdmmc_mem.c
--- sdmmc_mem.c 22 Apr 2015 04:02:06 - 1.21
+++ sdmmc_mem.c 28 Aug 2015 05:32:37 -
@@ -521,6 +521,10 @@ sdmmc_mem_send_op_cond(struct sdmmc_soft
cmd.c_arg = ocr;
cmd.c_flags = SCF_CMD_BCR | SCF_RSP_R3;
+ /* Set sector mode bit after 50 failures*/
+ if(i 50) {
+ cmd.c_arg |= MMC_OCR_SECTOR_MODE;
+ }
if (ISSET(sc-sc_flags, SMF_SD_MODE)) {
cmd.c_opcode = SD_APP_OP_COND;
error = sdmmc_app_command(sc, cmd);
Index: sdmmcreg.h
===
RCS file: /cvs/src/sys/dev/sdmmc/sdmmcreg.h,v
retrieving revision 1.6
diff -u -p -r1.6 sdmmcreg.h
--- sdmmcreg.h 23 Sep 2014 12:08:13 - 1.6
+++ sdmmcreg.h 28 Aug 2015 05:32:37 -
@@ -48,6 +48,7 @@
/* OCR bits */
#define MMC_OCR_MEM_READY (131) /* memory power-up status bit */
+#define MMC_OCR_SECTOR_MODE(130) /* Provided in argument
following CMD1 */
#define MMC_OCR_3_5V_3_6V (123)
#define MMC_OCR_3_4V_3_5V (122)
#define MMC_OCR_3_3V_3_4V (121)
Re: [patch] fix sdmmc bug affecting strictly sector-addressable eMMC chips
Nice work! I wonder why you chose to be so conservative (or arbitrary) about the after 50 failed attempts condition. :) My reasoning is that after 50 failed CMD1s, either two things are the case: The card will only initialize with the sector bit set, or it is a high capacity device taking a long time to become ready. In either of these cases, the card will be block/sector addressed anyway. Since sector mode is supported by the host software, shouldn't we always set the MMC_OCR_SECTOR_MODE bit and check the returned OCR value to see if the card also supports sector mode addressing? Related thread: http://marc.info/?l=linux-mmcm=129419687808618 That is a good point. I'll fix that.
