2022-12-21 20:11:19

by Nicolas Frayer

[permalink] [raw]
Subject: [PATCH] soc: ti: k3-ringacc: Add try_module_get() to k3_dmaring_request_dual_ring()

When the k3 ring accelerator driver has been modified to add module build
support, try_module_get() and module_put() have been added to update the
module refcnt. One code path has not been updated and it has introduced
an issue where the refcnt is decremented by module_put() in
k3_ringacc_ring_free() without being incremented previously.
Adding try_module_get() to k3_dmaring_request_dual_ring() ensures the
refcnt is kept up to date.

Signed-off-by: Nicolas Frayer <[email protected]>
---
drivers/soc/ti/k3-ringacc.c | 7 +++++++
1 file changed, 7 insertions(+)

diff --git a/drivers/soc/ti/k3-ringacc.c b/drivers/soc/ti/k3-ringacc.c
index e01e4d815230a..8f131368a7586 100644
--- a/drivers/soc/ti/k3-ringacc.c
+++ b/drivers/soc/ti/k3-ringacc.c
@@ -406,6 +406,11 @@ static int k3_dmaring_request_dual_ring(struct k3_ringacc *ringacc, int fwd_id,

mutex_lock(&ringacc->req_lock);

+ if (!try_module_get(ringacc->dev->driver->owner)) {
+ ret = -EINVAL;
+ goto err_module_get;
+ }
+
if (test_bit(fwd_id, ringacc->rings_inuse)) {
ret = -EBUSY;
goto error;
@@ -421,6 +426,8 @@ static int k3_dmaring_request_dual_ring(struct k3_ringacc *ringacc, int fwd_id,
return 0;

error:
+ module_put(ringacc->dev->driver->owner);
+err_module_get:
mutex_unlock(&ringacc->req_lock);
return ret;
}
--
2.25.1


2022-12-24 22:54:03

by Péter Ujfalusi

[permalink] [raw]
Subject: Re: [PATCH] soc: ti: k3-ringacc: Add try_module_get() to k3_dmaring_request_dual_ring()

Hi Nicolas,

On 21/12/2022 22:05, Nicolas Frayer wrote:
> When the k3 ring accelerator driver has been modified to add module build
> support, try_module_get() and module_put() have been added to update the
> module refcnt. One code path has not been updated and it has introduced
> an issue where the refcnt is decremented by module_put() in
> k3_ringacc_ring_free() without being incremented previously.
> Adding try_module_get() to k3_dmaring_request_dual_ring() ensures the
> refcnt is kept up to date.

Good catch!
I had never had my hands on a device with DMSS (BCDMA/PKDMA) and the
original module support was developed and tested on devices with UDMA.

can you add the fixes tag along with my reviewed by tag?

Fixes: c07f216a8b72 ("soc: ti: k3-ringacc: Allow the driver to be built as module")
Reviewed-by: Peter Ujfalusi <[email protected]>

>
> Signed-off-by: Nicolas Frayer <[email protected]>
> ---
> drivers/soc/ti/k3-ringacc.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/drivers/soc/ti/k3-ringacc.c b/drivers/soc/ti/k3-ringacc.c
> index e01e4d815230a..8f131368a7586 100644
> --- a/drivers/soc/ti/k3-ringacc.c
> +++ b/drivers/soc/ti/k3-ringacc.c
> @@ -406,6 +406,11 @@ static int k3_dmaring_request_dual_ring(struct k3_ringacc *ringacc, int fwd_id,
>
> mutex_lock(&ringacc->req_lock);
>
> + if (!try_module_get(ringacc->dev->driver->owner)) {
> + ret = -EINVAL;
> + goto err_module_get;
> + }
> +
> if (test_bit(fwd_id, ringacc->rings_inuse)) {
> ret = -EBUSY;
> goto error;
> @@ -421,6 +426,8 @@ static int k3_dmaring_request_dual_ring(struct k3_ringacc *ringacc, int fwd_id,
> return 0;
>
> error:
> + module_put(ringacc->dev->driver->owner);
> +err_module_get:
> mutex_unlock(&ringacc->req_lock);
> return ret;
> }

--
Péter

2022-12-30 00:19:14

by Nicolas Frayer

[permalink] [raw]
Subject: Re: [PATCH] soc: ti: k3-ringacc: Add try_module_get() to k3_dmaring_request_dual_ring()

Le sam. 24 déc. 2022 à 23:01, Péter Ujfalusi
<[email protected]> a écrit :
>
> Hi Nicolas,
>
> On 21/12/2022 22:05, Nicolas Frayer wrote:
> > When the k3 ring accelerator driver has been modified to add module build
> > support, try_module_get() and module_put() have been added to update the
> > module refcnt. One code path has not been updated and it has introduced
> > an issue where the refcnt is decremented by module_put() in
> > k3_ringacc_ring_free() without being incremented previously.
> > Adding try_module_get() to k3_dmaring_request_dual_ring() ensures the
> > refcnt is kept up to date.
>
> Good catch!
> I had never had my hands on a device with DMSS (BCDMA/PKDMA) and the
> original module support was developed and tested on devices with UDMA.
>
> can you add the fixes tag along with my reviewed by tag?
>
> Fixes: c07f216a8b72 ("soc: ti: k3-ringacc: Allow the driver to be built as module")
> Reviewed-by: Peter Ujfalusi <[email protected]>
>
> >
> > Signed-off-by: Nicolas Frayer <[email protected]>
> > ---
> > drivers/soc/ti/k3-ringacc.c | 7 +++++++
> > 1 file changed, 7 insertions(+)
> >
> > diff --git a/drivers/soc/ti/k3-ringacc.c b/drivers/soc/ti/k3-ringacc.c
> > index e01e4d815230a..8f131368a7586 100644
> > --- a/drivers/soc/ti/k3-ringacc.c
> > +++ b/drivers/soc/ti/k3-ringacc.c
> > @@ -406,6 +406,11 @@ static int k3_dmaring_request_dual_ring(struct k3_ringacc *ringacc, int fwd_id,
> >
> > mutex_lock(&ringacc->req_lock);
> >
> > + if (!try_module_get(ringacc->dev->driver->owner)) {
> > + ret = -EINVAL;
> > + goto err_module_get;
> > + }
> > +
> > if (test_bit(fwd_id, ringacc->rings_inuse)) {
> > ret = -EBUSY;
> > goto error;
> > @@ -421,6 +426,8 @@ static int k3_dmaring_request_dual_ring(struct k3_ringacc *ringacc, int fwd_id,
> > return 0;
> >
> > error:
> > + module_put(ringacc->dev->driver->owner);
> > +err_module_get:
> > mutex_unlock(&ringacc->req_lock);
> > return ret;
> > }
>
> --
> Péter
Thanks Peter, I'll send a v2 with the tags.