mtd: extend plat_nand for (read|write)_buf
Subsystems (6)
Alexander Clouter
maintainerPATCH 1/1 mtd: extend plat_nand for (read|write)_buf
This patch adds (write|read)_buf callbacks to plat_nand.
The NAND on the TS-7800 provisioned by the FPGA allows readw() and readl() to be used which gives a 2.5x speed up. To be able to use this from the plat_nand driver a hook for read_buf (and also write_buf whilst we are in there) need to be made available. This patch adds the hook.
This is a repost[1] of a feature asked for before[2], plus other helpful patches to plat_nand[3] would be welcomed if at all possible too.
[1] http://lists.infradead.org/pipermail/linux-mtd/2009-March/024747.html
[2] http://lists.infradead.org/pipermail/linux-mtd/2007-October/019659.html
[3] http://lists.infradead.org/pipermail/linux-mtd/2009-March/024780.html
---
drivers/mtd/nand/plat_nand.c +2 -0
@@ -61,6 +61,8 @@ static int __devinit plat_nand_probe(struct platform_device *pdev) data->chip.cmd_ctrl = pdata->ctrl.cmd_ctrl; data->chip.dev_ready = pdata->ctrl.dev_ready; data->chip.select_chip = pdata->ctrl.select_chip; + data->chip.write_buf = pdata->ctrl.write_buf; + data->chip.read_buf = pdata->ctrl.read_buf; data->chip.chip_delay = pdata->chip.chip_delay; data->chip.options |= pdata->chip.options;
include/linux/mtd/nand.h +6 -0
@@ -584,6 +584,8 @@ struct platform_nand_chip {* @select_chip: platform specific chip select function * @cmd_ctrl: platform specific function for controlling * ALE/CLE/nCE. Also used to write command and address +* @write_buf: platform specific function for write buffer +* @read_buf: platform specific function for read buffer * @priv: private data to transport driver specific settings * * All fields are optional and depend on the hardware driver requirements
@@ -594,6 +596,10 @@ struct platform_nand_ctrl {void (*select_chip)(struct mtd_info *mtd,int chip);void (*cmd_ctrl)(struct mtd_info *mtd,int dat,unsigned int ctrl); +void (*write_buf)(struct mtd_info *mtd, +const uint8_t *buf,int len); +void (*read_buf)(struct mtd_info *mtd, +uint8_t *buf,int len);void *priv; };
Quoted 7 lines · Alexander Clouter
This patch adds (write|read)_buf callbacks to
plat_nand.The NAND on the TS-7800 provisioned by the FPGA allows
readw()andreadl()to be used which gives a 2.5x speed up. To be able to use this from theplat_nanddriver a hook forread_buf(and alsowrite_bufwhilst we are in there) need to be made available. This patch adds the hook.
Quoted 8 lines
This is a repost[1] of a feature asked for before[2], plus other helpful patches to plat_nand[3] would be welcomed if at all possible too.
[1] http://lists.infradead.org/pipermail/linux-mtd/2009-March/024747.html
[2] http://lists.infradead.org/pipermail/linux-mtd/2007-October/019659.html
[3] http://lists.infradead.org/pipermail/linux-mtd/2009-March/024780.htmlSigned-off-by Alexander Clouter <[email protected]>
I think this should be submitted with the user at the same time. Adding hooks without having a user is not very good. E.g., you may find out later that this does not help, or cancel your project, or end up with different call-backs, etc.
You may keep this patch privately, work on your NAND driver, and when you have it ready, you submit it with this at the same time.
Signature · 2 lines
Best regards, Artem Bityutskiy (Битюцкий Артём)
Alexander Clouter
maintainerHi,
Quoted 16 lines · Artem Bityutskiy
Quoted 14 lines · Alexander Clouter
This patch adds (write|read)_buf callbacks to
plat_nand.The NAND on the TS-7800 provisioned by the FPGA allows
readw()andreadl()to be used which gives a 2.5x speed up. To be able to use this from theplat_nanddriver a hook forread_buf(and alsowrite_bufwhilst we are in there) need to be made available. This patch adds the hook.This is a repost[1] of a feature asked for before[2], plus other helpful patches to plat_nand[3] would be welcomed if at all possible too.
[1] http://lists.infradead.org/pipermail/linux-mtd/2009-March/024747.html
[2] http://lists.infradead.org/pipermail/linux-mtd/2007-October/019659.html
[3] http://lists.infradead.org/pipermail/linux-mtd/2009-March/024780.html
Quoted 8 lines
Quoted 1 line
Signed-off-by Alexander Clouter <[email protected]>I think this should be submitted with the user at the same time. Adding hooks without having a user is not very good. E.g., you may find out later that this does not help, or cancel your project, or end up with different call-backs, etc.
In the case of read/write_buf...there is no other way, I'm all ears if you know how I can do (read|write)(w|l) with plat_nand *now*?
I just pulled off this type of manoeuvre for a HW RNG driver:
http://lkml.indiana.edu/hypermail/linux/kernel/0902.2/02727.html
http://git.marvell.com/?p=orion.git;a=commit;h=8be7b57dea0467324ad4545334c5e35c18f0814b
I doubt the ARM people would be happy if I push ARM patches through the MTD sub-system :)
Quoted 4 lines
You may keep this patch privately, work on your NAND driver, and when you have it ready, you submit it with this at the same time.
Erm, it's been ready for a *long* time, about a year[1], and is being used. Recently the bits of my plat_nand impementation[2] is on it's way to going mainline (Orion -> ARM -> Linus):
http://git.marvell.com/?p=orion.git;a=commit;h=75bb6b9aab3255f440ef4e72a31978d1681105d6
Cheers
[1] http://tech.groups.yahoo.com/group/ts-7000/message/11616
[2] http://lists.infradead.org/pipermail/linux-mtd/2009-February/024555.html
Signature · 2 lines
Alexander Clouter .sigmonster says: Your fly might be open (but don't check it just now).
Alexander Clouter
maintainerHi,
Quoted 20 lines · Alexander Clouter
Quoted 18 lines · Artem Bityutskiy
Quoted 15 lines · Alexander Clouter
This patch adds (write|read)_buf callbacks to
plat_nand.The NAND on the TS-7800 provisioned by the FPGA allows
readw()andreadl()to be used which gives a 2.5x speed up. To be able to use this from theplat_nanddriver a hook forread_buf(and alsowrite_bufwhilst we are in there) need to be made available. This patch adds the hook.This is a repost[1] of a feature asked for before[2], plus other helpful patches to plat_nand[3] would be welcomed if at all possible too.
[1] http://lists.infradead.org/pipermail/linux-mtd/2009-March/024747.html
[2] http://lists.infradead.org/pipermail/linux-mtd/2007-October/019659.html
[3] http://lists.infradead.org/pipermail/linux-mtd/2009-March/024780.htmlSigned-off-by Alexander Clouter <[email protected]>
Quoted 9 lines
Quoted 6 lines
I think this should be submitted with the user at the same time. Adding hooks without having a user is not very good. E.g., you may find out later that this does not help, or cancel your project, or end up with different call-backs, etc.
In the case of read/write_buf...there is no other way, I'm all ears if you know how I can do (read|write)(w|l) with
plat_nand*now*?
For those interested, what I'm currently doing is attached inline below, this of course goes on top of the existing plat_nand driver hooks going mainline now:
I probably will be re-writing the platform specific functions to be more like:
I doubt (read|write)w is worth it for the extra function complexity. Feedback welcome on this of course though; I was thinking maybe to leave it in as DMA support is just around the corner but on my particular board there are no alignment requirements for using the DMA engines.
Cheers
Cheers
Signature · 2 lines
Alexander Clouter .sigmonster says: Yow! I want to mail a bronzed artichoke to Nicaragua!
arch/arm/mach-orion5x/ts78xx-setup.c +72 -1
@@ -191,6 +191,75 @@ static int ts78xx_ts_nand_dev_ready(struct mtd_info *mtd)return readb (TS_NAND_CTRL ) &0x20 ; } +/* +* FPGA will handle 8/16/32bit reads and writes +*/ +static void ts78xx_ts_nand_read_buf (struct mtd_info *mtd, +uint8_t *buf,int len) +{ +struct nand_chip *this = mtd->priv; + +if (len %2 ) { + *buf++ =readb (this->IO_ADDR_R); + len--; + } +if (len %4 ==2 ) { +uint16_t *t = (uint16_t *) buf; + *t =readw (this->IO_ADDR_R); + buf +=2 ; + len -=2 ; + } + +while (len >3 ) { +uint32_t *t = (uint32_t *) buf; + *t =readl (this->IO_ADDR_R); + buf +=4 ; + len -=4 ; + } + +if (len %4 ==2 ) { +uint16_t *t = (uint16_t *) buf; + *t =readw (this->IO_ADDR_R); + buf +=2 ; + len -=2 ; + } +if (len) + *buf =readb (this->IO_ADDR_R); +} + +static void ts78xx_ts_nand_write_buf (struct mtd_info *mtd, +const uint8_t *buf,int len) +{ +struct nand_chip *this = mtd->priv; + +if (len %2 ) { +writeb (*buf++, this->IO_ADDR_W); + len--; + } +if (len %4 ==2 ) { +uint16_t *t = (uint16_t *) buf; +writew (*t, this->IO_ADDR_W); + buf +=2 ; + len -=2 ; + } + +while (len >3 ) { +uint32_t *t = (uint32_t *) buf; +writel (*t, this->IO_ADDR_W); + buf +=4 ; + len -=4 ; + } + +if (len %4 ==2 ) { +uint16_t *t = (uint16_t *) buf; +writew (*t, this->IO_ADDR_W); + buf +=2 ; + len -=2 ; + } +if (len) +writeb (*buf, this->IO_ADDR_W); +} +const char *ts_nand_part_probes[] = {"cmdlinepart" ,NULL };static struct mtd_partition ts78xx_ts_nand_parts[] = {
@@ -230,8 +299,10 @@ static struct platform_nand_data ts78xx_ts_nand_data = {* e6cf5df1838c28bb060ac45b5585e48e71bbc740 so now there is * no performance advantage to be had so we no longer bother */ - .cmd_ctrl = ts78xx_ts_nand_cmd_ctrl, .dev_ready = ts78xx_ts_nand_dev_ready, + .cmd_ctrl = ts78xx_ts_nand_cmd_ctrl, + .read_buf = ts78xx_ts_nand_read_buf, + .write_buf = ts78xx_ts_nand_write_buf, }, };
drivers/mtd/nand/plat_nand.c +2 -0
@@ -61,6 +61,8 @@ static int __devinit plat_nand_probe(struct platform_device *pdev) data->chip.cmd_ctrl = pdata->ctrl.cmd_ctrl; data->chip.dev_ready = pdata->ctrl.dev_ready; data->chip.select_chip = pdata->ctrl.select_chip; + data->chip.write_buf = pdata->ctrl.write_buf; + data->chip.read_buf = pdata->ctrl.read_buf; data->chip.chip_delay = pdata->chip.chip_delay; data->chip.options |= pdata->chip.options;
include/linux/mtd/nand.h +6 -0
@@ -584,6 +584,8 @@ struct platform_nand_chip {* @select_chip: platform specific chip select function * @cmd_ctrl: platform specific function for controlling * ALE/CLE/nCE. Also used to write command and address +* @write_buf: platform specific function for write buffer +* @read_buf: platform specific function for read buffer * @priv: private data to transport driver specific settings * * All fields are optional and depend on the hardware driver requirements
@@ -594,6 +596,10 @@ struct platform_nand_ctrl {void (*select_chip)(struct mtd_info *mtd,int chip);void (*cmd_ctrl)(struct mtd_info *mtd,int dat,unsigned int ctrl); +void (*write_buf)(struct mtd_info *mtd, +const uint8_t *buf,int len); +void (*read_buf)(struct mtd_info *mtd, +uint8_t *buf,int len);void *priv; };
Quoted 7 lines · Alexander Clouter
Quoted 4 lines
You may keep this patch privately, work on your NAND driver, and when you have it ready, you submit it with this at the same time.
Erm, it's been ready for a *long* time, about a year[1], and is being used. Recently the bits of my
plat_nandimpementation[2] is on it's way to going mainline (Orion -> ARM -> Linus):
Ok, I thought you do not use this so far.
I've pushed your patch to l2-mtd-2.6.git. This does not guarantee anything, just means I'll ping David to look at it and send upstream.
Signature · 2 lines
Best regards, Artem Bityutskiy (Битюцкий Артём)