[PATCH] ARM: Tegra: DMA: Fail safe if initialization fails

[PATCH] ARM: Tegra: DMA: Fail safe if initialization fails

[Stephen Warren]

tegra_dma_init currently simply bails out early if any initialization fails.
This skips various data-structure initialization. In turn, this means that
tegra_dma_allocate_channel can still hand out channels. In this case, when
tegra_dma_free_channel is called, which calls tegra_dma_cancel, the walking
on ch->list will OOPS since the list's next/prev pointers may still be
NULL.

To solve this:
* Mark all possible channels as in-use before doing anything else in init.
* Only mark a channel as free once all channel-related initialization has
  completed.

This prevents allocate_channel from handing out uninitialized channels.

There is still one small hole; allocate_channel can't check the usage array
for the shared channel, since this channel is permanently marked in-use.
This could be solved using an explicit "init OK" flag that allocate_channel
could check.

Signed-off-by: Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
---
 arch/arm/mach-tegra/dma.c |   20 ++++++++------------
 1 files changed, 8 insertions(+), 12 deletions(-)

diff --git a/arch/arm/mach-tegra/dma.c b/arch/arm/mach-tegra/dma.c
index 2d720f2..79765ca 100644
--- a/arch/arm/mach-tegra/dma.c
+++ b/arch/arm/mach-tegra/dma.c
@@ -678,6 +678,12 @@ int __init tegra_dma_init(void)
 	void __iomem *addr;
 	struct clk *c;
 
+	memset(channel_usage, 0, sizeof(channel_usage));
+	memset(dma_channels, 0, sizeof(dma_channels));
+
+	for (i = 0; i < NV_DMA_MAX_CHANNELS; i++)
+		__set_bit(i, channel_usage);
+
 	c = clk_get_sys("tegra-dma", NULL);
 	if (IS_ERR(c)) {
 		pr_err("Unable to get clock for APB DMA\n");
@@ -696,18 +702,9 @@ int __init tegra_dma_init(void)
 	writel(0xFFFFFFFFul >> (31 - TEGRA_SYSTEM_DMA_CH_MAX),
 	       addr + APB_DMA_IRQ_MASK_SET);
 
-	memset(channel_usage, 0, sizeof(channel_usage));
-	memset(dma_channels, 0, sizeof(dma_channels));
-
-	/* Reserve all the channels we are not supposed to touch */
-	for (i = 0; i < TEGRA_SYSTEM_DMA_CH_MIN; i++)
-		__set_bit(i, channel_usage);
-
 	for (i = TEGRA_SYSTEM_DMA_CH_MIN; i <= TEGRA_SYSTEM_DMA_CH_MAX; i++) {
 		struct tegra_dma_channel *ch = &dma_channels[i];
 
-		__clear_bit(i, channel_usage);
-
 		ch->id = i;
 		snprintf(ch->name, TEGRA_DMA_NAME_SIZE, "dma_channel_%d", i);
 
@@ -726,13 +723,12 @@ int __init tegra_dma_init(void)
 			goto fail;
 		}
 		ch->irq = irq;
+
+		__clear_bit(i, channel_usage);
 	}
 	/* mark the shared channel allocated */
 	__set_bit(TEGRA_SYSTEM_DMA_CH_MIN, channel_usage);
 
-	for (i = TEGRA_SYSTEM_DMA_CH_MAX+1; i < NV_DMA_MAX_CHANNELS; i++)
-		__set_bit(i, channel_usage);
-
 	return ret;
 fail:
 	writel(0, addr + APB_DMA_GEN);
-- 
1.7.1

--
To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH] ARM: Tegra: DMA: Fail safe if initialization fails

[Stephen Warren]

tegra_dma_init currently simply bails out early if any initialization fails.
This skips various data-structure initialization. In turn, this means that
tegra_dma_allocate_channel can still hand out channels. In this case, when
tegra_dma_free_channel is called, which calls tegra_dma_cancel, the walking
on ch->list will OOPS since the list's next/prev pointers may still be
NULL.

To solve this:
* Mark all possible channels as in-use before doing anything else in init.
* Only mark a channel as free once all channel-related initialization has
  completed.

This prevents allocate_channel from handing out uninitialized channels.

There is still one small hole; allocate_channel can't check the usage array
for the shared channel, since this channel is permanently marked in-use.
This could be solved using an explicit "init OK" flag that allocate_channel
could check.

Signed-off-by: Stephen Warren <swarren@nvidia.com>
---
 arch/arm/mach-tegra/dma.c |   20 ++++++++------------
 1 files changed, 8 insertions(+), 12 deletions(-)

diff --git a/arch/arm/mach-tegra/dma.c b/arch/arm/mach-tegra/dma.c
index 2d720f2..79765ca 100644
--- a/arch/arm/mach-tegra/dma.c
+++ b/arch/arm/mach-tegra/dma.c
@@ -678,6 +678,12 @@ int __init tegra_dma_init(void)
 	void __iomem *addr;
 	struct clk *c;
 
+	memset(channel_usage, 0, sizeof(channel_usage));
+	memset(dma_channels, 0, sizeof(dma_channels));
+
+	for (i = 0; i < NV_DMA_MAX_CHANNELS; i++)
+		__set_bit(i, channel_usage);
+
 	c = clk_get_sys("tegra-dma", NULL);
 	if (IS_ERR(c)) {
 		pr_err("Unable to get clock for APB DMA\n");
@@ -696,18 +702,9 @@ int __init tegra_dma_init(void)
 	writel(0xFFFFFFFFul >> (31 - TEGRA_SYSTEM_DMA_CH_MAX),
 	       addr + APB_DMA_IRQ_MASK_SET);
 
-	memset(channel_usage, 0, sizeof(channel_usage));
-	memset(dma_channels, 0, sizeof(dma_channels));
-
-	/* Reserve all the channels we are not supposed to touch */
-	for (i = 0; i < TEGRA_SYSTEM_DMA_CH_MIN; i++)
-		__set_bit(i, channel_usage);
-
 	for (i = TEGRA_SYSTEM_DMA_CH_MIN; i <= TEGRA_SYSTEM_DMA_CH_MAX; i++) {
 		struct tegra_dma_channel *ch = &dma_channels[i];
 
-		__clear_bit(i, channel_usage);
-
 		ch->id = i;
 		snprintf(ch->name, TEGRA_DMA_NAME_SIZE, "dma_channel_%d", i);
 
@@ -726,13 +723,12 @@ int __init tegra_dma_init(void)
 			goto fail;
 		}
 		ch->irq = irq;
+
+		__clear_bit(i, channel_usage);
 	}
 	/* mark the shared channel allocated */
 	__set_bit(TEGRA_SYSTEM_DMA_CH_MIN, channel_usage);
 
-	for (i = TEGRA_SYSTEM_DMA_CH_MAX+1; i < NV_DMA_MAX_CHANNELS; i++)
-		__set_bit(i, channel_usage);
-
 	return ret;
 fail:
 	writel(0, addr + APB_DMA_GEN);
-- 
1.7.1

Re: [PATCH] ARM: Tegra: DMA: Fail safe if initialization fails

[Colin Cross]

On Wed, Feb 23, 2011 at 9:29 AM, Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org> wrote:
> tegra_dma_init currently simply bails out early if any initialization fails.
> This skips various data-structure initialization. In turn, this means that
> tegra_dma_allocate_channel can still hand out channels. In this case, when
> tegra_dma_free_channel is called, which calls tegra_dma_cancel, the walking
> on ch->list will OOPS since the list's next/prev pointers may still be
> NULL.
>
> To solve this:
> * Mark all possible channels as in-use before doing anything else in init.
> * Only mark a channel as free once all channel-related initialization has
>  completed.
>
> This prevents allocate_channel from handing out uninitialized channels.
>
> There is still one small hole; allocate_channel can't check the usage array
> for the shared channel, since this channel is permanently marked in-use.
> This could be solved using an explicit "init OK" flag that allocate_channel
> could check.
If we still need an init complete flag, why not skip this patch and
just add that?

>
> Signed-off-by: Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> ---
>  arch/arm/mach-tegra/dma.c |   20 ++++++++------------
>  1 files changed, 8 insertions(+), 12 deletions(-)
>
> diff --git a/arch/arm/mach-tegra/dma.c b/arch/arm/mach-tegra/dma.c
> index 2d720f2..79765ca 100644
> --- a/arch/arm/mach-tegra/dma.c
> +++ b/arch/arm/mach-tegra/dma.c
> @@ -678,6 +678,12 @@ int __init tegra_dma_init(void)
>        void __iomem *addr;
>        struct clk *c;
>
> +       memset(channel_usage, 0, sizeof(channel_usage));
> +       memset(dma_channels, 0, sizeof(dma_channels));
I noticed yesterday these memsets were unnecessary, channel_usage and
dma_channels will be allocated as 0.

> +
> +       for (i = 0; i < NV_DMA_MAX_CHANNELS; i++)
> +               __set_bit(i, channel_usage);
> +
>        c = clk_get_sys("tegra-dma", NULL);
>        if (IS_ERR(c)) {
>                pr_err("Unable to get clock for APB DMA\n");
> @@ -696,18 +702,9 @@ int __init tegra_dma_init(void)
>        writel(0xFFFFFFFFul >> (31 - TEGRA_SYSTEM_DMA_CH_MAX),
>               addr + APB_DMA_IRQ_MASK_SET);
>
> -       memset(channel_usage, 0, sizeof(channel_usage));
> -       memset(dma_channels, 0, sizeof(dma_channels));
> -
> -       /* Reserve all the channels we are not supposed to touch */
> -       for (i = 0; i < TEGRA_SYSTEM_DMA_CH_MIN; i++)
> -               __set_bit(i, channel_usage);
> -
>        for (i = TEGRA_SYSTEM_DMA_CH_MIN; i <= TEGRA_SYSTEM_DMA_CH_MAX; i++) {
>                struct tegra_dma_channel *ch = &dma_channels[i];
>
> -               __clear_bit(i, channel_usage);
> -
>                ch->id = i;
>                snprintf(ch->name, TEGRA_DMA_NAME_SIZE, "dma_channel_%d", i);
>
> @@ -726,13 +723,12 @@ int __init tegra_dma_init(void)
>                        goto fail;
>                }
>                ch->irq = irq;
> +
> +               __clear_bit(i, channel_usage);
>        }
>        /* mark the shared channel allocated */
>        __set_bit(TEGRA_SYSTEM_DMA_CH_MIN, channel_usage);
>
> -       for (i = TEGRA_SYSTEM_DMA_CH_MAX+1; i < NV_DMA_MAX_CHANNELS; i++)
> -               __set_bit(i, channel_usage);
> -
>        return ret;
>  fail:
>        writel(0, addr + APB_DMA_GEN);
> --
> 1.7.1
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
--
To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH] ARM: Tegra: DMA: Fail safe if initialization fails

[Colin Cross]

On Wed, Feb 23, 2011 at 9:29 AM, Stephen Warren <swarren@nvidia.com> wrote:
> tegra_dma_init currently simply bails out early if any initialization fails.
> This skips various data-structure initialization. In turn, this means that
> tegra_dma_allocate_channel can still hand out channels. In this case, when
> tegra_dma_free_channel is called, which calls tegra_dma_cancel, the walking
> on ch->list will OOPS since the list's next/prev pointers may still be
> NULL.
>
> To solve this:
> * Mark all possible channels as in-use before doing anything else in init.
> * Only mark a channel as free once all channel-related initialization has
> ?completed.
>
> This prevents allocate_channel from handing out uninitialized channels.
>
> There is still one small hole; allocate_channel can't check the usage array
> for the shared channel, since this channel is permanently marked in-use.
> This could be solved using an explicit "init OK" flag that allocate_channel
> could check.
If we still need an init complete flag, why not skip this patch and
just add that?

>
> Signed-off-by: Stephen Warren <swarren@nvidia.com>
> ---
> ?arch/arm/mach-tegra/dma.c | ? 20 ++++++++------------
> ?1 files changed, 8 insertions(+), 12 deletions(-)
>
> diff --git a/arch/arm/mach-tegra/dma.c b/arch/arm/mach-tegra/dma.c
> index 2d720f2..79765ca 100644
> --- a/arch/arm/mach-tegra/dma.c
> +++ b/arch/arm/mach-tegra/dma.c
> @@ -678,6 +678,12 @@ int __init tegra_dma_init(void)
> ? ? ? ?void __iomem *addr;
> ? ? ? ?struct clk *c;
>
> + ? ? ? memset(channel_usage, 0, sizeof(channel_usage));
> + ? ? ? memset(dma_channels, 0, sizeof(dma_channels));
I noticed yesterday these memsets were unnecessary, channel_usage and
dma_channels will be allocated as 0.

> +
> + ? ? ? for (i = 0; i < NV_DMA_MAX_CHANNELS; i++)
> + ? ? ? ? ? ? ? __set_bit(i, channel_usage);
> +
> ? ? ? ?c = clk_get_sys("tegra-dma", NULL);
> ? ? ? ?if (IS_ERR(c)) {
> ? ? ? ? ? ? ? ?pr_err("Unable to get clock for APB DMA\n");
> @@ -696,18 +702,9 @@ int __init tegra_dma_init(void)
> ? ? ? ?writel(0xFFFFFFFFul >> (31 - TEGRA_SYSTEM_DMA_CH_MAX),
> ? ? ? ? ? ? ? addr + APB_DMA_IRQ_MASK_SET);
>
> - ? ? ? memset(channel_usage, 0, sizeof(channel_usage));
> - ? ? ? memset(dma_channels, 0, sizeof(dma_channels));
> -
> - ? ? ? /* Reserve all the channels we are not supposed to touch */
> - ? ? ? for (i = 0; i < TEGRA_SYSTEM_DMA_CH_MIN; i++)
> - ? ? ? ? ? ? ? __set_bit(i, channel_usage);
> -
> ? ? ? ?for (i = TEGRA_SYSTEM_DMA_CH_MIN; i <= TEGRA_SYSTEM_DMA_CH_MAX; i++) {
> ? ? ? ? ? ? ? ?struct tegra_dma_channel *ch = &dma_channels[i];
>
> - ? ? ? ? ? ? ? __clear_bit(i, channel_usage);
> -
> ? ? ? ? ? ? ? ?ch->id = i;
> ? ? ? ? ? ? ? ?snprintf(ch->name, TEGRA_DMA_NAME_SIZE, "dma_channel_%d", i);
>
> @@ -726,13 +723,12 @@ int __init tegra_dma_init(void)
> ? ? ? ? ? ? ? ? ? ? ? ?goto fail;
> ? ? ? ? ? ? ? ?}
> ? ? ? ? ? ? ? ?ch->irq = irq;
> +
> + ? ? ? ? ? ? ? __clear_bit(i, channel_usage);
> ? ? ? ?}
> ? ? ? ?/* mark the shared channel allocated */
> ? ? ? ?__set_bit(TEGRA_SYSTEM_DMA_CH_MIN, channel_usage);
>
> - ? ? ? for (i = TEGRA_SYSTEM_DMA_CH_MAX+1; i < NV_DMA_MAX_CHANNELS; i++)
> - ? ? ? ? ? ? ? __set_bit(i, channel_usage);
> -
> ? ? ? ?return ret;
> ?fail:
> ? ? ? ?writel(0, addr + APB_DMA_GEN);
> --
> 1.7.1
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
> the body of a message to majordomo at vger.kernel.org
> More majordomo info at ?http://vger.kernel.org/majordomo-info.html
>

RE: [PATCH] ARM: Tegra: DMA: Fail safe if initialization fails

[Stephen Warren]

Colin Cross wrote at Wednesday, February 23, 2011 10:38 AM:
> 
> On Wed, Feb 23, 2011 at 9:29 AM, Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org> wrote:
> > tegra_dma_init currently simply bails out early if any initialization fails.
> > This skips various data-structure initialization. In turn, this means that
> > tegra_dma_allocate_channel can still hand out channels. In this case, when
> > tegra_dma_free_channel is called, which calls tegra_dma_cancel, the walking
> > on ch->list will OOPS since the list's next/prev pointers may still be
> > NULL.
> >
> > To solve this:
> > * Mark all possible channels as in-use before doing anything else in init.
> > * Only mark a channel as free once all channel-related initialization has
> >  completed.
> >
> > This prevents allocate_channel from handing out uninitialized channels.
> >
> > There is still one small hole; allocate_channel can't check the usage array
> > for the shared channel, since this channel is permanently marked in-use.
> > This could be solved using an explicit "init OK" flag that allocate_channel
> > could check.
>
> If we still need an init complete flag, why not skip this patch and
> just add that?

I tend to like the cleanup in this patch; it simplifies the channel_usage
initialization. I'll let you make the call on whether to take this or not.

I can certainly throw together a second patch, or modify this patch, to also
use an init_ok flag to completely solve the shared channel case too though.

-- 
nvpublic

--
To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH] ARM: Tegra: DMA: Fail safe if initialization fails

[Stephen Warren]

Colin Cross wrote at Wednesday, February 23, 2011 10:38 AM:
> 
> On Wed, Feb 23, 2011 at 9:29 AM, Stephen Warren <swarren@nvidia.com> wrote:
> > tegra_dma_init currently simply bails out early if any initialization fails.
> > This skips various data-structure initialization. In turn, this means that
> > tegra_dma_allocate_channel can still hand out channels. In this case, when
> > tegra_dma_free_channel is called, which calls tegra_dma_cancel, the walking
> > on ch->list will OOPS since the list's next/prev pointers may still be
> > NULL.
> >
> > To solve this:
> > * Mark all possible channels as in-use before doing anything else in init.
> > * Only mark a channel as free once all channel-related initialization has
> > ?completed.
> >
> > This prevents allocate_channel from handing out uninitialized channels.
> >
> > There is still one small hole; allocate_channel can't check the usage array
> > for the shared channel, since this channel is permanently marked in-use.
> > This could be solved using an explicit "init OK" flag that allocate_channel
> > could check.
>
> If we still need an init complete flag, why not skip this patch and
> just add that?

I tend to like the cleanup in this patch; it simplifies the channel_usage
initialization. I'll let you make the call on whether to take this or not.

I can certainly throw together a second patch, or modify this patch, to also
use an init_ok flag to completely solve the shared channel case too though.

-- 
nvpublic

Re: [PATCH] ARM: Tegra: DMA: Fail safe if initialization fails

[Colin Cross]

On Wed, Feb 23, 2011 at 9:54 AM, Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org> wrote:
> Colin Cross wrote at Wednesday, February 23, 2011 10:38 AM:
>>
>> On Wed, Feb 23, 2011 at 9:29 AM, Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org> wrote:
>> > tegra_dma_init currently simply bails out early if any initialization fails.
>> > This skips various data-structure initialization. In turn, this means that
>> > tegra_dma_allocate_channel can still hand out channels. In this case, when
>> > tegra_dma_free_channel is called, which calls tegra_dma_cancel, the walking
>> > on ch->list will OOPS since the list's next/prev pointers may still be
>> > NULL.
>> >
>> > To solve this:
>> > * Mark all possible channels as in-use before doing anything else in init.
>> > * Only mark a channel as free once all channel-related initialization has
>> >  completed.
>> >
>> > This prevents allocate_channel from handing out uninitialized channels.
>> >
>> > There is still one small hole; allocate_channel can't check the usage array
>> > for the shared channel, since this channel is permanently marked in-use.
>> > This could be solved using an explicit "init OK" flag that allocate_channel
>> > could check.
>>
>> If we still need an init complete flag, why not skip this patch and
>> just add that?
>
> I tend to like the cleanup in this patch; it simplifies the channel_usage
> initialization. I'll let you make the call on whether to take this or not.
Ok, it seems like a reasonable cleanup.  Can I request a few changes?
Get rid of the redundant memsets
Replace the initial for loop to set all the bits with bitmap_set
Might as well add the init complete flag to this patch as well.  You
could also do it by checking the channel_usage bit for the shared
channel, and never setting it in tegra_dma_allocate_channel.

> I can certainly throw together a second patch, or modify this patch, to also
> use an init_ok flag to completely solve the shared channel case too though.
>
> --
> nvpublic
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH] ARM: Tegra: DMA: Fail safe if initialization fails

[Colin Cross]

On Wed, Feb 23, 2011 at 9:54 AM, Stephen Warren <swarren@nvidia.com> wrote:
> Colin Cross wrote at Wednesday, February 23, 2011 10:38 AM:
>>
>> On Wed, Feb 23, 2011 at 9:29 AM, Stephen Warren <swarren@nvidia.com> wrote:
>> > tegra_dma_init currently simply bails out early if any initialization fails.
>> > This skips various data-structure initialization. In turn, this means that
>> > tegra_dma_allocate_channel can still hand out channels. In this case, when
>> > tegra_dma_free_channel is called, which calls tegra_dma_cancel, the walking
>> > on ch->list will OOPS since the list's next/prev pointers may still be
>> > NULL.
>> >
>> > To solve this:
>> > * Mark all possible channels as in-use before doing anything else in init.
>> > * Only mark a channel as free once all channel-related initialization has
>> > ?completed.
>> >
>> > This prevents allocate_channel from handing out uninitialized channels.
>> >
>> > There is still one small hole; allocate_channel can't check the usage array
>> > for the shared channel, since this channel is permanently marked in-use.
>> > This could be solved using an explicit "init OK" flag that allocate_channel
>> > could check.
>>
>> If we still need an init complete flag, why not skip this patch and
>> just add that?
>
> I tend to like the cleanup in this patch; it simplifies the channel_usage
> initialization. I'll let you make the call on whether to take this or not.
Ok, it seems like a reasonable cleanup.  Can I request a few changes?
Get rid of the redundant memsets
Replace the initial for loop to set all the bits with bitmap_set
Might as well add the init complete flag to this patch as well.  You
could also do it by checking the channel_usage bit for the shared
channel, and never setting it in tegra_dma_allocate_channel.

> I can certainly throw together a second patch, or modify this patch, to also
> use an init_ok flag to completely solve the shared channel case too though.
>
> --
> nvpublic
>
>