A few months ago, hans de Goede and I where talking about the behavior of HPI on
eMMC modules. We came to the conclusion that more often then not, the mmc
chipset might be the problem and not the eMMC module itself. To remedy this the
broken-hpi parameter is now also checked on the chipset level.
On the sunxi platform, all the eMMC users with emmc modules that support HPI
need the broken-hpi flag, indicating that the sunxi mmc controller is likely
not properly supporting HPI. We disable this on the chipset level for these
chipsets.
As for the broken-hpi flag in the devicetree's, those should in theory be
removed from the boards in question in a separate patch when in agreement, as
technically, the eMMC module supports HPI just fine and thus the we are lying
about what really is broken.
This was tested on an OLinuXino Lime2 with 4GB industrial grade Micron eMMC
flash.
Olliver Schinagl (2):
mmc: core: Improve marking broken HPI through devicetree
mmc: sunxi: Mark the HPI of the mmc controller broken by default
drivers/mmc/core/host.c | 2 ++
drivers/mmc/core/mmc.c | 2 +-
drivers/mmc/host/sunxi-mmc.c | 4 ++--
include/linux/mmc/host.h | 1 +
4 files changed, 6 insertions(+), 3 deletions(-)
--
2.8.0.rc3
So far it seems all sunxi MMC controllers have a broken HPI. Until
proved otherwise, mark all sunxi mmc controllers as having a broken HPI.
Signed-off-by: Olliver Schinagl <[email protected]>
---
drivers/mmc/host/sunxi-mmc.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/mmc/host/sunxi-mmc.c b/drivers/mmc/host/sunxi-mmc.c
index 8372a41..acf24fb 100644
--- a/drivers/mmc/host/sunxi-mmc.c
+++ b/drivers/mmc/host/sunxi-mmc.c
@@ -1126,8 +1126,8 @@ static int sunxi_mmc_probe(struct platform_device *pdev)
mmc->f_min = 400000;
mmc->f_max = 52000000;
mmc->caps |= MMC_CAP_MMC_HIGHSPEED | MMC_CAP_SD_HIGHSPEED |
- MMC_CAP_1_8V_DDR |
- MMC_CAP_ERASE | MMC_CAP_SDIO_IRQ;
+ MMC_CAP_1_8V_DDR | MMC_CAP_ERASE |
+ MMC_CAP_SDIO_IRQ | MMC_CAP_BROKEN_HPI;
ret = mmc_of_parse(mmc);
if (ret)
--
2.8.0.rc3
In patch 81f8a7be66 Hans de Goede added a patch to allow marking an mmc
device as to having an broken HPI implementation. After talking some
with Hans, we now think it is actually the mmc controller that can be
broken and not support broken HPI's.
This patch adds a new capability, mmc-broken-hpi, which allows us to
mark a broken hpi implementation on the host level.
Signed-off-by: Olliver Schinagl <[email protected]>
---
drivers/mmc/core/host.c | 2 ++
drivers/mmc/core/mmc.c | 2 +-
include/linux/mmc/host.h | 1 +
3 files changed, 4 insertions(+), 1 deletion(-)
diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
index 6e4c55a..9b63b36 100644
--- a/drivers/mmc/core/host.c
+++ b/drivers/mmc/core/host.c
@@ -270,6 +270,8 @@ int mmc_of_parse(struct mmc_host *host)
host->caps |= MMC_CAP_HW_RESET;
if (of_property_read_bool(np, "cap-sdio-irq"))
host->caps |= MMC_CAP_SDIO_IRQ;
+ if (of_property_read_bool(np, "mmc-broken-hpi"))
+ host->caps |= MMC_CAP_BROKEN_HPI;
if (of_property_read_bool(np, "full-pwr-cycle"))
host->caps2 |= MMC_CAP2_FULL_PWR_CYCLE;
if (of_property_read_bool(np, "keep-power-in-suspend"))
diff --git a/drivers/mmc/core/mmc.c b/drivers/mmc/core/mmc.c
index 4dbe3df..9a19562 100644
--- a/drivers/mmc/core/mmc.c
+++ b/drivers/mmc/core/mmc.c
@@ -1592,7 +1592,7 @@ static int mmc_init_card(struct mmc_host *host, u32 ocr,
/*
* Enable HPI feature (if supported)
*/
- if (card->ext_csd.hpi) {
+ if (card->ext_csd.hpi && !(host->caps & MMC_CAP_BROKEN_HPI)) {
err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
EXT_CSD_HPI_MGMT, 1,
card->ext_csd.generic_cmd6_time);
diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
index 8dd4d29..20f758e 100644
--- a/include/linux/mmc/host.h
+++ b/include/linux/mmc/host.h
@@ -264,6 +264,7 @@ struct mmc_host {
#define MMC_CAP_DRIVER_TYPE_A (1 << 23) /* Host supports Driver Type A */
#define MMC_CAP_DRIVER_TYPE_C (1 << 24) /* Host supports Driver Type C */
#define MMC_CAP_DRIVER_TYPE_D (1 << 25) /* Host supports Driver Type D */
+#define MMC_CAP_BROKEN_HPI (1 << 29) /* Host support for HPI is broken */
#define MMC_CAP_CMD23 (1 << 30) /* CMD23 supported. */
#define MMC_CAP_HW_RESET (1 << 31) /* Hardware reset */
--
2.8.0.rc3
On 19 April 2016 at 09:12, Olliver Schinagl <[email protected]> wrote:
> In patch 81f8a7be66 Hans de Goede added a patch to allow marking an mmc
> device as to having an broken HPI implementation. After talking some
> with Hans, we now think it is actually the mmc controller that can be
> broken and not support broken HPI's.
I don't want us to invent a DT binding for something you *think* is a
HW controller issue.
Have you really excluded that this isn't a software issue? Me
personally haven't been using HPI that much so I can't really tell
about the code robustness from the mmc core (mmc protocol point of
view).
Kind regards
Uffe
>
> This patch adds a new capability, mmc-broken-hpi, which allows us to
> mark a broken hpi implementation on the host level.
>
> Signed-off-by: Olliver Schinagl <[email protected]>
> ---
> drivers/mmc/core/host.c | 2 ++
> drivers/mmc/core/mmc.c | 2 +-
> include/linux/mmc/host.h | 1 +
> 3 files changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
> index 6e4c55a..9b63b36 100644
> --- a/drivers/mmc/core/host.c
> +++ b/drivers/mmc/core/host.c
> @@ -270,6 +270,8 @@ int mmc_of_parse(struct mmc_host *host)
> host->caps |= MMC_CAP_HW_RESET;
> if (of_property_read_bool(np, "cap-sdio-irq"))
> host->caps |= MMC_CAP_SDIO_IRQ;
> + if (of_property_read_bool(np, "mmc-broken-hpi"))
> + host->caps |= MMC_CAP_BROKEN_HPI;
> if (of_property_read_bool(np, "full-pwr-cycle"))
> host->caps2 |= MMC_CAP2_FULL_PWR_CYCLE;
> if (of_property_read_bool(np, "keep-power-in-suspend"))
> diff --git a/drivers/mmc/core/mmc.c b/drivers/mmc/core/mmc.c
> index 4dbe3df..9a19562 100644
> --- a/drivers/mmc/core/mmc.c
> +++ b/drivers/mmc/core/mmc.c
> @@ -1592,7 +1592,7 @@ static int mmc_init_card(struct mmc_host *host, u32 ocr,
> /*
> * Enable HPI feature (if supported)
> */
> - if (card->ext_csd.hpi) {
> + if (card->ext_csd.hpi && !(host->caps & MMC_CAP_BROKEN_HPI)) {
> err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
> EXT_CSD_HPI_MGMT, 1,
> card->ext_csd.generic_cmd6_time);
> diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
> index 8dd4d29..20f758e 100644
> --- a/include/linux/mmc/host.h
> +++ b/include/linux/mmc/host.h
> @@ -264,6 +264,7 @@ struct mmc_host {
> #define MMC_CAP_DRIVER_TYPE_A (1 << 23) /* Host supports Driver Type A */
> #define MMC_CAP_DRIVER_TYPE_C (1 << 24) /* Host supports Driver Type C */
> #define MMC_CAP_DRIVER_TYPE_D (1 << 25) /* Host supports Driver Type D */
> +#define MMC_CAP_BROKEN_HPI (1 << 29) /* Host support for HPI is broken */
> #define MMC_CAP_CMD23 (1 << 30) /* CMD23 supported. */
> #define MMC_CAP_HW_RESET (1 << 31) /* Hardware reset */
>
> --
> 2.8.0.rc3
>
Hi Ulf,
On 19-04-16 11:29, Ulf Hansson wrote:
> On 19 April 2016 at 09:12, Olliver Schinagl <[email protected]> wrote:
>> In patch 81f8a7be66 Hans de Goede added a patch to allow marking an mmc
>> device as to having an broken HPI implementation. After talking some
>> with Hans, we now think it is actually the mmc controller that can be
>> broken and not support broken HPI's.
> I don't want us to invent a DT binding for something you *think* is a
> HW controller issue.
>
> Have you really excluded that this isn't a software issue? Me
> personally haven't been using HPI that much so I can't really tell
> about the code robustness from the mmc core (mmc protocol point of
> view).
Well this patch goes hand in hand so to speak with the broken-hpi patch
introduced by him, he did most of the investigation. We just discussed
how to handle it and asked me to cook up the patch.
@hans, what do you think?
>
> Kind regards
> Uffe
>
>> This patch adds a new capability, mmc-broken-hpi, which allows us to
>> mark a broken hpi implementation on the host level.
>>
>> Signed-off-by: Olliver Schinagl <[email protected]>
>> ---
>> drivers/mmc/core/host.c | 2 ++
>> drivers/mmc/core/mmc.c | 2 +-
>> include/linux/mmc/host.h | 1 +
>> 3 files changed, 4 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
>> index 6e4c55a..9b63b36 100644
>> --- a/drivers/mmc/core/host.c
>> +++ b/drivers/mmc/core/host.c
>> @@ -270,6 +270,8 @@ int mmc_of_parse(struct mmc_host *host)
>> host->caps |= MMC_CAP_HW_RESET;
>> if (of_property_read_bool(np, "cap-sdio-irq"))
>> host->caps |= MMC_CAP_SDIO_IRQ;
>> + if (of_property_read_bool(np, "mmc-broken-hpi"))
>> + host->caps |= MMC_CAP_BROKEN_HPI;
>> if (of_property_read_bool(np, "full-pwr-cycle"))
>> host->caps2 |= MMC_CAP2_FULL_PWR_CYCLE;
>> if (of_property_read_bool(np, "keep-power-in-suspend"))
>> diff --git a/drivers/mmc/core/mmc.c b/drivers/mmc/core/mmc.c
>> index 4dbe3df..9a19562 100644
>> --- a/drivers/mmc/core/mmc.c
>> +++ b/drivers/mmc/core/mmc.c
>> @@ -1592,7 +1592,7 @@ static int mmc_init_card(struct mmc_host *host, u32 ocr,
>> /*
>> * Enable HPI feature (if supported)
>> */
>> - if (card->ext_csd.hpi) {
>> + if (card->ext_csd.hpi && !(host->caps & MMC_CAP_BROKEN_HPI)) {
>> err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
>> EXT_CSD_HPI_MGMT, 1,
>> card->ext_csd.generic_cmd6_time);
>> diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
>> index 8dd4d29..20f758e 100644
>> --- a/include/linux/mmc/host.h
>> +++ b/include/linux/mmc/host.h
>> @@ -264,6 +264,7 @@ struct mmc_host {
>> #define MMC_CAP_DRIVER_TYPE_A (1 << 23) /* Host supports Driver Type A */
>> #define MMC_CAP_DRIVER_TYPE_C (1 << 24) /* Host supports Driver Type C */
>> #define MMC_CAP_DRIVER_TYPE_D (1 << 25) /* Host supports Driver Type D */
>> +#define MMC_CAP_BROKEN_HPI (1 << 29) /* Host support for HPI is broken */
>> #define MMC_CAP_CMD23 (1 << 30) /* CMD23 supported. */
>> #define MMC_CAP_HW_RESET (1 << 31) /* Hardware reset */
>>
>> --
>> 2.8.0.rc3
>>
Hi
On 04/19/2016 06:42 PM, Olliver Schinagl wrote:
> Hi Ulf,
>
> On 19-04-16 11:29, Ulf Hansson wrote:
>> On 19 April 2016 at 09:12, Olliver Schinagl <[email protected]> wrote:
>>> In patch 81f8a7be66 Hans de Goede added a patch to allow marking an mmc
>>> device as to having an broken HPI implementation. After talking some
>>> with Hans, we now think it is actually the mmc controller that can be
>>> broken and not support broken HPI's.
>> I don't want us to invent a DT binding for something you *think* is a
>> HW controller issue.
>>
>> Have you really excluded that this isn't a software issue? Me
>> personally haven't been using HPI that much so I can't really tell
>> about the code robustness from the mmc core (mmc protocol point of
>> view).
> Well this patch goes hand in hand so to speak with the broken-hpi patch introduced by him, he did most of the investigation. We just discussed how to handle it and asked me to cook up the patch.
I didn't understand why add this property. Is this same patch?
commit 81f8a7be6642b4c26ab681b2e0f4c4120a6de1b0
Author: Hans de Goede <[email protected]>
Date: Wed Apr 1 17:26:23 2015 +0200
mmc: Add support for marking hpi as broken through devicetree
The eMMC on a tablet I've will stop working / communicating as soon as
the kernel executes:
mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
EXT_CSD_HPI_MGMT, 1,
card->ext_csd.generic_cmd6_time);
There seems to be no way to reliable identify eMMC-s which have a broken
hpi implementation, but at least for eMMC's which are soldered onto a board
we can work around this by specifying that hpi is broken in devicetree.
>
> @hans, what do you think?
>>
>> Kind regards
>> Uffe
>>
>>> This patch adds a new capability, mmc-broken-hpi, which allows us to
>>> mark a broken hpi implementation on the host level.
>>>
>>> Signed-off-by: Olliver Schinagl <[email protected]>
>>> ---
>>> drivers/mmc/core/host.c | 2 ++
>>> drivers/mmc/core/mmc.c | 2 +-
>>> include/linux/mmc/host.h | 1 +
>>> 3 files changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
>>> index 6e4c55a..9b63b36 100644
>>> --- a/drivers/mmc/core/host.c
>>> +++ b/drivers/mmc/core/host.c
>>> @@ -270,6 +270,8 @@ int mmc_of_parse(struct mmc_host *host)
>>> host->caps |= MMC_CAP_HW_RESET;
>>> if (of_property_read_bool(np, "cap-sdio-irq"))
>>> host->caps |= MMC_CAP_SDIO_IRQ;
>>> + if (of_property_read_bool(np, "mmc-broken-hpi"))
>>> + host->caps |= MMC_CAP_BROKEN_HPI;
>>> if (of_property_read_bool(np, "full-pwr-cycle"))
>>> host->caps2 |= MMC_CAP2_FULL_PWR_CYCLE;
>>> if (of_property_read_bool(np, "keep-power-in-suspend"))
>>> diff --git a/drivers/mmc/core/mmc.c b/drivers/mmc/core/mmc.c
>>> index 4dbe3df..9a19562 100644
>>> --- a/drivers/mmc/core/mmc.c
>>> +++ b/drivers/mmc/core/mmc.c
>>> @@ -1592,7 +1592,7 @@ static int mmc_init_card(struct mmc_host *host, u32 ocr,
>>> /*
>>> * Enable HPI feature (if supported)
>>> */
>>> - if (card->ext_csd.hpi) {
>>> + if (card->ext_csd.hpi && !(host->caps & MMC_CAP_BROKEN_HPI)) {
>>> err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
>>> EXT_CSD_HPI_MGMT, 1,
>>> card->ext_csd.generic_cmd6_time);
>>> diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
>>> index 8dd4d29..20f758e 100644
>>> --- a/include/linux/mmc/host.h
>>> +++ b/include/linux/mmc/host.h
>>> @@ -264,6 +264,7 @@ struct mmc_host {
>>> #define MMC_CAP_DRIVER_TYPE_A (1 << 23) /* Host supports Driver Type A */
>>> #define MMC_CAP_DRIVER_TYPE_C (1 << 24) /* Host supports Driver Type C */
>>> #define MMC_CAP_DRIVER_TYPE_D (1 << 25) /* Host supports Driver Type D */
>>> +#define MMC_CAP_BROKEN_HPI (1 << 29) /* Host support for HPI is broken */
>>> #define MMC_CAP_CMD23 (1 << 30) /* CMD23 supported. */
>>> #define MMC_CAP_HW_RESET (1 << 31) /* Hardware reset */
>>>
>>> --
>>> 2.8.0.rc3
>>>
>
>
>
On 19 April 2016 at 11:42, Olliver Schinagl <[email protected]> wrote:
> Hi Ulf,
>
> On 19-04-16 11:29, Ulf Hansson wrote:
>>
>> On 19 April 2016 at 09:12, Olliver Schinagl <[email protected]> wrote:
>>>
>>> In patch 81f8a7be66 Hans de Goede added a patch to allow marking an mmc
>>> device as to having an broken HPI implementation. After talking some
>>> with Hans, we now think it is actually the mmc controller that can be
>>> broken and not support broken HPI's.
>>
>> I don't want us to invent a DT binding for something you *think* is a
>> HW controller issue.
>>
>> Have you really excluded that this isn't a software issue? Me
>> personally haven't been using HPI that much so I can't really tell
>> about the code robustness from the mmc core (mmc protocol point of
>> view).
>
> Well this patch goes hand in hand so to speak with the broken-hpi patch
> introduced by him, he did most of the investigation. We just discussed how
> to handle it and asked me to cook up the patch.
Well, my point is that it's more understandable about having a broken
HPI implementation for eMMC cards, but for host controllers I am not
so sure.
I don't think there is an electrical change required by the host
controller to support HPI, is just like any other command, right?
Unless I am missing something, of course.
Kind regards
Uffe
Hello.
On 4/19/2016 10:12 AM, Olliver Schinagl wrote:
> In patch 81f8a7be66 Hans de Goede added a patch to allow marking an mmc
scripts/checkpatch.pl now enforces certain commit citing style, the commit
summary should be specified too.
> device as to having an broken HPI implementation. After talking some
> with Hans, we now think it is actually the mmc controller that can be
> broken and not support broken HPI's.
>
> This patch adds a new capability, mmc-broken-hpi, which allows us to
> mark a broken hpi implementation on the host level.
>
> Signed-off-by: Olliver Schinagl <[email protected]>
[...]
MBR, Sergei
Hi,
On 19-04-16 11:42, Olliver Schinagl wrote:
> Hi Ulf,
>
> On 19-04-16 11:29, Ulf Hansson wrote:
>> On 19 April 2016 at 09:12, Olliver Schinagl <[email protected]> wrote:
>>> In patch 81f8a7be66 Hans de Goede added a patch to allow marking an mmc
>>> device as to having an broken HPI implementation. After talking some
>>> with Hans, we now think it is actually the mmc controller that can be
>>> broken and not support broken HPI's.
>>
>> I don't want us to invent a DT binding for something you *think* is a
>> HW controller issue.
I agree we should not add a dt flag for this, we can simply set the
flag in the host driver if we believe this is a host issue.
>> Have you really excluded that this isn't a software issue? Me
>> personally haven't been using HPI that much so I can't really tell
>> about the code robustness from the mmc core (mmc protocol point of
>> view).
> Well this patch goes hand in hand so to speak with the broken-hpi patch introduced by him, he did most of the investigation. We just discussed how to handle it and asked me to cook up the patch.
>
> @hans, what do you think?
When we discussed this a while back we had a pretty small sample
set of sunxi boards with emmc, and it seemed that hpi was broken
on all of them. But recently we've been seeing eMMC-s on a lot more
boards and they all work fine, except the one on my Utoo A13 tablets,
and the one you are using, so this does really seem be an eMMC
specific problem. That or it is a problem with the host on sun4i/sun5i/sun7i
which is not present on sun6i, sun8i and later ...
But given how rare eMMC-s are on sun4i/sun5i/sun7i I think the current
solution where we set a flag on the emmc dt node rather then on the
host node / in the host driver is fine.
Taking your case into account too, that will bring us up to 2 cases
where we set the broken-hpi flag on the emmc node, which does not
really seem like a number to worry about.
TL;DR: Thanks for writing this patch set, but given recent developments
I believe that it is best to keep handling broken-hpi as we are doing
in current kernels and no changes are necessary.
Regards,
Hans
>>
>> Kind regards
>> Uffe
>>
>>> This patch adds a new capability, mmc-broken-hpi, which allows us to
>>> mark a broken hpi implementation on the host level.
>>>
>>> Signed-off-by: Olliver Schinagl <[email protected]>
>>> ---
>>> drivers/mmc/core/host.c | 2 ++
>>> drivers/mmc/core/mmc.c | 2 +-
>>> include/linux/mmc/host.h | 1 +
>>> 3 files changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
>>> index 6e4c55a..9b63b36 100644
>>> --- a/drivers/mmc/core/host.c
>>> +++ b/drivers/mmc/core/host.c
>>> @@ -270,6 +270,8 @@ int mmc_of_parse(struct mmc_host *host)
>>> host->caps |= MMC_CAP_HW_RESET;
>>> if (of_property_read_bool(np, "cap-sdio-irq"))
>>> host->caps |= MMC_CAP_SDIO_IRQ;
>>> + if (of_property_read_bool(np, "mmc-broken-hpi"))
>>> + host->caps |= MMC_CAP_BROKEN_HPI;
>>> if (of_property_read_bool(np, "full-pwr-cycle"))
>>> host->caps2 |= MMC_CAP2_FULL_PWR_CYCLE;
>>> if (of_property_read_bool(np, "keep-power-in-suspend"))
>>> diff --git a/drivers/mmc/core/mmc.c b/drivers/mmc/core/mmc.c
>>> index 4dbe3df..9a19562 100644
>>> --- a/drivers/mmc/core/mmc.c
>>> +++ b/drivers/mmc/core/mmc.c
>>> @@ -1592,7 +1592,7 @@ static int mmc_init_card(struct mmc_host *host, u32 ocr,
>>> /*
>>> * Enable HPI feature (if supported)
>>> */
>>> - if (card->ext_csd.hpi) {
>>> + if (card->ext_csd.hpi && !(host->caps & MMC_CAP_BROKEN_HPI)) {
>>> err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
>>> EXT_CSD_HPI_MGMT, 1,
>>> card->ext_csd.generic_cmd6_time);
>>> diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
>>> index 8dd4d29..20f758e 100644
>>> --- a/include/linux/mmc/host.h
>>> +++ b/include/linux/mmc/host.h
>>> @@ -264,6 +264,7 @@ struct mmc_host {
>>> #define MMC_CAP_DRIVER_TYPE_A (1 << 23) /* Host supports Driver Type A */
>>> #define MMC_CAP_DRIVER_TYPE_C (1 << 24) /* Host supports Driver Type C */
>>> #define MMC_CAP_DRIVER_TYPE_D (1 << 25) /* Host supports Driver Type D */
>>> +#define MMC_CAP_BROKEN_HPI (1 << 29) /* Host support for HPI is broken */
>>> #define MMC_CAP_CMD23 (1 << 30) /* CMD23 supported. */
>>> #define MMC_CAP_HW_RESET (1 << 31) /* Hardware reset */
>>>
>>> --
>>> 2.8.0.rc3
>>>
>
Hey Jaehoon,
On 19-04-16 11:49, Jaehoon Chung wrote:
> Hi
>
> On 04/19/2016 06:42 PM, Olliver Schinagl wrote:
>> Hi Ulf,
>>
>> On 19-04-16 11:29, Ulf Hansson wrote:
>>> On 19 April 2016 at 09:12, Olliver Schinagl <[email protected]> wrote:
>>>> In patch 81f8a7be66 Hans de Goede added a patch to allow marking an mmc
>>>> device as to having an broken HPI implementation. After talking some
>>>> with Hans, we now think it is actually the mmc controller that can be
>>>> broken and not support broken HPI's.
>>> I don't want us to invent a DT binding for something you *think* is a
>>> HW controller issue.
>>>
>>> Have you really excluded that this isn't a software issue? Me
>>> personally haven't been using HPI that much so I can't really tell
>>> about the code robustness from the mmc core (mmc protocol point of
>>> view).
>> Well this patch goes hand in hand so to speak with the broken-hpi patch introduced by him, he did most of the investigation. We just discussed how to handle it and asked me to cook up the patch.
> I didn't understand why add this property. Is this same patch?
>
> commit 81f8a7be6642b4c26ab681b2e0f4c4120a6de1b0
> Author: Hans de Goede <[email protected]>
> Date: Wed Apr 1 17:26:23 2015 +0200
>
> mmc: Add support for marking hpi as broken through devicetree
>
> The eMMC on a tablet I've will stop working / communicating as soon as
> the kernel executes:
>
> mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
> EXT_CSD_HPI_MGMT, 1,
> card->ext_csd.generic_cmd6_time);
>
> There seems to be no way to reliable identify eMMC-s which have a broken
> hpi implementation, but at least for eMMC's which are soldered onto a board
> we can work around this by specifying that hpi is broken in devicetree.
Similar, this does it on the card level, e.g. you mark a card as having
a broken HPI implementation.
The truth is, the card may not be broken at all, but the mmc controller
so we add a property on the controller level.
What Ulf is pondering, if this is not a software stack problem, rather
then a hardware implementation, which I don't know.
>> @hans, what do you think?
>>> Kind regards
>>> Uffe
>>>
>>>> This patch adds a new capability, mmc-broken-hpi, which allows us to
>>>> mark a broken hpi implementation on the host level.
>>>>
>>>> Signed-off-by: Olliver Schinagl <[email protected]>
>>>> ---
>>>> drivers/mmc/core/host.c | 2 ++
>>>> drivers/mmc/core/mmc.c | 2 +-
>>>> include/linux/mmc/host.h | 1 +
>>>> 3 files changed, 4 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
>>>> index 6e4c55a..9b63b36 100644
>>>> --- a/drivers/mmc/core/host.c
>>>> +++ b/drivers/mmc/core/host.c
>>>> @@ -270,6 +270,8 @@ int mmc_of_parse(struct mmc_host *host)
>>>> host->caps |= MMC_CAP_HW_RESET;
>>>> if (of_property_read_bool(np, "cap-sdio-irq"))
>>>> host->caps |= MMC_CAP_SDIO_IRQ;
>>>> + if (of_property_read_bool(np, "mmc-broken-hpi"))
>>>> + host->caps |= MMC_CAP_BROKEN_HPI;
>>>> if (of_property_read_bool(np, "full-pwr-cycle"))
>>>> host->caps2 |= MMC_CAP2_FULL_PWR_CYCLE;
>>>> if (of_property_read_bool(np, "keep-power-in-suspend"))
>>>> diff --git a/drivers/mmc/core/mmc.c b/drivers/mmc/core/mmc.c
>>>> index 4dbe3df..9a19562 100644
>>>> --- a/drivers/mmc/core/mmc.c
>>>> +++ b/drivers/mmc/core/mmc.c
>>>> @@ -1592,7 +1592,7 @@ static int mmc_init_card(struct mmc_host *host, u32 ocr,
>>>> /*
>>>> * Enable HPI feature (if supported)
>>>> */
>>>> - if (card->ext_csd.hpi) {
>>>> + if (card->ext_csd.hpi && !(host->caps & MMC_CAP_BROKEN_HPI)) {
>>>> err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
>>>> EXT_CSD_HPI_MGMT, 1,
>>>> card->ext_csd.generic_cmd6_time);
>>>> diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
>>>> index 8dd4d29..20f758e 100644
>>>> --- a/include/linux/mmc/host.h
>>>> +++ b/include/linux/mmc/host.h
>>>> @@ -264,6 +264,7 @@ struct mmc_host {
>>>> #define MMC_CAP_DRIVER_TYPE_A (1 << 23) /* Host supports Driver Type A */
>>>> #define MMC_CAP_DRIVER_TYPE_C (1 << 24) /* Host supports Driver Type C */
>>>> #define MMC_CAP_DRIVER_TYPE_D (1 << 25) /* Host supports Driver Type D */
>>>> +#define MMC_CAP_BROKEN_HPI (1 << 29) /* Host support for HPI is broken */
>>>> #define MMC_CAP_CMD23 (1 << 30) /* CMD23 supported. */
>>>> #define MMC_CAP_HW_RESET (1 << 31) /* Hardware reset */
>>>>
>>>> --
>>>> 2.8.0.rc3
>>>>
>>
>>
Hey Sergei,
On 19-04-16 12:52, Sergei Shtylyov wrote:
> Hello.
>
> On 4/19/2016 10:12 AM, Olliver Schinagl wrote:
>
>> In patch 81f8a7be66 Hans de Goede added a patch to allow marking an mmc
>
> scripts/checkpatch.pl now enforces certain commit citing style, the
> commit summary should be specified too.
I used checkpatch from 4.6.0-rc2(or3) and did not notice it. But I'll
add it to v2.
>
>> device as to having an broken HPI implementation. After talking some
>> with Hans, we now think it is actually the mmc controller that can be
>> broken and not support broken HPI's.
>>
>> This patch adds a new capability, mmc-broken-hpi, which allows us to
>> mark a broken hpi implementation on the host level.
>>
>> Signed-off-by: Olliver Schinagl <[email protected]>
> [...]
>
> MBR, Sergei
>
Hi Olliver,
On 04/19/2016 09:20 PM, Olliver Schinagl wrote:
> Hey Jaehoon,
>
> On 19-04-16 11:49, Jaehoon Chung wrote:
>> Hi
>>
>> On 04/19/2016 06:42 PM, Olliver Schinagl wrote:
>>> Hi Ulf,
>>>
>>> On 19-04-16 11:29, Ulf Hansson wrote:
>>>> On 19 April 2016 at 09:12, Olliver Schinagl <[email protected]> wrote:
>>>>> In patch 81f8a7be66 Hans de Goede added a patch to allow marking an mmc
>>>>> device as to having an broken HPI implementation. After talking some
>>>>> with Hans, we now think it is actually the mmc controller that can be
>>>>> broken and not support broken HPI's.
>>>> I don't want us to invent a DT binding for something you *think* is a
>>>> HW controller issue.
>>>>
>>>> Have you really excluded that this isn't a software issue? Me
>>>> personally haven't been using HPI that much so I can't really tell
>>>> about the code robustness from the mmc core (mmc protocol point of
>>>> view).
>>> Well this patch goes hand in hand so to speak with the broken-hpi patch introduced by him, he did most of the investigation. We just discussed how to handle it and asked me to cook up the patch.
>> I didn't understand why add this property. Is this same patch?
>>
>> commit 81f8a7be6642b4c26ab681b2e0f4c4120a6de1b0
>> Author: Hans de Goede <[email protected]>
>> Date: Wed Apr 1 17:26:23 2015 +0200
>>
>> mmc: Add support for marking hpi as broken through devicetree
>>
>> The eMMC on a tablet I've will stop working / communicating as soon as
>> the kernel executes:
>>
>> mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
>> EXT_CSD_HPI_MGMT, 1,
>> card->ext_csd.generic_cmd6_time);
>>
>> There seems to be no way to reliable identify eMMC-s which have a broken
>> hpi implementation, but at least for eMMC's which are soldered onto a board
>> we can work around this by specifying that hpi is broken in devicetree.
> Similar, this does it on the card level, e.g. you mark a card as having a broken HPI implementation.
>
> The truth is, the card may not be broken at all, but the mmc controller so we add a property on the controller level.
>
> What Ulf is pondering, if this is not a software stack problem, rather then a hardware implementation, which I don't know.
Hmm..problem is mmc controller side..it's strange.
How did you know this problem is on mmc controller side?
Well, i should be also missing something for HPI feature..but it's not relevant to mmc controller.
Ok. i will check more..
Best Regards,
Jaehoon Chung
>>> @hans, what do you think?
>>>> Kind regards
>>>> Uffe
>>>>
>>>>> This patch adds a new capability, mmc-broken-hpi, which allows us to
>>>>> mark a broken hpi implementation on the host level.
>>>>>
>>>>> Signed-off-by: Olliver Schinagl <[email protected]>
>>>>> ---
>>>>> drivers/mmc/core/host.c | 2 ++
>>>>> drivers/mmc/core/mmc.c | 2 +-
>>>>> include/linux/mmc/host.h | 1 +
>>>>> 3 files changed, 4 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
>>>>> index 6e4c55a..9b63b36 100644
>>>>> --- a/drivers/mmc/core/host.c
>>>>> +++ b/drivers/mmc/core/host.c
>>>>> @@ -270,6 +270,8 @@ int mmc_of_parse(struct mmc_host *host)
>>>>> host->caps |= MMC_CAP_HW_RESET;
>>>>> if (of_property_read_bool(np, "cap-sdio-irq"))
>>>>> host->caps |= MMC_CAP_SDIO_IRQ;
>>>>> + if (of_property_read_bool(np, "mmc-broken-hpi"))
>>>>> + host->caps |= MMC_CAP_BROKEN_HPI;
>>>>> if (of_property_read_bool(np, "full-pwr-cycle"))
>>>>> host->caps2 |= MMC_CAP2_FULL_PWR_CYCLE;
>>>>> if (of_property_read_bool(np, "keep-power-in-suspend"))
>>>>> diff --git a/drivers/mmc/core/mmc.c b/drivers/mmc/core/mmc.c
>>>>> index 4dbe3df..9a19562 100644
>>>>> --- a/drivers/mmc/core/mmc.c
>>>>> +++ b/drivers/mmc/core/mmc.c
>>>>> @@ -1592,7 +1592,7 @@ static int mmc_init_card(struct mmc_host *host, u32 ocr,
>>>>> /*
>>>>> * Enable HPI feature (if supported)
>>>>> */
>>>>> - if (card->ext_csd.hpi) {
>>>>> + if (card->ext_csd.hpi && !(host->caps & MMC_CAP_BROKEN_HPI)) {
>>>>> err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
>>>>> EXT_CSD_HPI_MGMT, 1,
>>>>> card->ext_csd.generic_cmd6_time);
>>>>> diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
>>>>> index 8dd4d29..20f758e 100644
>>>>> --- a/include/linux/mmc/host.h
>>>>> +++ b/include/linux/mmc/host.h
>>>>> @@ -264,6 +264,7 @@ struct mmc_host {
>>>>> #define MMC_CAP_DRIVER_TYPE_A (1 << 23) /* Host supports Driver Type A */
>>>>> #define MMC_CAP_DRIVER_TYPE_C (1 << 24) /* Host supports Driver Type C */
>>>>> #define MMC_CAP_DRIVER_TYPE_D (1 << 25) /* Host supports Driver Type D */
>>>>> +#define MMC_CAP_BROKEN_HPI (1 << 29) /* Host support for HPI is broken */
>>>>> #define MMC_CAP_CMD23 (1 << 30) /* CMD23 supported. */
>>>>> #define MMC_CAP_HW_RESET (1 << 31) /* Hardware reset */
>>>>>
>>>>> --
>>>>> 2.8.0.rc3
>>>>>
>>>
>>>
>
>
>
Hey Hans,
On 19-04-16 13:22, Hans de Goede wrote:
> Hi,
>
> On 19-04-16 11:42, Olliver Schinagl wrote:
>> Hi Ulf,
>>
>> On 19-04-16 11:29, Ulf Hansson wrote:
>>> On 19 April 2016 at 09:12, Olliver Schinagl <[email protected]> wrote:
>>>> In patch 81f8a7be66 Hans de Goede added a patch to allow marking an
>>>> mmc
>>>> device as to having an broken HPI implementation. After talking some
>>>> with Hans, we now think it is actually the mmc controller that can be
>>>> broken and not support broken HPI's.
> >>
>>> I don't want us to invent a DT binding for something you *think* is a
>>> HW controller issue.
>
> I agree we should not add a dt flag for this, we can simply set the
> flag in the host driver if we believe this is a host issue.
Okay, I can remove the dt flag
>
>>> Have you really excluded that this isn't a software issue? Me
>>> personally haven't been using HPI that much so I can't really tell
>>> about the code robustness from the mmc core (mmc protocol point of
>>> view).
>> Well this patch goes hand in hand so to speak with the broken-hpi
>> patch introduced by him, he did most of the investigation. We just
>> discussed how to handle it and asked me to cook up the patch.
>>
>> @hans, what do you think?
>
> When we discussed this a while back we had a pretty small sample
> set of sunxi boards with emmc, and it seemed that hpi was broken
> on all of them. But recently we've been seeing eMMC-s on a lot more
> boards and they all work fine, except the one on my Utoo A13 tablets,
> and the one you are using, so this does really seem be an eMMC
> specific problem. That or it is a problem with the host on
> sun4i/sun5i/sun7i
> which is not present on sun6i, sun8i and later ...
Well I think we still have a very small sample size, the sun4i, sun5i
and sun7i boards (all using the same mmc controller afaik) have the
broken-hpi set, the sun6i and sun8i seem to be working fine (different
mmc controller?).
I'm not so sure it is an eMMC specific problem though. The module we're
using is a high-end micron part, industrial grade. Granted, it could be
still broken there, but I find it less likely. Micron has an
interessting technical document:
"TN-52-05 e.EMMC Linux Enablement"
talking specifically about HPI on eMMC.
It also mentions HPI is a mmc/jedic 4.41 thing, afaik our controller is
only 4.3 (which might not even matter according to Ulf if it is just a
command).
The datasheet of the chip I use, MTFC4GACAANA, also mentions explicitly
that it supports HPI.
Granted, it could still be broken, but I have doubts.
>
> But given how rare eMMC-s are on sun4i/sun5i/sun7i I think the current
> solution where we set a flag on the emmc dt node rather then on the
> host node / in the host driver is fine.
Yeah it is very limited, that is true, and I suppose I can live with that.
>
> Taking your case into account too, that will bring us up to 2 cases
> where we set the broken-hpi flag on the emmc node, which does not
> really seem like a number to worry about.
Actually, 4 :), there are 3 sun5i (tablets?) devices that suffer from
this and my device now. The sun6i and sun8i devices are only 2 (the
sinlinx devices in the current kernel) that a very quick grep (mmc-card
) showed. Grepping for non-removable yielded a bit more, like the chip
sun5i-like device with a "non-removable" mmc, not sure what to make of
that though.
>
> TL;DR: Thanks for writing this patch set, but given recent developments
> I believe that it is best to keep handling broken-hpi as we are doing
> in current kernels and no changes are necessary.
I would still recommend to add the capability and raise the flag for the
sun[457]i devices though, as my gut thinks it's a problem with the sunxi
IP there. But with the emmc-card level work around, it does solve/fix
it, so what is the best way?
Olliver
>
> Regards,
>
> Hans
>
>
>>>
>>> Kind regards
>>> Uffe
>>>
>>>> This patch adds a new capability, mmc-broken-hpi, which allows us to
>>>> mark a broken hpi implementation on the host level.
>>>>
>>>> Signed-off-by: Olliver Schinagl <[email protected]>
>>>> ---
>>>> drivers/mmc/core/host.c | 2 ++
>>>> drivers/mmc/core/mmc.c | 2 +-
>>>> include/linux/mmc/host.h | 1 +
>>>> 3 files changed, 4 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
>>>> index 6e4c55a..9b63b36 100644
>>>> --- a/drivers/mmc/core/host.c
>>>> +++ b/drivers/mmc/core/host.c
>>>> @@ -270,6 +270,8 @@ int mmc_of_parse(struct mmc_host *host)
>>>> host->caps |= MMC_CAP_HW_RESET;
>>>> if (of_property_read_bool(np, "cap-sdio-irq"))
>>>> host->caps |= MMC_CAP_SDIO_IRQ;
>>>> + if (of_property_read_bool(np, "mmc-broken-hpi"))
>>>> + host->caps |= MMC_CAP_BROKEN_HPI;
>>>> if (of_property_read_bool(np, "full-pwr-cycle"))
>>>> host->caps2 |= MMC_CAP2_FULL_PWR_CYCLE;
>>>> if (of_property_read_bool(np, "keep-power-in-suspend"))
>>>> diff --git a/drivers/mmc/core/mmc.c b/drivers/mmc/core/mmc.c
>>>> index 4dbe3df..9a19562 100644
>>>> --- a/drivers/mmc/core/mmc.c
>>>> +++ b/drivers/mmc/core/mmc.c
>>>> @@ -1592,7 +1592,7 @@ static int mmc_init_card(struct mmc_host
>>>> *host, u32 ocr,
>>>> /*
>>>> * Enable HPI feature (if supported)
>>>> */
>>>> - if (card->ext_csd.hpi) {
>>>> + if (card->ext_csd.hpi && !(host->caps & MMC_CAP_BROKEN_HPI)) {
>>>> err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
>>>> EXT_CSD_HPI_MGMT, 1,
>>>> card->ext_csd.generic_cmd6_time);
>>>> diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
>>>> index 8dd4d29..20f758e 100644
>>>> --- a/include/linux/mmc/host.h
>>>> +++ b/include/linux/mmc/host.h
>>>> @@ -264,6 +264,7 @@ struct mmc_host {
>>>> #define MMC_CAP_DRIVER_TYPE_A (1 << 23) /* Host supports
>>>> Driver Type A */
>>>> #define MMC_CAP_DRIVER_TYPE_C (1 << 24) /* Host supports
>>>> Driver Type C */
>>>> #define MMC_CAP_DRIVER_TYPE_D (1 << 25) /* Host supports
>>>> Driver Type D */
>>>> +#define MMC_CAP_BROKEN_HPI (1 << 29) /* Host support for
>>>> HPI is broken */
>>>> #define MMC_CAP_CMD23 (1 << 30) /* CMD23
>>>> supported. */
>>>> #define MMC_CAP_HW_RESET (1 << 31) /* Hardware reset */
>>>>
>>>> --
>>>> 2.8.0.rc3
>>>>
>>
On 4/19/2016 3:21 PM, Olliver Schinagl wrote:
>>> In patch 81f8a7be66 Hans de Goede added a patch to allow marking an mmc
>>
>> scripts/checkpatch.pl now enforces certain commit citing style, the
>> commit summary should be specified too.
> I used checkpatch from 4.6.0-rc2(or3) and did not notice it. But I'll add it
> to v2.
Maybe it just didn't recognize your commit citing? Anyway, the format is
the same as for the "Fixes:" tag.
>>> device as to having an broken HPI implementation. After talking some
>>> with Hans, we now think it is actually the mmc controller that can be
>>> broken and not support broken HPI's.
>>>
>>> This patch adds a new capability, mmc-broken-hpi, which allows us to
>>> mark a broken hpi implementation on the host level.
>>>
>>> Signed-off-by: Olliver Schinagl <[email protected]>
>> [...]
MBR, Sergei
[...]
> Well I think we still have a very small sample size, the sun4i, sun5i and
> sun7i boards (all using the same mmc controller afaik) have the broken-hpi
> set, the sun6i and sun8i seem to be working fine (different mmc
> controller?).
> I'm not so sure it is an eMMC specific problem though. The module we're
> using is a high-end micron part, industrial grade. Granted, it could be
> still broken there, but I find it less likely. Micron has an interessting
> technical document:
> "TN-52-05 e.EMMC Linux Enablement"
> talking specifically about HPI on eMMC.
> It also mentions HPI is a mmc/jedic 4.41 thing, afaik our controller is only
> 4.3 (which might not even matter according to Ulf if it is just a command).
>
> The datasheet of the chip I use, MTFC4GACAANA, also mentions explicitly that
> it supports HPI.
>
> Granted, it could still be broken, but I have doubts.
Perhaps the same eMMC is used without errors on other controllers?
>
>>
>> But given how rare eMMC-s are on sun4i/sun5i/sun7i I think the current
>> solution where we set a flag on the emmc dt node rather then on the
>> host node / in the host driver is fine.
>
> Yeah it is very limited, that is true, and I suppose I can live with that.
>>
>>
>> Taking your case into account too, that will bring us up to 2 cases
>> where we set the broken-hpi flag on the emmc node, which does not
>> really seem like a number to worry about.
>
> Actually, 4 :), there are 3 sun5i (tablets?) devices that suffer from this
> and my device now. The sun6i and sun8i devices are only 2 (the sinlinx
> devices in the current kernel) that a very quick grep (mmc-card ) showed.
> Grepping for non-removable yielded a bit more, like the chip sun5i-like
> device with a "non-removable" mmc, not sure what to make of that though.
>>
>>
>> TL;DR: Thanks for writing this patch set, but given recent developments
>> I believe that it is best to keep handling broken-hpi as we are doing
>> in current kernels and no changes are necessary.
>
> I would still recommend to add the capability and raise the flag for the
> sun[457]i devices though, as my gut thinks it's a problem with the sunxi IP
> there. But with the emmc-card level work around, it does solve/fix it, so
> what is the best way?
The best is clearly to make a proper debug investigation before we
decide to add a DT binding for the host.
I don't know on what level you are able to measure signals on the HW,
but if not, perhaps the newly TRACE support in the MMC core can help.
Kind regards
Uffe