[#6285] MMC/SD can't work on bf537-stamp when either icache or dcache is off

Submitted By: Vivi Li

Open Date

2010-10-09 06:24:06

Priority:

Low Assignee:

Nobody

Category:

N/A Fixed In Release:

N/A

Found In Release:

2010R1 Status:

Open

Board:

N/A Processor:

BF537

Silicon Revision:

Resolution:

Assigned (Not Start)

Is the bug repeatable?:

N/A

Summary: MMC/SD can't work on bf537-stamp when either icache or dcache is off

Details:

MMC/SD can't work on bf537-stamp when either icache or dcache is off. This happens on both branch 2010r1 and trunk.

Last passed version is 2405; first failed version is 2411.

Bellow is the log:

U-Boot 2010.06-svn2467 (ADI-2010R1) (Oct 09 2010 - 14:29:38)

CPU: ADSP bf537-0.2 (Detected Rev: 0.2) (bypass boot)

Board: ADI BF537 stamp board

Support: http://blackfin.uclinux.org/

Clock: VCO: 400 MHz, Core: 200 MHz, System: 100 MHz

RAM: 64 MiB

Flash: 4 MiB

In: serial

Out: serial

Err: serial

KGDB: [on serial] ready

Net: bfin_mac

Hit any key to stop autoboot: 0

bfin> icache on

Instruction Cache is ON

bfin> dcache off

Data (writethrough) Cache is OFF

bfin> set mmc_cs 4; set mmc_hz 20000000

bfin> printenv mmc_cs mmc_hz

mmc_cs=4

mmc_hz=20000000

bfin> mmc init

using spi0.4 at 20000000 hz with mode 3

mmc1 is available

bfin> fatls mmc 0

** Unable to use mmc 0:1 for fatls **

bfin>

Follow-ups

--- Sonic Zhang 2010-10-12 06:10:55

Hi Mike,

This bug was introduced by your latest change to the SPI DMA code before code

freeze. Could you take a look?

--- Mike Frysinger 2010-10-12 21:10:14

i'm not worried about cache off behavior. does the issue only crop up when

caches are off ?

--- Vivi Li 2010-10-13 23:52:32

Yes, it happens only when either one of the caches is off.

--- Sonic Zhang 2010-10-18 22:32:47

Hi Mike,

I guess the conclusion is that we should not worry about spi device with cache

off? If so, Vivi may disable SPI cases with cache off.

--- Mike Frysinger 2011-05-07 18:09:57

i dont think any customer will ever run with caches turned off since the speed

loss is so high. it is useful to test with caches off as the increased latency

sometimes can detect race conditions in the code.

does this still fail with the newer mmc/spi driver ? i think this test was run

against the old one which no longer exists ...

Files

Changes

Commits

Dependencies

Duplicates

Associations