mmc: mxs-mmc: add mmc host driver for i.MX23/28

This adds the mmc host driver for Freescale MXS-based SoC i.MX23/28. The driver calls into mxs-dma via generic dmaengine api for both pio and data transfer.

Thanks Chris Ball for the indentation patch.

---
 arch/arm/mach-mxs/include/mach/mmc.h |   18 +
 drivers/mmc/host/Kconfig             |    9 +
 drivers/mmc/host/Makefile            |    1 +
 drivers/mmc/host/mxs-mmc.c           |  874 ++++++++++++++++++++++++++++++++++
 4 files changed, 902 insertions(+), 0 deletions(-)
 create mode 100644 arch/arm/mach-mxs/include/mach/mmc.h
 create mode 100644 drivers/mmc/host/mxs-mmc.c

new file mode 100644
index 0000000..211547a
@@ -0,0 +1,18 @@
+/*
+ * Copyright 2011 Freescale Semiconductor, Inc. All Rights Reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef __MACH_MXS_MMC_H__
+#define __MACH_MXS_MMC_H__
+
+struct mxs_mmc_platform_data {
+	int wp_gpio;	/* write protect pin */
+	unsigned int flags;
+#define SLOTF_4_BIT_CAPABLE	(1 << 0)
+#define SLOTF_8_BIT_CAPABLE	(1 << 1)
+};
+#endif /* __MACH_MXS_MMC_H__ */

index afe8c6f..47ef75d 100644
@@ -319,6 +319,15 @@ config MMC_MXC
 
 	  If unsure, say N.
 
+config MMC_MXS
+	tristate "Freescale MXS Multimedia Card Interface support"
+	depends on ARCH_MXS && MXS_DMA
+	help
+	  This selects the Freescale SSP MMC controller found on MXS based
+	  platforms like mx23/28.
+
+	  If unsure, say N.
+
 config MMC_TIFM_SD
 	tristate "TI Flash Media MMC/SD Interface support  (EXPERIMENTAL)"
 	depends on EXPERIMENTAL && PCI

index e834fb2..30aa686 100644
@@ -6,6 +6,7 @@ obj-$(CONFIG_MMC_ARMMMCI)	+= mmci.o
 obj-$(CONFIG_MMC_PXA)		+= pxamci.o
 obj-$(CONFIG_MMC_IMX)		+= imxmmc.o
 obj-$(CONFIG_MMC_MXC)		+= mxcmmc.o
+obj-$(CONFIG_MMC_MXS)		+= mxs-mmc.o
 obj-$(CONFIG_MMC_SDHCI)		+= sdhci.o
 obj-$(CONFIG_MMC_SDHCI_PCI)	+= sdhci-pci.o
 obj-$(CONFIG_MMC_SDHCI_PXA)	+= sdhci-pxa.o

1.7.1

Shawn,

What throughput rates do you get on a mx28-regulated slot serving a card at 50Mhz? This is poor here (55kb/s) while it is OK at 25Mhz. Could be a board issue though (or we can fix it later), rest works fine, thus:

Pengutronix e.K.                           | Wolfram Sang                |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20110221/8244f36f/attachment.sig>

Hi Wolfram,

To make sure we are on the same page, can you share me the test you ran? And I can give it a try on my board.

Regards,
Shawn

Please ignore the message which is the one I sent a few days ago and I thought it did not get through.

Regards,
Shawn

Hi Wolfram,

To make sure we are on the same test, can you share the test case you ran to get the result? So that I can run it here for the result on my board.

Regards,
Shawn

# mount -t vfat /dev/mmcblk0p1 /mnt
# dd if=/mnt/test.dat of=/tmp/a bs=1024
# umount /mnt
# ./dd --version
dd (coreutils) 8.5
...

'test.dat' was a 8MB file of random data on a 4GB SDHC card (extrememory) on a slot powered by the mx28. Operating the card at 50Mhz, dd will report ~55 kb/s. If I add 'rate /= 2;' to the beginning of set_clk_rate, so it will operate at 25Mhz, I have 8mb/s.

Regards,

Wolfram

Pengutronix e.K.                           | Wolfram Sang                |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20110222/067cb096/attachment.sig>

My busybox version of dd does not report the speed. But I tried to dd 8MB data and saw it's done in 1~2 seconds.

Also here is mmc host driver test performance related result on my board if you are interested.

mmc0: Test case 23. Best-case read performance...
mmc0: Transfer of 6630 sectors (3315 KiB) took 0.101999209 seconds (33280 kB/s, 32500 KiB/s)
mmc0: Result: OK
mmc0: Test case 24. Best-case write performance...
mmc0: Transfer of 6630 sectors (3315 KiB) took 0.388688317 seconds (8733 kB/s, 8 528 KiB/s)
mmc0: Result: OK
mmc0: Test case 25. Best-case read performance into scattered pages...
mmc0: Transfer of 416 sectors (208 KiB) took 0.006997933 seconds (30436 kB/s, 29 723 KiB/s)
mmc0: Result: OK
mmc0: Test case 26. Best-case write performance from scattered pages...
mmc0: Transfer of 416 sectors (208 KiB) took 0.020431605 seconds (10424 kB/s, 10 180 KiB/s)
mmc0: Result: OK
mmc0: Test case 27. Single read performance by transfer size...
mmc0: Transfer of 1 sectors (0.5 KiB) took 0.000906524 seconds (564 kB/s, 551 Ki B/s)
mmc0: Transfer of 2 sectors (1 KiB) took 0.000625190 seconds (1637 kB/s, 1599 Ki B/s)
mmc0: Transfer of 4 sectors (2 KiB) took 0.000687706 seconds (2978 kB/s, 2908 Ki B/s)
mmc0: Transfer of 8 sectors (4 KiB) took 0.000718959 seconds (5697 kB/s, 5563 Ki B/s)
mmc0: Transfer of 16 sectors (8 KiB) took 0.000812729 seconds (10079 kB/s, 9843 KiB/s)
mmc0: Transfer of 32 sectors (16 KiB) took 0.001094051 seconds (14975 kB/s, 1462 4 KiB/s)
mmc0: Transfer of 64 sectors (32 KiB) took 0.001625434 seconds (20159 kB/s, 1968 7 KiB/s)
mmc0: Transfer of 128 sectors (64 KiB) took 0.002594421 seconds (25260 kB/s, 246 68 KiB/s)
mmc0: Transfer of 256 sectors (128 KiB) took 0.004563644 seconds (28720 kB/s, 28 047 KiB/s)
mmc0: Transfer of 512 sectors (256 KiB) took 0.008439521 seconds (31061 kB/s, 30 333 KiB/s)
mmc0: Transfer of 1024 sectors (512 KiB) took 0.016253622 seconds (32256 kB/s, 3 1500 KiB/s)
mmc0: Transfer of 2048 sectors (1024 KiB) took 0.031886943 seconds (32884 kB/s, 32113 KiB/s)
mmc0: Transfer of 4096 sectors (2048 KiB) took 0.063176177 seconds (33195 kB/s, 32417 KiB/s)
mmc0: Transfer of 6630 sectors (3315 KiB) took 0.101735514 seconds (33366 kB/s, 32584 KiB/s)
mmc0: Result: OK
mmc0: Test case 28. Single write performance by transfer size...
mmc0: Transfer of 1 sectors (0.5 KiB) took 0.001781476 seconds (287 kB/s, 280 Ki B/s)
mmc0: Transfer of 2 sectors (1 KiB) took 0.002250249 seconds (455 kB/s, 444 KiB/
s)
mmc0: Transfer of 4 sectors (2 KiB) took 0.002312719 seconds (885 kB/s, 864 KiB/
s)
mmc0: Transfer of 8 sectors (4 KiB) took 0.003437771 seconds (1191 kB/s, 1163 Ki B/s)
mmc0: Transfer of 16 sectors (8 KiB) took 0.002718920 seconds (3012 kB/s, 2942 K iB/s)
mmc0: Transfer of 32 sectors (16 KiB) took 0.001500142 seconds (10921 kB/s, 1066 5 KiB/s)
mmc0: Transfer of 64 sectors (32 KiB) took 0.003875336 seconds (8455 kB/s, 8257 KiB/s)
mmc0: Transfer of 128 sectors (64 KiB) took 0.006250472 seconds (10484 kB/s, 102 39 KiB/s)
mmc0: Transfer of 256 sectors (128 KiB) took 0.011563288 seconds (11335 kB/s, 11 069 KiB/s)
mmc0: Transfer of 512 sectors (256 KiB) took 0.022187432 seconds (11814 kB/s, 11 538 KiB/s)
mmc0: Transfer of 1024 sectors (512 KiB) took 0.049155884 seconds (10665 kB/s, 1 0415 KiB/s)
mmc0: Transfer of 2048 sectors (1024 KiB) took 0.085375335 seconds (12281 kB/s, 11994 KiB/s)
mmc0: Transfer of 4096 sectors (2048 KiB) took 0.262563987 seconds (7987 kB/s, 7 800 KiB/s)
mmc0: Transfer of 6630 sectors (3315 KiB) took 0.284787038 seconds (11919 kB/s, 11640 KiB/s)
mmc0: Result: OK
mmc0: Test case 29. Single trim performance by transfer size...
mmc0: Result: UNSUPPORTED (by card)
mmc0: Test case 30. Consecutive read performance by transfer size...
mmc0: Transfer of 8192 x 1 sectors (8192 x 0.5 KiB) took 4.091141441 seconds (10 25 kB/s, 1001 KiB/s)
mmc0: Transfer of 4096 x 2 sectors (4096 x 1 KiB) took 2.490875000 seconds (1683
kB/s, 1644 KiB/s)
mmc0: Transfer of 2048 x 4 sectors (2048 x 2 KiB) took 1.316531250 seconds (3185
kB/s, 3111 KiB/s)
mmc0: Transfer of 1024 x 8 sectors (1024 x 4 KiB) took 0.727437500 seconds (5765
kB/s, 5630 KiB/s)
mmc0: Transfer of 512 x 16 sectors (512 x 8 KiB) took 0.437000000 seconds (9597 kB/s, 9372 KiB/s)
mmc0: Transfer of 256 x 32 sectors (256 x 16 KiB) took 0.283156250 seconds (1481 2 kB/s, 14465 KiB/s)
mmc0: Transfer of 128 x 64 sectors (128 x 32 KiB) took 0.205812500 seconds (2037 9 kB/s, 19901 KiB/s)
mmc0: Transfer of 64 x 128 sectors (64 x 64 KiB) took 0.165750000 seconds (25305
kB/s, 24711 KiB/s)
mmc0: Transfer of 32 x 256 sectors (32 x 128 KiB) took 0.145343750 seconds (2885 7 kB/s, 28181 KiB/s)
mmc0: Transfer of 16 x 512 sectors (16 x 256 KiB) took 0.134968750 seconds (3107 6 kB/s, 30347 KiB/s)
mmc0: Transfer of 8 x 1024 sectors (8 x 512 KiB) took 0.130125000 seconds (32232
kB/s, 31477 KiB/s)
mmc0: Transfer of 4 x 2048 sectors (4 x 1024 KiB) took 0.127531249 seconds (3288 8 kB/s, 32117 KiB/s)
mmc0: Transfer of 2 x 4096 sectors (2 x 2048 KiB) took 0.126343751 seconds (3319 7 kB/s, 32419 KiB/s)
mmc0: Transfer of 1 x 6630 sectors (1 x 3315 KiB) took 0.101687494 seconds (3338 2 kB/s, 32599 KiB/s)
mmc0: Result: OK
mmc0: Test case 31. Consecutive write performance by transfer size...
mmc0: Transfer of 8192 x 1 sectors (8192 x 0.5 KiB) took 14.309562500 seconds (2 93 kB/s, 286 KiB/s)
mmc0: Transfer of 4096 x 2 sectors (4096 x 1 KiB) took 9.871062499 seconds (424 kB/s, 414 KiB/s)
mmc0: Transfer of 2048 x 4 sectors (2048 x 2 KiB) took 4.814593750 seconds (871 kB/s, 850 KiB/s)
mmc0: Transfer of 1024 x 8 sectors (1024 x 4 KiB) took 2.296875000 seconds (1826
kB/s, 1783 KiB/s)
mmc0: Transfer of 512 x 16 sectors (512 x 8 KiB) took 1.072843750 seconds (3909 kB/s, 3817 KiB/s)
mmc0: Transfer of 256 x 32 sectors (256 x 16 KiB) took 0.596312500 seconds (7033
kB/s, 6868 KiB/s)
mmc0: Transfer of 128 x 64 sectors (128 x 32 KiB) took 0.455375000 seconds (9210
kB/s, 8994 KiB/s)
mmc0: Transfer of 64 x 128 sectors (64 x 64 KiB) took 0.402875000 seconds (10410
kB/s, 10166 KiB/s)
mmc0: Transfer of 32 x 256 sectors (32 x 128 KiB) took 0.375531250 seconds (1116 8 kB/s, 10907 KiB/s)
mmc0: Transfer of 16 x 512 sectors (16 x 256 KiB) took 0.362562500 seconds (1156 8 kB/s, 11297 KiB/s)
mmc0: Transfer of 8 x 1024 sectors (8 x 512 KiB) took 0.350500000 seconds (11966
kB/s, 11686 KiB/s)
mmc0: Transfer of 4 x 2048 sectors (4 x 1024 KiB) took 0.351593796 seconds (1192 9 kB/s, 11649 KiB/s)
mmc0: Transfer of 2 x 4096 sectors (2 x 2048 KiB) took 0.346812955 seconds (1209 3 kB/s, 11810 KiB/s)
mmc0: Transfer of 1 x 6630 sectors (1 x 3315 KiB) took 0.280838757 seconds (1208 7 kB/s, 11803 KiB/s)
mmc0: Result: OK
mmc0: Test case 32. Consecutive trim performance by transfer size...
mmc0: Result: UNSUPPORTED (by card)
mmc0: Tests completed.

Regards,
Shawn

OK, looks good. Let's merge the driver then and I will check my board.

Thanks!

Wolfram

Pengutronix e.K.                           | Wolfram Sang                |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20110222/1d31de3a/attachment.sig>

In general, you should try to rule out file system, garbage collection, partitioning and cache effects.

With coreutils dd, a good command is

dd if=/dev/zero of=/dev/mmcblk0 bs=64K count=64 seek=256 oflag=direct

* Always write multiples of 4 MB to avoid garbage collection
* smaller block sizes may be faster, so writing 64 times 64K may be better than writing 4 MB once, depending on the card
* don't write to the FAT area, so start writing at 16 MB into the device.
* use direct I/O to bypass the page cache.
* read from /dev/zero instead of a disk file to avoid measuring the wrong thing

Arnd

Yes, right, I am aware of that (well, most of it ;)). I simplified my testcase here, because I was sure by previous tests it was only dependent on the frequency requested by the card. Remember, I wasn't after maximum values but chasing a difference in read performance of two magnitudes.

Thanks,

Wolfram

Pengutronix e.K.                           | Wolfram Sang                |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20110222/e97edfa7/attachment.sig>

Hi Shawn,

Thanks very much, pushed to mmc-next for .39.

Chris Ball   <[email protected]>   <http://printf.net/>
One Laptop Per Child