split MMC_CAP_4_BIT_DATA

The on-chip Blackfin MMC/SD/SDIO host controller has the ability to do 1-bit MMC, 1-bit/4-bit SD, and 1-bit/4-bit SDIO. Thus the current convention of MMC_CAP_4_BIT_DATA meaning "your host controller can do 1-bit or 4-bit for all modes" is insufficient for our needs. The attached patch splits MMC_CAP_4_BIT_DATA into MMC_CAP_MMC_4_BIT_DATA and MMC_CAP_SD_4_BIT_DATA and updates all host controllers to include these in their caps and then changes existing code to check the new defines. At the moment, SD/SDIO are lumped into MMC_CAP_SD_4_BIT_DATA ... should I bother with splitting that into SD and SDIO as well while I'm doing this ?

Signed-off-by Mike Frysinger <[email protected]>

Quoted 11 lines · Mike Frysinger

The on-chip Blackfin MMC/SD/SDIO host controller has the ability to do 1-bit MMC, 1-bit/4-bit SD, and 1-bit/4-bit SDIO. Thus the current convention of MMC_CAP_4_BIT_DATA meaning "your host controller can do 1-bit or 4-bit for all modes" is insufficient for our needs. The attached patch splits MMC_CAP_4_BIT_DATA into MMC_CAP_MMC_4_BIT_DATA and MMC_CAP_SD_4_BIT_DATA and updates all host controllers to include these in their caps and then changes existing code to check the new defines. At the moment, SD/SDIO are lumped into MMC_CAP_SD_4_BIT_DATA ... should I bother with splitting that into SD and SDIO as well while I'm doing this ?

Quoted 1 line

Signed-off-by Mike Frysinger <[email protected]>

I fail to see why you need to split MMC and SD. Could you elaborate why the controller won't work with MMC cards? I haven't seen any differences from SD.

i dont understand what's confusing. the Blackfin on chip host controller only supports 1-bit MMC, but it supports 4-bit SD/SDIO. this is a fact. while it may be a stupid decision, it is what it is, and i need the framework made more flexible in order to get the Blackfin driver merged cleanly. we do software for hardware, we dont do hardware.

Quoted 5 lines · On Tue, 8 Jan 2008 14:40:49 -0500 Mike Frysinger <[email protected]> wrote:

i dont understand what's confusing. the Blackfin on chip host controller only supports 1-bit MMC, but it supports 4-bit SD/SDIO. this is a fact. while it may be a stupid decision, it is what it is, and i need the framework made more flexible in order to get the Blackfin driver merged cleanly. we do software for hardware, we dont do hardware.

Well, since I've seen no _hardware_ differences between 4-bit MMC and 4-bit SD, "support" in this case must me "vendor will guarantee it works". And that is not the kind of "support" that needs a distinction in the code.

So, again, if you feel that there is a hardware difference between 4-bit MMC and 4-bit SD then please elaborate as it is my understanding that they are identical.

Quoted 1 line · Pierre Ossman

So, again, if you feel that there is a hardware difference between 4-bit MMC and 4-bit SD then please elaborate as it is my understanding that they are identical.

you may be 100% correct, i have no idea, i'm not really familiar with MMC/SD/SDIO at all.

Quoted 6 lines · On Tue, 8 Jan 2008 14:40:49 -0500 Mike Frysinger <[email protected]> wrote:

i dont understand what's confusing. the Blackfin on chip host controller only supports 1-bit MMC, but it supports 4-bit SD/SDIO. this is a fact. while it may be a stupid decision, it is what it is, and i need the framework made more flexible in order to get the Blackfin driver merged cleanly. we do software for hardware, we dont do hardware.

Well, since I've seen no _hardware_ differences between 4-bit MMC and 4-bit SD, "support" in this case must me "vendor will guarantee it works". And that is not the kind of "support" that needs a distinction in the code.

So, again, if you feel that there is a hardware difference between 4-bit MMC and 4-bit SD then please elaborate as it is my understanding that they are identical.

Quoted 6 lines · On Tue, 8 Jan 2008 14:40:49 -0500 Mike Frysinger <[email protected]> wrote:

i dont understand what's confusing. the Blackfin on chip host controller only supports 1-bit MMC, but it supports 4-bit SD/SDIO. this is a fact. while it may be a stupid decision, it is what it is, and i need the framework made more flexible in order to get the Blackfin driver merged cleanly. we do software for hardware, we

Well, since I've seen no _hardware_ differences between 4-bit MMC and

So, again, if you feel that there is a hardware difference between

Hi,all

I'd like to say something about this issue.
Currently,the blackfin on chip SD host ONLY support 1-bit MMC while support 1-bit/4-bit SD/SDIO.
And we want our driver to support both 1-bit MMC and 4-bit SD/SDIO.but the current MMC driver framework
Only allow us to set one kind of bus width,either 1-bit or 4-bit.So in order to meet our case,we need more flexible mechanism
To inform the upper commom driver to know our situation.

Quoted 12 lines · Cliff

Hi,all

I'd like to say something about this issue.
Currently,the blackfin on chip SD host ONLY support 1-bit MMC while support 1-bit/4-bit SD/SDIO.
And we want our driver to support both 1-bit MMC and 4-bit SD/SDIO.but the current MMC driver framework
Only allow us to set one kind of bus width,either 1-bit or 4-bit.So in order to meet our case,we need more flexible mechanism
To inform the upper commom driver to know our situation.

That's just iterating what's already been said. My claim is that 4-bit is 4-bit, regardless if it's MMC or SD. So if you want this patch to go in you need to explain why there is a difference for the blackfin controller.

Actually, Blackfin BF54x on-chip SDIO host controller is supposed to support 1-bit and 4-bit MMC or SD. But in the real platform, when MMC works at 4-bit mode there will be a FIFO underrun error
(http://blackfin.uclinux.org/gf/project/uclinux-dist/tracker/?action=TrackerItemEdit&tracker_item_id=3535).
While SD works at 4-bit mode is OK. We believe this is a hardware issue of the SDIO host controller of BF54x.

We intend to workaround in our own BF54x SDIO host controller driver, but there is no place for us to determine MMC or SD. One choice is to hack common code like this:

Quoted 1 line · On Thu, 10 Jan 2008 00:45:27 +0800 "Bryan Wu" <[email protected]> wrote:

Quoted 6 lines

Actually, Blackfin BF54x on-chip SDIO host controller is supposed to support 1-bit and 4-bit MMC or SD. But in the real platform, when MMC works at 4-bit mode there will be a FIFO underrun error
(http://blackfin.uclinux.org/gf/project/uclinux-dist/tracker/?action=TrackerItemEdit&tracker_item_id=3535).
While SD works at 4-bit mode is OK. We believe this is a hardware issue of the SDIO host controller of BF54x.

Odd. As there shouldn't be a difference, I'm inclined to believe that this is card dependent. So it could as well appear with some SD cards as well.

How much have you tested this issue? That tracker entry isn't very verbose.

Also, how can an underrun appear? The system should have flow control. And an underrun does not sound like a bus problem, rather that you have a card with different speed characteristics than the hardware/driver can handle.

Quoted 4 lines

We intend to workaround in our own BF54x SDIO host controller driver, but there is no place for us to determine MMC or SD. One choice is to hack common code like this:

_If_ it truly cannot handle 4-bit MMC, then Mike's patch is the way to go. But I'm not yet convinced that the problem is related to MMC vs SD.

Rgds
--
-- Pierre Ossman

Linux kernel, MMC maintainer http://www.kernel.org
PulseAudio, core developer http://pulseaudio.org
rdesktop, core developer http://www.rdesktop.org

At page 4-3 of ADSP-BF54x Blackfin(R) Processor Peripheral Hardware Reference, there is a table which guide us the SDH controller does not support 4-bit mode MMC card. Please found the table picture in the attachment or get the document at:
http://www.analog.com/UploadedFiles/Associated_Docs/61460151169789ADSP_BF54x_Blackfin_Processor_Peripheral_Hardware_Reference.pdf

Thanks, actually we are not yet convinced by our hardware designer why BF54x SDH does not support 4-bit MMC.

Quoted 1 line · On Thu, 10 Jan 2008 17:22:55 +0800 "Bryan Wu" <[email protected]> wrote:

Quoted 6 lines

At page 4-3 of ADSP-BF54x Blackfin(R) Processor Peripheral Hardware Reference, there is a table which guide us the SDH controller does not support 4-bit mode MMC card. Please found the table picture in the attachment or get the document at:
http://www.analog.com/UploadedFiles/Associated_Docs/61460151169789ADSP_BF54x_Blackfin_Processor_Peripheral_Hardware_Reference.pdf

Ok, but this just means the controller wasn't designed with 4-bit MMC in mind. As several other "SD-only" controller have been tested with modern MMC cards without a problem, this is not sufficient to explain any problem.

Quoted 4 lines

Thanks, actually we are not yet convinced by our hardware designer why BF54x SDH does not support 4-bit MMC.

Please keep me informed on how it progresses. I'd like an at least plausible explanation, preferably also some empirical data, before I'm ready to accept Mike's patch.

Quoted 4 lines · Pierre Ossman

Quoted 1 line · On Thu, 10 Jan 2008 17:22:55 +0800 "Bryan Wu" <[email protected]> wrote:

Quoted 9 lines

Quoted 6 lines

At page 4-3 of ADSP-BF54x Blackfin(R) Processor Peripheral Hardware Reference, there is a table which guide us the SDH controller does not support 4-bit mode MMC card. Please found the table picture in the attachment or get the document at:
http://www.analog.com/UploadedFiles/Associated_Docs/61460151169789ADSP_BF54x_Blackfin_Processor_Peripheral_Hardware_Reference.pdf

Ok, but this just means the controller wasn't designed with 4-bit MMC in mind. As several other "SD-only" controller have been tested with modern MMC cards without a problem, this is not sufficient to explain any problem.

This should be some HW design issue and BF54x including SDH controller is in mass production.
There is no chance to change the silicon, we just wanna use software driver to workaround this issue.
So Mike's patch is here.

Quoted 5 lines

Quoted 4 lines

Thanks, actually we are not yet convinced by our hardware designer why BF54x SDH does not support 4-bit MMC.

Quoted 2 lines

Please keep me informed on how it progresses. I'd like an at least plausible explanation, preferably also some empirical data, before I'm ready to accept Mike's patch.

We were told this is an hardware design issue, so please help us to workaround it in software side with Mike's patch.
And how do you think my suggestion to Mike's patch,
#define MMC_CAP_4_BIT_DATA ( MMC_CAP_SD_4_BIT_DATA | MMC_CAP_MMC_4_BIT_DATA) Then no need to change other MMC host drivers

We were told this is an hardware design issue, so please help us to workaround it in software side with Mike's patch.

Quoted 2 lines · Bryan Wu

We were told this is an hardware design issue, so please help us to workaround it in software side with Mike's patch.

So it's far more probable that you've misdiagnosed your error than this being the actual problem.

The fact that 4-bit MMC support has been present for some time and you're only now seeing a problem further supports that.

Quoted 1 line · Pierre Ossman

So it's far more probable that you've misdiagnosed your error than this being the actual problem.

the guys who design the silicon are telling us it doesnt work. our tests show it not working. i'm not sure what else you want here.

Quoted 1 line

The fact that 4-bit MMC support has been present for some time and you're only now seeing a problem further supports that.

not in the least actually. we just started developing the driver. it has seen no real world use. we've been testing things internally to make sure the setup is sane. 4-bit MMC is coming up as not working.

Quoted 5 lines · Mike Frysinger

Quoted 1 line · Pierre Ossman

So it's far more probable that you've misdiagnosed your error than this being the actual problem.

the guys who design the silicon are telling us it doesnt work. our tests show it not working. i'm not sure what else you want here.

More information. You have not provided any speculation as to why it doesn't work, or what you've done to figure it out. You haven't even described how it fails, just that it does. No information about what cards you've tested and how you've tested them to confirm that your theory on 4-bit data is accurate.

You have no track record with me, and you're proposing something that goes against all known information. So you have to understand that I'm sceptical about your claims. I won't dismiss them off hand, but I need to be convinced you haven't made a mistake when coming to this conclusion.

Cliff should be able to enumerate the cards he has tested and the issues he's run into. i'll see if i cant get more in depth information from the hardware guys beyond the documentation on the sdh already available in the hardware manual (which i'm guessing has not been enough to appease you).

Quoted 6 lines · On Fri, 11 Jan 2008 04:47:44 -0500 "Mike Frysinger" <[email protected]>…

Cliff should be able to enumerate the cards he has tested and the issues he's run into. i'll see if i cant get more in depth information from the hardware guys beyond the documentation on the sdh already available in the hardware manual (which i'm guessing has not been enough to appease you).

Great. And you're right in that the spec did not give any indication that MMC shouldn't work.

I did notice one thing though. The pull-ups and pull-downs seem to be controllable. If I remember correctly, DAT3 needs to have a controller-side pull-up for MMC cards. And the docs said the default was not to provide that (in order to use the internal pull-up that most, but not all, SD cards have for detection). Might be something to look at if you haven't already.

I guess it is for MMC/SD card insert detection. Is it related with the 4-bit MMC and 4-bit SD?
Actually, there are some issues about the card insert detection on BF54x SDH. Following is some comments of our bfin-sdh driver which is in the development.

Quoted 2 lines · On Fri, 11 Jan 2008 04:08:53 -0500 "Mike Frysinger" <[email protected]>…

Quoted 1 line · Pierre Ossman

So it's far more probable that you've misdiagnosed your error than

this being the actual problem.

Quoted 4 lines

the guys who design the silicon are telling us it doesnt work. our tests show it not working. i'm not sure what else you want here.

More information. You have not provided any speculation as to why it doesn't work, or what you've done to figure it out.

On Fri 11 Jan 2008 04:35, Pierre Ossman pondered:

Quoted 5 lines

Quoted 2 lines · On Fri, 11 Jan 2008 04:08:53 -0500 "Mike Frysinger" <[email protected]>…

Quoted 1 line · Pierre Ossman

So it's far more probable that you've misdiagnosed your error than

Quoted 8 lines

this being the actual problem.

Quoted 4 lines

the guys who design the silicon are telling us it doesnt work. our tests show it not working. i'm not sure what else you want here.

More information. You have not provided any speculation as to why it doesn't work, or what you've done to figure it out.

I have been trying to get the detailed information from the hardware (silicon designer) guy about why he believes support for 4-bit MMC is not there, but I have not heard back yet.

As soon as I hear from them - and either understand why 4-bit MMC doesn't work, or if the Blackfin docs are incorrect (either is OK outcome) - I will let you know.

On Fri 11 Jan 2008 12:52, Robin Getz pondered:

Quoted 19 lines

On Fri 11 Jan 2008 04:35, Pierre Ossman pondered:

Quoted 5 lines

Quoted 2 lines · On Fri, 11 Jan 2008 04:08:53 -0500 "Mike Frysinger" <[email protected]>…

Quoted 1 line · Pierre Ossman

So it's far more probable that you've misdiagnosed your error than

Quoted 8 lines

this being the actual problem.

Quoted 4 lines

the guys who design the silicon are telling us it doesnt work. our tests show it not working. i'm not sure what else you want here.

More information. You have not provided any speculation as to why it doesn't work, or what you've done to figure it out.

I have been trying to get the detailed information from the hardware (silicon designer) guy about why he believes support for 4-bit MMC is not there, but I have not heard back yet.

Quoted 3 lines

As soon as I hear from them - and either understand why 4-bit MMC doesn't work, or if the Blackfin docs are incorrect (either is OK outcome) - I will let you know.

According to the HW folks - it is exactly as Pierre indicated - in theory it should work, 4-bit MMC requires usage of different set of commands as compared to 4-bit SD, so it should be just software - although no one has tested it yet.

So I guess this is back on Bryan/Cliff to make work....

Thanks for pushing back.

According to the HW folks - it is exactly as Pierre indicated - in theory it should work, 4-bit MMC requires usage of different set of commands as compared to 4-bit SD, so it should be just software - although no one has tested it yet.

So I guess this is back on Bryan/Cliff to make work....

Thanks for pushing back.

The on-chip Blackfin MMC/SD/SDIO host controller has the ability to do 1-bit MMC, 1-bit/4-bit SD, and 1-bit/4-bit SDIO. Thus the current convention of MMC_CAP_4_BIT_DATA meaning "your host controller can do 1-bit or 4-bit for all modes" is insufficient for our needs. The attached patch splits MMC_CAP_4_BIT_DATA into MMC_CAP_MMC_4_BIT_DATA and MMC_CAP_SD_4_BIT_DATA and updates all host controllers to include these in their caps and then changes existing code to check the new defines. At the moment, SD/SDIO are lumped into MMC_CAP_SD_4_BIT_DATA ... should I bother with splitting that into SD and SDIO as well while I'm doing this ?

Signed-off-by Mike Frysinger <[email protected]>

---

drivers/mmc/core/mmc.c +1 -1

index 68c0e3b..ca12db7 100644
@@ -397,7 +397,7 @@ static int mmc_init_card(struct mmc_host *host, u32 ocr,
         * Activate wide bus (if supported).
         */
        if ((card->csd.mmca_vsn >= CSD_SPEC_VER_4) &&
-               (host->caps & MMC_CAP_4_BIT_DATA)) {
+               (host->caps & MMC_CAP_MMC_4_BIT_DATA)) {
                err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
                        EXT_CSD_BUS_WIDTH, EXT_CSD_BUS_WIDTH_4);
                if (err)

drivers/mmc/core/sd.c +1 -1

index d1c1e0f..974b63d 100644
@@ -441,7 +441,7 @@ static int mmc_sd_init_card(struct mmc_host *host, u32 ocr,
        /*
         * Switch to wider bus (if supported).
         */
-       if ((host->caps & MMC_CAP_4_BIT_DATA) &&
+       if ((host->caps & MMC_CAP_SD_4_BIT_DATA) &&
                (card->scr.bus_widths & SD_SCR_BUS_WIDTH_4)) {
                err = mmc_app_set_bus_width(card, MMC_BUS_WIDTH_4);
                if (err)

drivers/mmc/core/sdio.c +1 -1

index 87a50f4..1d389c8 100644
@@ -143,7 +143,7 @@ static int sdio_enable_wide(struct mmc_card *card)
        int ret;
        u8 ctrl;
 
-       if (!(card->host->caps & MMC_CAP_4_BIT_DATA))
+       if (!(card->host->caps & MMC_CAP_SD_4_BIT_DATA))
                return 0;
 
        if (card->cccr.low_speed && !card->cccr.wide_bus)

drivers/mmc/host/at91_mci.c +1 -1

index b1edcef..63d89b0 100644
@@ -851,7 +851,7 @@ static int __init at91_mci_probe(struct platform_device *pdev)
        host->board = pdev->dev.platform_data;
        if (host->board->wire4) {
                if (cpu_is_at91sam9260() || cpu_is_at91sam9263())
-                       mmc->caps |= MMC_CAP_4_BIT_DATA;
+                       mmc->caps |= (MMC_CAP_SD_4_BIT_DATA | MMC_CAP_MMC_4_BIT_DATA);
                else
                        printk("AT91 MMC: 4 wire bus mode not supported"
                                " - using 1 wire\n");

drivers/mmc/host/imxmmc.c +1 -1

index f2070a1..67d4bc0 100644
@@ -975,7 +975,7 @@ static int imxmci_probe(struct platform_device *pdev)
        mmc->f_min = 150000;
        mmc->f_max = CLK_RATE/2;
        mmc->ocr_avail = MMC_VDD_32_33;
-       mmc->caps = MMC_CAP_4_BIT_DATA;
+       mmc->caps = (MMC_CAP_SD_4_BIT_DATA | MMC_CAP_MMC_4_BIT_DATA);
 
        /* MMC core transfer sizes tunable parameters */
        mmc->max_hw_segs = 64;

drivers/mmc/host/omap.c +1 -1

index 971e18b..b1ae793 100644
@@ -1079,7 +1079,7 @@ static int __init mmc_omap_probe(struct platform_device *pdev)
        mmc->caps = MMC_CAP_MULTIWRITE | MMC_CAP_BYTEBLOCK;
 
        if (minfo->wire4)
-                mmc->caps |= MMC_CAP_4_BIT_DATA;
+                mmc->caps |= (MMC_CAP_SD_4_BIT_DATA | MMC_CAP_MMC_4_BIT_DATA);
 
        /* Use scatterlist DMA to reduce per-transfer costs.
         * NOTE max_seg_size assumption that small blocks aren't

drivers/mmc/host/pxamci.c +1 -1

index 1654a33..4fa00f1 100644
@@ -527,7 +527,7 @@ static int pxamci_probe(struct platform_device *pdev)
        mmc->caps = 0;
        host->cmdat = 0;
        if (!cpu_is_pxa21x() && !cpu_is_pxa25x()) {
-               mmc->caps |= MMC_CAP_4_BIT_DATA | MMC_CAP_SDIO_IRQ;
+               mmc->caps |= MMC_CAP_SD_4_BIT_DATA | MMC_CAP_MMC_4_BIT_DATA | MMC_CAP_SDIO_IRQ;
                host->cmdat |= CMDAT_SDIO_INT_EN;
        }
 

drivers/mmc/host/sdhci.c +1 -1

index 785bbdc..09c4358 100644
@@ -1405,7 +1405,7 @@ static int __devinit sdhci_probe_slot(struct pci_dev *pdev, int slot)
        mmc->ops = &sdhci_ops;
        mmc->f_min = host->max_clk / 256;
        mmc->f_max = host->max_clk;
-       mmc->caps = MMC_CAP_4_BIT_DATA | MMC_CAP_MULTIWRITE | MMC_CAP_SDIO_IRQ;
+       mmc->caps = MMC_CAP_SD_4_BIT_DATA | MMC_CAP_MMC_4_BIT_DATA | MMC_CAP_MULTIWRITE | MMC_CAP_SDIO_IRQ;
 
        if (caps & SDHCI_CAN_DO_HISPD)
                mmc->caps |= MMC_CAP_SD_HIGHSPEED;

drivers/mmc/host/tifm_sd.c +1 -1

index 20d5c7b..78cad6b 100644
@@ -973,7 +973,7 @@ static int tifm_sd_probe(struct tifm_dev *sock)
 
        mmc->ops = &tifm_sd_ops;
        mmc->ocr_avail = MMC_VDD_32_33 | MMC_VDD_33_34;
-       mmc->caps = MMC_CAP_4_BIT_DATA | MMC_CAP_MULTIWRITE;
+       mmc->caps = MMC_CAP_SD_4_BIT_DATA | MMC_CAP_MMC_4_BIT_DATA | MMC_CAP_MULTIWRITE;
        mmc->f_min = 20000000 / 60;
        mmc->f_max = 24000000;
 

drivers/mmc/host/wbsd.c +1 -1

index 4d5f374..70d6071 100644
@@ -1219,7 +1219,7 @@ static int __devinit wbsd_alloc_mmc(struct device *dev)
        mmc->f_min = 375000;
        mmc->f_max = 24000000;
        mmc->ocr_avail = MMC_VDD_32_33 | MMC_VDD_33_34;
-       mmc->caps = MMC_CAP_4_BIT_DATA | MMC_CAP_MULTIWRITE;
+       mmc->caps = MMC_CAP_SD_4_BIT_DATA | MMC_CAP_MMC_4_BIT_DATA | MMC_CAP_MULTIWRITE;
 
        spin_lock_init(&host->lock);
 

include/linux/mmc/host.h +0 -0

index 7ab962f..142fa69 100644
@@ -88,12 +88,13 @@ struct mmc_host {
 
        unsigned long           caps;           /* Host capabilities */