[PATCH] mmc: dw_mmc: Avoid hung state if GEN_CMD transfer fails

[PATCH] mmc: dw_mmc: Avoid hung state if GEN_CMD transfer fails

[Mårten Lindahl]

If we get a data error during a block transfer command, a stop command
(CMD12) is normally initiated. But this does not work for the general
command (CMD56), but instead the action is ignored and an uninitialized
command struct is used for the stop action, with unexpected result.

Fix this by adding a check for GEN_CMD when preparing stop transmission.

Signed-off-by: Mårten Lindahl <marten.lindahl@axis.com>
---
 drivers/mmc/host/dw_mmc.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
index 6578cc64ae9e..988c32e93e03 100644
--- a/drivers/mmc/host/dw_mmc.c
+++ b/drivers/mmc/host/dw_mmc.c
@@ -335,7 +335,8 @@ static u32 dw_mci_prep_stop_abort(struct dw_mci *host, struct mmc_command *cmd)
 	    cmdr == MMC_WRITE_BLOCK ||
 	    cmdr == MMC_WRITE_MULTIPLE_BLOCK ||
 	    cmdr == MMC_SEND_TUNING_BLOCK ||
-	    cmdr == MMC_SEND_TUNING_BLOCK_HS200) {
+	    cmdr == MMC_SEND_TUNING_BLOCK_HS200 ||
+	    cmdr == MMC_GEN_CMD) {
 		stop->opcode = MMC_STOP_TRANSMISSION;
 		stop->arg = 0;
 		stop->flags = MMC_RSP_R1B | MMC_CMD_AC;
-- 
2.20.1

Re: [PATCH] mmc: dw_mmc: Avoid hung state if GEN_CMD transfer fails

[Vincent Whitchurch]

On Wed, Nov 03, 2021 at 07:27:16PM +0100, Mårten Lindahl wrote:
> If we get a data error during a block transfer command, a stop command
> (CMD12) is normally initiated. But this does not work for the general
> command (CMD56), but instead the action is ignored and an uninitialized
> command struct is used for the stop action, with unexpected result.
> 
> Fix this by adding a check for GEN_CMD when preparing stop transmission.
> 
> Signed-off-by: Mårten Lindahl <marten.lindahl@axis.com>
> ---
>  drivers/mmc/host/dw_mmc.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
> index 6578cc64ae9e..988c32e93e03 100644
> --- a/drivers/mmc/host/dw_mmc.c
> +++ b/drivers/mmc/host/dw_mmc.c
> @@ -335,7 +335,8 @@ static u32 dw_mci_prep_stop_abort(struct dw_mci *host, struct mmc_command *cmd)
>  	    cmdr == MMC_WRITE_BLOCK ||
>  	    cmdr == MMC_WRITE_MULTIPLE_BLOCK ||
>  	    cmdr == MMC_SEND_TUNING_BLOCK ||
> -	    cmdr == MMC_SEND_TUNING_BLOCK_HS200) {
> +	    cmdr == MMC_SEND_TUNING_BLOCK_HS200 ||
> +	    cmdr == MMC_GEN_CMD) {
>  		stop->opcode = MMC_STOP_TRANSMISSION;
>  		stop->arg = 0;
>  		stop->flags = MMC_RSP_R1B | MMC_CMD_AC;

While this fix looks correct for CMD56, the "Data transfer mode"
sections of the eMMC and SD specifications list several more data
commands, all of which can be aborted by CMD12, but which aren't handled
in the if above.

If I'm not mistaken, those will also result in an uninitialized stop
command being sent in the case of an error, since the driver calls
send_stop_abort() on any data error.

Is there a reason why those other commands should not be in the list
above, or should we fix this list so that CMD12 is initialized for all
data commands except SD_IO_RW_EXTENDED?

Re: [PATCH] mmc: dw_mmc: Avoid hung state if GEN_CMD transfer fails

[Marten Lindahl]

On Thu, Nov 11, 2021 at 04:40:20PM +0100, Vincent Whitchurch wrote:
> On Wed, Nov 03, 2021 at 07:27:16PM +0100, Mårten Lindahl wrote:
> > If we get a data error during a block transfer command, a stop command
> > (CMD12) is normally initiated. But this does not work for the general
> > command (CMD56), but instead the action is ignored and an uninitialized
> > command struct is used for the stop action, with unexpected result.
> > 
> > Fix this by adding a check for GEN_CMD when preparing stop transmission.
> > 
> > Signed-off-by: Mårten Lindahl <marten.lindahl@axis.com>
> > ---
> >  drivers/mmc/host/dw_mmc.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
> > index 6578cc64ae9e..988c32e93e03 100644
> > --- a/drivers/mmc/host/dw_mmc.c
> > +++ b/drivers/mmc/host/dw_mmc.c
> > @@ -335,7 +335,8 @@ static u32 dw_mci_prep_stop_abort(struct dw_mci *host, struct mmc_command *cmd)
> >  	    cmdr == MMC_WRITE_BLOCK ||
> >  	    cmdr == MMC_WRITE_MULTIPLE_BLOCK ||
> >  	    cmdr == MMC_SEND_TUNING_BLOCK ||
> > -	    cmdr == MMC_SEND_TUNING_BLOCK_HS200) {
> > +	    cmdr == MMC_SEND_TUNING_BLOCK_HS200 ||
> > +	    cmdr == MMC_GEN_CMD) {
> >  		stop->opcode = MMC_STOP_TRANSMISSION;
> >  		stop->arg = 0;
> >  		stop->flags = MMC_RSP_R1B | MMC_CMD_AC;

Hi!
> 
> While this fix looks correct for CMD56, the "Data transfer mode"
> sections of the eMMC and SD specifications list several more data
> commands, all of which can be aborted by CMD12, but which aren't handled
> in the if above.
> 
> If I'm not mistaken, those will also result in an uninitialized stop
> command being sent in the case of an error, since the driver calls
> send_stop_abort() on any data error.
> 
> Is there a reason why those other commands should not be in the list
> above, or should we fix this list so that CMD12 is initialized for all
> data commands except SD_IO_RW_EXTENDED?

I agree that there are more commands that according to the specification
should be included in the list. The most obvious ones that I identify are:
CMD26, CMD27, CMD30, CMD46, and CMD47.

But the reason why I added just CMD56 is that I have an easy way to test
this specific command, and which was how I saw the need for this fix.

My suggestion is to separate adding the other commands to one (or more)
separate patch(es), to when there are good test cases for those.

Kind regards
Mårten

Re: [PATCH] mmc: dw_mmc: Avoid hung state if GEN_CMD transfer fails

[Ulf Hansson]

On Fri, 12 Nov 2021 at 13:46, Marten Lindahl <martenli@axis.com> wrote:
>
> On Thu, Nov 11, 2021 at 04:40:20PM +0100, Vincent Whitchurch wrote:
> > On Wed, Nov 03, 2021 at 07:27:16PM +0100, Mårten Lindahl wrote:
> > > If we get a data error during a block transfer command, a stop command
> > > (CMD12) is normally initiated. But this does not work for the general
> > > command (CMD56), but instead the action is ignored and an uninitialized
> > > command struct is used for the stop action, with unexpected result.
> > >
> > > Fix this by adding a check for GEN_CMD when preparing stop transmission.
> > >
> > > Signed-off-by: Mårten Lindahl <marten.lindahl@axis.com>
> > > ---
> > >  drivers/mmc/host/dw_mmc.c | 3 ++-
> > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
> > > index 6578cc64ae9e..988c32e93e03 100644
> > > --- a/drivers/mmc/host/dw_mmc.c
> > > +++ b/drivers/mmc/host/dw_mmc.c
> > > @@ -335,7 +335,8 @@ static u32 dw_mci_prep_stop_abort(struct dw_mci *host, struct mmc_command *cmd)
> > >         cmdr == MMC_WRITE_BLOCK ||
> > >         cmdr == MMC_WRITE_MULTIPLE_BLOCK ||
> > >         cmdr == MMC_SEND_TUNING_BLOCK ||
> > > -       cmdr == MMC_SEND_TUNING_BLOCK_HS200) {
> > > +       cmdr == MMC_SEND_TUNING_BLOCK_HS200 ||
> > > +       cmdr == MMC_GEN_CMD) {
> > >             stop->opcode = MMC_STOP_TRANSMISSION;
> > >             stop->arg = 0;
> > >             stop->flags = MMC_RSP_R1B | MMC_CMD_AC;
>
> Hi!
> >
> > While this fix looks correct for CMD56, the "Data transfer mode"
> > sections of the eMMC and SD specifications list several more data
> > commands, all of which can be aborted by CMD12, but which aren't handled
> > in the if above.
> >
> > If I'm not mistaken, those will also result in an uninitialized stop
> > command being sent in the case of an error, since the driver calls
> > send_stop_abort() on any data error.
> >
> > Is there a reason why those other commands should not be in the list
> > above, or should we fix this list so that CMD12 is initialized for all
> > data commands except SD_IO_RW_EXTENDED?
>
> I agree that there are more commands that according to the specification
> should be included in the list. The most obvious ones that I identify are:
> CMD26, CMD27, CMD30, CMD46, and CMD47.
>
> But the reason why I added just CMD56 is that I have an easy way to test
> this specific command, and which was how I saw the need for this fix.
>
> My suggestion is to separate adding the other commands to one (or more)
> separate patch(es), to when there are good test cases for those.

I agree with both of you, it would be nice to make a change that takes
care of all the relevant commands at once, rather than having to come
back to this later again.

On the other hand, as it's not always straightforward to test the
other commands, I am fine with this as well.

So, applied for next, thanks!

Kind regards
Uffe