diff mbox series

[v2,5/9] coresight: Dynamically add connections

Message ID 20230310160610.742382-6-james.clark@arm.com (mailing list archive)
State New, archived
Headers show
Series coresight: Fix CTI module refcount leak by making it a helper device | expand

Commit Message

James Clark March 10, 2023, 4:06 p.m. UTC
Add a function for adding connections dynamically. This also removes
the 1:1 mapping between port number and the index into the connections
array. The only place this mapping was used was in the warning for
duplicate output ports, which has been replaced by a search. Other
uses of the port number already use the port member variable.

Being able to dynamically add connections will allow other devices like
CTI to re-use the connection mechanism despite not having explicit
connections described in the DT.

Signed-off-by: James Clark <james.clark@arm.com>
---
 .../hwtracing/coresight/coresight-platform.c  | 77 ++++++++++++++-----
 include/linux/coresight.h                     |  7 +-
 2 files changed, 64 insertions(+), 20 deletions(-)

Comments

Suzuki K Poulose March 16, 2023, 5:12 p.m. UTC | #1
On 10/03/2023 16:06, James Clark wrote:
> Add a function for adding connections dynamically. This also removes
> the 1:1 mapping between port number and the index into the connections
> array. The only place this mapping was used was in the warning for
> duplicate output ports, which has been replaced by a search. Other
> uses of the port number already use the port member variable.
> 
> Being able to dynamically add connections will allow other devices like
> CTI to re-use the connection mechanism despite not having explicit
> connections described in the DT.
> 
> Signed-off-by: James Clark <james.clark@arm.com>
> ---
>   .../hwtracing/coresight/coresight-platform.c  | 77 ++++++++++++++-----
>   include/linux/coresight.h                     |  7 +-
>   2 files changed, 64 insertions(+), 20 deletions(-)
> 
> diff --git a/drivers/hwtracing/coresight/coresight-platform.c b/drivers/hwtracing/coresight/coresight-platform.c
> index c77238cdf448..16553f7dde12 100644
> --- a/drivers/hwtracing/coresight/coresight-platform.c
> +++ b/drivers/hwtracing/coresight/coresight-platform.c
> @@ -27,8 +27,9 @@ static int coresight_alloc_conns(struct device *dev,
>   				 struct coresight_platform_data *pdata)
>   {
>   	if (pdata->nr_outconns) {
> -		pdata->out_conns = devm_kcalloc(dev, pdata->nr_outconns,
> -					    sizeof(*pdata->out_conns), GFP_KERNEL);
> +		pdata->out_conns = devm_krealloc_array(
> +			dev, pdata->out_conns, pdata->nr_outconns,

super minor nit:
		pdata->out_conns = devm_krealloc_array(dev,

		
> +			sizeof(*pdata->out_conns), GFP_KERNEL | __GFP_ZERO);
>   		if (!pdata->out_conns)
>   			return -ENOMEM;
>   	}
> @@ -36,6 +37,48 @@ static int coresight_alloc_conns(struct device *dev,
>   	return 0;
>   }
>   
> +/*
> + * Add a connection in the first free slot, or realloc
> + * if there is no space. @conn's contents is copied into the new slot.
> + *
> + * If the output port is already assigned on this device, return -EINVAL
> + */
> +int coresight_add_conn(struct device *dev,
> +		       struct coresight_platform_data *pdata,
> +		       const struct coresight_connection *conn)
> +{
> +	int ret;
> +	struct coresight_connection *free_conn = NULL;
> +	struct coresight_connection *i;
> +
> +	/*
> +	 * Search for a free slot, and while looking for one, warn
> +	 * on any existing duplicate output port.
> +	 */
> +	for (i = pdata->out_conns; i < pdata->out_conns + pdata->nr_outconns;
> +	     ++i) {

minor nit: I see why you have gone against using "i" as index into
the array. But I think having that as the index is still better
readable.

	for (i = 0; i < pdata->nr_outconns; i++) {
		struct coresight_connection *c = &pdata->out_conns[i];

> +		if (i->remote_fwnode && conn->port != -1 &&
> +		    i->port == conn->port) {
> +			dev_warn(dev, "Duplicate output port %d\n", i->port);
> +			return -EINVAL;
> +		}
> +		if (!i->remote_fwnode && !free_conn)
> +			free_conn = i;
> +	}
> +
> +	if (!free_conn) {

and:
	/* No free slots */
	if (i == pdata->nr_outconns) {

> +		pdata->nr_outconns++;
> +		ret = coresight_alloc_conns(dev, pdata);
> +		if (ret)
> +			return ret;
> +		free_conn = &pdata->out_conns[pdata->nr_outconns - 1];
> +	}
> +

and:
	pdata->out_conns[i] = *conn;


Otherwise looks good to me.

Suzuki
Mike Leach March 21, 2023, 5:56 p.m. UTC | #2
Hi James

On Thu, 16 Mar 2023 at 17:12, Suzuki K Poulose <suzuki.poulose@arm.com> wrote:
>
> On 10/03/2023 16:06, James Clark wrote:
> > Add a function for adding connections dynamically. This also removes
> > the 1:1 mapping between port number and the index into the connections
> > array. The only place this mapping was used was in the warning for
> > duplicate output ports, which has been replaced by a search. Other
> > uses of the port number already use the port member variable.
> >
> > Being able to dynamically add connections will allow other devices like
> > CTI to re-use the connection mechanism despite not having explicit
> > connections described in the DT.
> >
> > Signed-off-by: James Clark <james.clark@arm.com>
> > ---
> >   .../hwtracing/coresight/coresight-platform.c  | 77 ++++++++++++++-----
> >   include/linux/coresight.h                     |  7 +-
> >   2 files changed, 64 insertions(+), 20 deletions(-)
> >
> > diff --git a/drivers/hwtracing/coresight/coresight-platform.c b/drivers/hwtracing/coresight/coresight-platform.c
> > index c77238cdf448..16553f7dde12 100644
> > --- a/drivers/hwtracing/coresight/coresight-platform.c
> > +++ b/drivers/hwtracing/coresight/coresight-platform.c
> > @@ -27,8 +27,9 @@ static int coresight_alloc_conns(struct device *dev,
> >                                struct coresight_platform_data *pdata)
> >   {
> >       if (pdata->nr_outconns) {
> > -             pdata->out_conns = devm_kcalloc(dev, pdata->nr_outconns,
> > -                                         sizeof(*pdata->out_conns), GFP_KERNEL);
> > +             pdata->out_conns = devm_krealloc_array(
> > +                     dev, pdata->out_conns, pdata->nr_outconns,
>
> super minor nit:
>                 pdata->out_conns = devm_krealloc_array(dev,
>
>
> > +                     sizeof(*pdata->out_conns), GFP_KERNEL | __GFP_ZERO);
> >               if (!pdata->out_conns)
> >                       return -ENOMEM;
> >       }
> > @@ -36,6 +37,48 @@ static int coresight_alloc_conns(struct device *dev,
> >       return 0;
> >   }
> >
> > +/*
> > + * Add a connection in the first free slot, or realloc
> > + * if there is no space. @conn's contents is copied into the new slot.
> > + *
> > + * If the output port is already assigned on this device, return -EINVAL
> > + */
> > +int coresight_add_conn(struct device *dev,
> > +                    struct coresight_platform_data *pdata,
> > +                    const struct coresight_connection *conn)
> > +{
> > +     int ret;
> > +     struct coresight_connection *free_conn = NULL;
> > +     struct coresight_connection *i;
> > +
> > +     /*
> > +      * Search for a free slot, and while looking for one, warn
> > +      * on any existing duplicate output port.
> > +      */
> > +     for (i = pdata->out_conns; i < pdata->out_conns + pdata->nr_outconns;
> > +          ++i) {
>
> minor nit: I see why you have gone against using "i" as index into
> the array. But I think having that as the index is still better
> readable.
>
>         for (i = 0; i < pdata->nr_outconns; i++) {
>                 struct coresight_connection *c = &pdata->out_conns[i];
>
> > +             if (i->remote_fwnode && conn->port != -1 &&
> > +                 i->port == conn->port) {
> > +                     dev_warn(dev, "Duplicate output port %d\n", i->port);
> > +                     return -EINVAL;
> > +             }

This code assumes that slots are filled sequentially and that it is
not possible to release slots out of order - i.e. if we find a free
slot, there is not a match in a later slot.
I can't think how this could happen but a comment to confirm this
might be needed here.

When we had 1:1 port / array index then this check was guaranteed

Mike



> > +             if (!i->remote_fwnode && !free_conn)
> > +                     free_conn = i;
> > +     }
> > +
> > +     if (!free_conn) {
>
> and:
>         /* No free slots */
>         if (i == pdata->nr_outconns) {
>
> > +             pdata->nr_outconns++;
> > +             ret = coresight_alloc_conns(dev, pdata);
> > +             if (ret)
> > +                     return ret;
> > +             free_conn = &pdata->out_conns[pdata->nr_outconns - 1];
> > +     }
> > +
>
> and:
>         pdata->out_conns[i] = *conn;
>
>
> Otherwise looks good to me.
>
> Suzuki
>
>


--
Mike Leach
Principal Engineer, ARM Ltd.
Manchester Design Centre. UK
James Clark March 23, 2023, 10:49 a.m. UTC | #3
On 21/03/2023 17:56, Mike Leach wrote:
> Hi James
> 
> On Thu, 16 Mar 2023 at 17:12, Suzuki K Poulose <suzuki.poulose@arm.com> wrote:
>>
>> On 10/03/2023 16:06, James Clark wrote:
>>> Add a function for adding connections dynamically. This also removes
>>> the 1:1 mapping between port number and the index into the connections
>>> array. The only place this mapping was used was in the warning for
>>> duplicate output ports, which has been replaced by a search. Other
>>> uses of the port number already use the port member variable.
>>>
>>> Being able to dynamically add connections will allow other devices like
>>> CTI to re-use the connection mechanism despite not having explicit
>>> connections described in the DT.
>>>
>>> Signed-off-by: James Clark <james.clark@arm.com>
>>> ---
>>>   .../hwtracing/coresight/coresight-platform.c  | 77 ++++++++++++++-----
>>>   include/linux/coresight.h                     |  7 +-
>>>   2 files changed, 64 insertions(+), 20 deletions(-)
>>>
>>> diff --git a/drivers/hwtracing/coresight/coresight-platform.c b/drivers/hwtracing/coresight/coresight-platform.c
>>> index c77238cdf448..16553f7dde12 100644
>>> --- a/drivers/hwtracing/coresight/coresight-platform.c
>>> +++ b/drivers/hwtracing/coresight/coresight-platform.c
>>> @@ -27,8 +27,9 @@ static int coresight_alloc_conns(struct device *dev,
>>>                                struct coresight_platform_data *pdata)
>>>   {
>>>       if (pdata->nr_outconns) {
>>> -             pdata->out_conns = devm_kcalloc(dev, pdata->nr_outconns,
>>> -                                         sizeof(*pdata->out_conns), GFP_KERNEL);
>>> +             pdata->out_conns = devm_krealloc_array(
>>> +                     dev, pdata->out_conns, pdata->nr_outconns,
>>
>> super minor nit:
>>                 pdata->out_conns = devm_krealloc_array(dev,
>>
>>
>>> +                     sizeof(*pdata->out_conns), GFP_KERNEL | __GFP_ZERO);
>>>               if (!pdata->out_conns)
>>>                       return -ENOMEM;
>>>       }
>>> @@ -36,6 +37,48 @@ static int coresight_alloc_conns(struct device *dev,
>>>       return 0;
>>>   }
>>>
>>> +/*
>>> + * Add a connection in the first free slot, or realloc
>>> + * if there is no space. @conn's contents is copied into the new slot.
>>> + *
>>> + * If the output port is already assigned on this device, return -EINVAL
>>> + */
>>> +int coresight_add_conn(struct device *dev,
>>> +                    struct coresight_platform_data *pdata,
>>> +                    const struct coresight_connection *conn)
>>> +{
>>> +     int ret;
>>> +     struct coresight_connection *free_conn = NULL;
>>> +     struct coresight_connection *i;
>>> +
>>> +     /*
>>> +      * Search for a free slot, and while looking for one, warn
>>> +      * on any existing duplicate output port.
>>> +      */
>>> +     for (i = pdata->out_conns; i < pdata->out_conns + pdata->nr_outconns;
>>> +          ++i) {
>>
>> minor nit: I see why you have gone against using "i" as index into
>> the array. But I think having that as the index is still better
>> readable.
>>
>>         for (i = 0; i < pdata->nr_outconns; i++) {
>>                 struct coresight_connection *c = &pdata->out_conns[i];
>>
>>> +             if (i->remote_fwnode && conn->port != -1 &&
>>> +                 i->port == conn->port) {
>>> +                     dev_warn(dev, "Duplicate output port %d\n", i->port);
>>> +                     return -EINVAL;
>>> +             }
> 
> This code assumes that slots are filled sequentially and that it is
> not possible to release slots out of order - i.e. if we find a free
> slot, there is not a match in a later slot.
> I can't think how this could happen but a comment to confirm this
> might be needed here.
> 
> When we had 1:1 port / array index then this check was guaranteed
>> Mike

I thought about this but I couldn't see an issue here. The loop always
runs to the end even if a free slot is found so it should find
duplicates in any order. Unless I'm missing some other edge case?

> 
> 
> 
>>> +             if (!i->remote_fwnode && !free_conn)
>>> +                     free_conn = i;
>>> +     }
>>> +
>>> +     if (!free_conn) {
>>
>> and:
>>         /* No free slots */
>>         if (i == pdata->nr_outconns) {
>>
>>> +             pdata->nr_outconns++;
>>> +             ret = coresight_alloc_conns(dev, pdata);
>>> +             if (ret)
>>> +                     return ret;
>>> +             free_conn = &pdata->out_conns[pdata->nr_outconns - 1];
>>> +     }
>>> +
>>
>> and:
>>         pdata->out_conns[i] = *conn;
>>
>>
>> Otherwise looks good to me.
>>
>> Suzuki
>>
>>
> 
> 
> --
> Mike Leach
> Principal Engineer, ARM Ltd.
> Manchester Design Centre. UK
James Clark March 29, 2023, 12:02 p.m. UTC | #4
On 16/03/2023 17:12, Suzuki K Poulose wrote:
> On 10/03/2023 16:06, James Clark wrote:
>> Add a function for adding connections dynamically. This also removes
>> the 1:1 mapping between port number and the index into the connections
>> array. The only place this mapping was used was in the warning for
>> duplicate output ports, which has been replaced by a search. Other
>> uses of the port number already use the port member variable.
>>
>> Being able to dynamically add connections will allow other devices like
>> CTI to re-use the connection mechanism despite not having explicit
>> connections described in the DT.
>>
>> Signed-off-by: James Clark <james.clark@arm.com>
>> ---
>>   .../hwtracing/coresight/coresight-platform.c  | 77 ++++++++++++++-----
>>   include/linux/coresight.h                     |  7 +-
>>   2 files changed, 64 insertions(+), 20 deletions(-)
>>
>> diff --git a/drivers/hwtracing/coresight/coresight-platform.c
>> b/drivers/hwtracing/coresight/coresight-platform.c
>> index c77238cdf448..16553f7dde12 100644
>> --- a/drivers/hwtracing/coresight/coresight-platform.c
>> +++ b/drivers/hwtracing/coresight/coresight-platform.c
>> @@ -27,8 +27,9 @@ static int coresight_alloc_conns(struct device *dev,
>>                    struct coresight_platform_data *pdata)
>>   {
>>       if (pdata->nr_outconns) {
>> -        pdata->out_conns = devm_kcalloc(dev, pdata->nr_outconns,
>> -                        sizeof(*pdata->out_conns), GFP_KERNEL);
>> +        pdata->out_conns = devm_krealloc_array(
>> +            dev, pdata->out_conns, pdata->nr_outconns,
> 
> super minor nit:
>         pdata->out_conns = devm_krealloc_array(dev,
> 

This is actually clang-format's doing when using the kernel rules in the
root of the repo. I started using it because of some other style
comments I got before. Not sure if this time it's just done something
bad or it's technically ok.

I formatted everything in V3 with it so it should at least all be
consistent.

>        
>> +            sizeof(*pdata->out_conns), GFP_KERNEL | __GFP_ZERO);
>>           if (!pdata->out_conns)
>>               return -ENOMEM;
>>       }
>> @@ -36,6 +37,48 @@ static int coresight_alloc_conns(struct device *dev,
>>       return 0;
>>   }
>>   +/*
>> + * Add a connection in the first free slot, or realloc
>> + * if there is no space. @conn's contents is copied into the new slot.
>> + *
>> + * If the output port is already assigned on this device, return -EINVAL
>> + */
>> +int coresight_add_conn(struct device *dev,
>> +               struct coresight_platform_data *pdata,
>> +               const struct coresight_connection *conn)
>> +{
>> +    int ret;
>> +    struct coresight_connection *free_conn = NULL;
>> +    struct coresight_connection *i;
>> +
>> +    /*
>> +     * Search for a free slot, and while looking for one, warn
>> +     * on any existing duplicate output port.
>> +     */
>> +    for (i = pdata->out_conns; i < pdata->out_conns +
>> pdata->nr_outconns;
>> +         ++i) {
> 
> minor nit: I see why you have gone against using "i" as index into
> the array. But I think having that as the index is still better
> readable.
>     for (i = 0; i < pdata->nr_outconns; i++) {
>         struct coresight_connection *c = &pdata->out_conns[i];
> >> +        if (i->remote_fwnode && conn->port != -1 &&
>> +            i->port == conn->port) {
>> +            dev_warn(dev, "Duplicate output port %d\n", i->port);
>> +            return -EINVAL;
>> +        }
>> +        if (!i->remote_fwnode && !free_conn)
>> +            free_conn = i;
>> +    }
>> +
>> +    if (!free_conn) {
> 
> and:
>     /* No free slots */
>     if (i == pdata->nr_outconns) {
> 
>> +        pdata->nr_outconns++;
>> +        ret = coresight_alloc_conns(dev, pdata);
>> +        if (ret)
>> +            return ret;
>> +        free_conn = &pdata->out_conns[pdata->nr_outconns - 1];
>> +    }
>> +
> 
> and:
>     pdata->out_conns[i] = *conn;
> 
> > Otherwise looks good to me.
> 
> Suzuki
> 
>
diff mbox series

Patch

diff --git a/drivers/hwtracing/coresight/coresight-platform.c b/drivers/hwtracing/coresight/coresight-platform.c
index c77238cdf448..16553f7dde12 100644
--- a/drivers/hwtracing/coresight/coresight-platform.c
+++ b/drivers/hwtracing/coresight/coresight-platform.c
@@ -27,8 +27,9 @@  static int coresight_alloc_conns(struct device *dev,
 				 struct coresight_platform_data *pdata)
 {
 	if (pdata->nr_outconns) {
-		pdata->out_conns = devm_kcalloc(dev, pdata->nr_outconns,
-					    sizeof(*pdata->out_conns), GFP_KERNEL);
+		pdata->out_conns = devm_krealloc_array(
+			dev, pdata->out_conns, pdata->nr_outconns,
+			sizeof(*pdata->out_conns), GFP_KERNEL | __GFP_ZERO);
 		if (!pdata->out_conns)
 			return -ENOMEM;
 	}
@@ -36,6 +37,48 @@  static int coresight_alloc_conns(struct device *dev,
 	return 0;
 }
 
+/*
+ * Add a connection in the first free slot, or realloc
+ * if there is no space. @conn's contents is copied into the new slot.
+ *
+ * If the output port is already assigned on this device, return -EINVAL
+ */
+int coresight_add_conn(struct device *dev,
+		       struct coresight_platform_data *pdata,
+		       const struct coresight_connection *conn)
+{
+	int ret;
+	struct coresight_connection *free_conn = NULL;
+	struct coresight_connection *i;
+
+	/*
+	 * Search for a free slot, and while looking for one, warn
+	 * on any existing duplicate output port.
+	 */
+	for (i = pdata->out_conns; i < pdata->out_conns + pdata->nr_outconns;
+	     ++i) {
+		if (i->remote_fwnode && conn->port != -1 &&
+		    i->port == conn->port) {
+			dev_warn(dev, "Duplicate output port %d\n", i->port);
+			return -EINVAL;
+		}
+		if (!i->remote_fwnode && !free_conn)
+			free_conn = i;
+	}
+
+	if (!free_conn) {
+		pdata->nr_outconns++;
+		ret = coresight_alloc_conns(dev, pdata);
+		if (ret)
+			return ret;
+		free_conn = &pdata->out_conns[pdata->nr_outconns - 1];
+	}
+
+	*free_conn = *conn;
+	return 0;
+}
+EXPORT_SYMBOL_GPL(coresight_add_conn);
+
 static struct device *
 coresight_find_device_by_fwnode(struct fwnode_handle *fwnode)
 {
@@ -224,7 +267,7 @@  static int of_coresight_parse_endpoint(struct device *dev,
 	struct device_node *rep = NULL;
 	struct device *rdev = NULL;
 	struct fwnode_handle *rdev_fwnode;
-	struct coresight_connection *conn;
+	struct coresight_connection conn;
 
 	do {
 		/* Parse the local port details */
@@ -251,14 +294,7 @@  static int of_coresight_parse_endpoint(struct device *dev,
 			break;
 		}
 
-		conn = &pdata->out_conns[endpoint.port];
-		if (conn->remote_fwnode) {
-			dev_warn(dev, "Duplicate output port %d\n",
-				 endpoint.port);
-			ret = -EINVAL;
-			break;
-		}
-		conn->port = endpoint.port;
+		conn.port = endpoint.port;
 		/*
 		 * Hold the refcount to the target device. This could be
 		 * released via:
@@ -267,8 +303,14 @@  static int of_coresight_parse_endpoint(struct device *dev,
 		 * 2) While removing the target device via
 		 *    coresight_remove_match()
 		 */
-		conn->remote_fwnode = fwnode_handle_get(rdev_fwnode);
-		conn->remote_port = rendpoint.port;
+		conn.remote_fwnode = fwnode_handle_get(rdev_fwnode);
+		conn.remote_port = rendpoint.port;
+
+		ret = coresight_add_conn(dev, pdata, &conn);
+		if (ret) {
+			fwnode_handle_put(conn.remote_fwnode);
+			return ret;
+		}
 		/* Connection record updated */
 	} while (0);
 
@@ -741,13 +783,10 @@  static int acpi_coresight_parse_graph(struct acpi_device *adev,
 
 	/* Copy the connection information to the final location */
 	for (i = 0; conns + i < ptr; i++) {
-		int port = conns[i].port;
-
-		/* Duplicate output port */
-		WARN_ON(pdata->out_conns[port].remote_fwnode);
-		pdata->out_conns[port] = conns[i];
+		rc = coresight_add_conn(&adev->dev, pdata, &conns[i]);
+		if (rc)
+			return rc;
 	}
-
 	devm_kfree(&adev->dev, conns);
 	return 0;
 }
diff --git a/include/linux/coresight.h b/include/linux/coresight.h
index 64bb5fc95afa..47fa58d6981d 100644
--- a/include/linux/coresight.h
+++ b/include/linux/coresight.h
@@ -173,7 +173,9 @@  struct coresight_desc {
  *		 on the remote device.
  * @remote_fwnode: remote component's fwnode handle.
  * @remote_dev:  a @coresight_device representation of the component
- *		 connected to @port.
+ *               connected to @port. Will be looked up using
+ *               @remote_fwnode once the remote's coresight_device has
+ *               been created.
  * @link: Representation of the connection as a sysfs link.
  */
 struct coresight_connection {
@@ -614,5 +616,8 @@  static inline void coresight_write64(struct coresight_device *csdev, u64 val, u3
 extern int coresight_get_cpu(struct device *dev);
 
 struct coresight_platform_data *coresight_get_platform_data(struct device *dev);
+int coresight_add_conn(struct device *dev,
+		       struct coresight_platform_data *pdata,
+		       const struct coresight_connection *conn);
 
 #endif		/* _LINUX_COREISGHT_H */