[SRU][N][PATCH v2 1/1] coresight: Remove atomic type from refcnt

Noah Wager noah.wager at canonical.com
Tue Feb 24 18:06:40 UTC 2026


On Tue, Feb 24, 2026 at 09:56:10AM +0100, Stefan Bader wrote:
> On 24/02/2026 08:33, Noah Wager wrote:
> > From: James Clark <james.clark at arm.com>
> >
> > Refcnt is only ever accessed from either inside the coresight_mutex, or
> > the device's spinlock, making the atomic type and atomic_dec_return()
>
> I would say above is the hard question. Is refcnt really ever accessed under
> protection in the 6.8 code base? It is probably hard to validate as access
> is happening around enable / disable operations. Not sure how often that
> takes places. Ensuring the assumption is true is the problematic part and
> usually too complicated to be done within the time limits of upstream stable
> work. I don't want to say no as you spent a lot of time on the backport
> already. It would be good to point out that it has been confirmed that the
> accessors all have some lock protection.
>
> The alternative to this would be to adjust / fix up the upstream stable
> patch. As
>
> coresight: catu: Introduce refcount and spinlock for enabling/disabling
>
> primarily wants to introduce refcount/lock protection it might just be safer
> to adapt the freshly introduced code to what everything around it does
> instead of the other way round. So just replacing direct access to refcnt by
> atomic_* functions.
>
> -Stefan
>

All references of refcnt occur in the same scope as an acquired spinlock
except one, which I mentioned inline below. Going by the author's word,
it seems likely all the references are protected for this tree, but I
can't say for certain.

A SAUCE patch, that adds the missing atomic_read/inc() calls to the
breaking patch, would indeed be the smallest change to the tree. I did
want to avoid a SAUCE patch if possible, but you raise a good point in
that I am putting a lot of trust in the assumption that the commit
message applies to the noble tree.

I guess my reply doesn't have any concrete answers, so I'd like your
advice: do you think the small SAUCE fixup is the way to go, as opposed
to trying to do backport acrobatics for this particular issue?

- Noah

> > calls confusing and unnecessary. The only point of synchronisation
> > outside of these two types of locks is already done with a compare and
> > swap on 'mode', which a comment has been added for.
> >
> > There was one instance of refcnt being used outside of a lock in TPIU,
> > but that can easily be fixed by making it the same as all the other
> > devices and adding a spinlock. Potentially in the future all the
> > refcounting and locking can be moved up into the core code, and all the
> > mostly duplicate code from the individual devices can be removed.
> >
> > Signed-off-by: James Clark <james.clark at arm.com>
> > Link: https://lore.kernel.org/r/20240129154050.569566-8-james.clark@arm.com
> > Signed-off-by: Suzuki K Poulose <suzuki.poulose at arm.com>
> > (backported from commit 4545b38ef004a586295750ea49a505b6396a7c90)
> > [nwager: Context changes due to missing patchset:
> >   "coresight: Separate sysfs and Perf usage and some other cleanups".
> >   Only this commit was a missing dependency of "coresight: catu: Introduce
> >   refcount and spinlock for enabling/disabling", and since this change is
> >   simple (basically a find/replace of refcnt atomic_read/inc to int accesses),
> >   I just fixed up the context changes instead of porting the five previous
> >   patches from the series. Here is the gist of the changes I made:
> >   - previous line context changes due to different way of checking CS_MODE_SYSFS
> >   - some occurrences were removed before this patch arrived upstream, so
> >     replace those occurrences in my fixup
> >   - upstream moved some hunks to coresight-sysfs.c, so I needed to
> >     replace the occurrences in the original locations
> >   - drop the 'mode' comment due to missing commit from series]
> > Signed-off-by: Noah Wager <noah.wager at canonical.com>
> >
> > Signed-off-by: Noah Wager <noah.wager at canonical.com>
> > ---
> >   drivers/hwtracing/coresight/coresight-core.c  |  9 +++++----
> >   drivers/hwtracing/coresight/coresight-etb10.c | 11 +++++-----
> >   .../hwtracing/coresight/coresight-tmc-etf.c   | 20 ++++++++++---------
> >   .../hwtracing/coresight/coresight-tmc-etr.c   | 13 ++++++------
> >   drivers/hwtracing/coresight/coresight-tpiu.c  | 14 +++++++++++--
> >   drivers/hwtracing/coresight/ultrasoc-smb.c    |  9 +++++----
> >   include/linux/coresight.h                     |  7 +++++--
> >   7 files changed, 51 insertions(+), 32 deletions(-)
> >
> > diff --git a/drivers/hwtracing/coresight/coresight-core.c b/drivers/hwtracing/coresight/coresight-core.c
> > index fc0e7f18d95b..52d50545d5c9 100644
> > --- a/drivers/hwtracing/coresight/coresight-core.c
> > +++ b/drivers/hwtracing/coresight/coresight-core.c
> > @@ -370,7 +370,7 @@ static void coresight_disable_link(struct coresight_device *csdev,
> >   			    0)
> >   				return;
> >   	} else {
> > -		if (atomic_read(&csdev->refcnt) != 0)
> > +		if (csdev->refcnt != 0)
> >   			return;

This is the only use of csdev->refcnt that doesn't explicitly acquire a
lock in the same scope. However, the line just below it is an access of
csdev->enable, which is said to only be used when locked in the upstream
commit that removes the ->enable field (from the same patch series):

d5e83f97eb56 ("coresight: Remove the 'enable' field.")

The 'enable' field has existed from the beginning, so I believe the
author's explanation should hold for 6.8. However, you are right in that
it is unlikely I will be able to independently verify this, as I am not
a coresight expert and it is difficult to make sweeping generalizations
about the control flow of enable/disable events.

> >   	}
> > @@ -391,7 +391,7 @@ int coresight_enable_source(struct coresight_device *csdev, enum cs_mode mode,
> >   		csdev->enable = true;
> >   	}
> > -	atomic_inc(&csdev->refcnt);
> > +	csdev->refcnt--;
> >   	return 0;
> >   }
> > @@ -473,7 +473,8 @@ EXPORT_SYMBOL_GPL(coresight_disable_source);
> >   static bool coresight_disable_source_sysfs(struct coresight_device *csdev,
> >   					   void *data)
> >   {
> > -	if (atomic_dec_return(&csdev->refcnt) == 0) {
> > +	csdev->refcnt--;
> > +	if (csdev->refcnt == 0) {
> >   		coresight_disable_source(csdev, data);
> >   		csdev->enable = false;
> >   	}
> > @@ -1142,7 +1143,7 @@ int coresight_enable(struct coresight_device *csdev)
> >   		 * source is already enabled.
> >   		 */
> >   		if (subtype == CORESIGHT_DEV_SUBTYPE_SOURCE_SOFTWARE)
> > -			atomic_inc(&csdev->refcnt);
> > +			csdev->refcnt++;
> >   		goto out;
> >   	}
> > diff --git a/drivers/hwtracing/coresight/coresight-etb10.c b/drivers/hwtracing/coresight/coresight-etb10.c
> > index d83d52dbd1a2..b2d0069bd21c 100644
> > --- a/drivers/hwtracing/coresight/coresight-etb10.c
> > +++ b/drivers/hwtracing/coresight/coresight-etb10.c
> > @@ -163,7 +163,7 @@ static int etb_enable_sysfs(struct coresight_device *csdev)
> >   		drvdata->mode = CS_MODE_SYSFS;
> >   	}
> > -	atomic_inc(&csdev->refcnt);
> > +	csdev->refcnt++;
> >   out:
> >   	raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
> >   	return ret;
> > @@ -199,7 +199,7 @@ static int etb_enable_perf(struct coresight_device *csdev, void *data)
> >   	 * use for this session.
> >   	 */
> >   	if (drvdata->pid == pid) {
> > -		atomic_inc(&csdev->refcnt);
> > +		csdev->refcnt++;
> >   		goto out;
> >   	}
> > @@ -217,7 +217,7 @@ static int etb_enable_perf(struct coresight_device *csdev, void *data)
> >   		/* Associate with monitored process. */
> >   		drvdata->pid = pid;
> >   		drvdata->mode = CS_MODE_PERF;
> > -		atomic_inc(&csdev->refcnt);
> > +		csdev->refcnt++;
> >   	}
> >   out:
> > @@ -356,7 +356,8 @@ static int etb_disable(struct coresight_device *csdev)
> >   	raw_spin_lock_irqsave(&drvdata->spinlock, flags);
> > -	if (atomic_dec_return(&csdev->refcnt)) {
> > +	csdev->refcnt--;
> > +	if (csdev->refcnt) {
> >   		raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
> >   		return -EBUSY;
> >   	}
> > @@ -447,7 +448,7 @@ static unsigned long etb_update_buffer(struct coresight_device *csdev,
> >   	raw_spin_lock_irqsave(&drvdata->spinlock, flags);
> >   	/* Don't do anything if another tracer is using this sink */
> > -	if (atomic_read(&csdev->refcnt) != 1)
> > +	if (csdev->refcnt != 1)
> >   		goto out;
> >   	__etb_disable_hw(drvdata);
> > diff --git a/drivers/hwtracing/coresight/coresight-tmc-etf.c b/drivers/hwtracing/coresight/coresight-tmc-etf.c
> > index 7406b65e2cdd..b7ecf660309a 100644
> > --- a/drivers/hwtracing/coresight/coresight-tmc-etf.c
> > +++ b/drivers/hwtracing/coresight/coresight-tmc-etf.c
> > @@ -206,7 +206,7 @@ static int tmc_enable_etf_sink_sysfs(struct coresight_device *csdev)
> >   	 * touched.
> >   	 */
> >   	if (drvdata->mode == CS_MODE_SYSFS) {
> > -		atomic_inc(&csdev->refcnt);
> > +		csdev->refcnt++;
> >   		goto out;
> >   	}
> > @@ -229,7 +229,7 @@ static int tmc_enable_etf_sink_sysfs(struct coresight_device *csdev)
> >   	ret = tmc_etb_enable_hw(drvdata);
> >   	if (!ret) {
> >   		drvdata->mode = CS_MODE_SYSFS;
> > -		atomic_inc(&csdev->refcnt);
> > +		csdev->refcnt++;
> >   	} else {
> >   		/* Free up the buffer if we failed to enable */
> >   		used = false;
> > @@ -284,7 +284,7 @@ static int tmc_enable_etf_sink_perf(struct coresight_device *csdev, void *data)
> >   		 * use for this session.
> >   		 */
> >   		if (drvdata->pid == pid) {
> > -			atomic_inc(&csdev->refcnt);
> > +			csdev->refcnt++;
> >   			break;
> >   		}
> > @@ -293,7 +293,7 @@ static int tmc_enable_etf_sink_perf(struct coresight_device *csdev, void *data)
> >   			/* Associate with monitored process. */
> >   			drvdata->pid = pid;
> >   			drvdata->mode = CS_MODE_PERF;
> > -			atomic_inc(&csdev->refcnt);
> > +			csdev->refcnt++;
> >   		}
> >   	} while (0);
> >   	spin_unlock_irqrestore(&drvdata->spinlock, flags);
> > @@ -338,7 +338,8 @@ static int tmc_disable_etf_sink(struct coresight_device *csdev)
> >   		return -EBUSY;
> >   	}
> > -	if (atomic_dec_return(&csdev->refcnt)) {
> > +	csdev->refcnt--;
> > +	if (csdev->refcnt) {
> >   		spin_unlock_irqrestore(&drvdata->spinlock, flags);
> >   		return -EBUSY;
> >   	}
> > @@ -371,7 +372,7 @@ static int tmc_enable_etf_link(struct coresight_device *csdev,
> >   		return -EBUSY;
> >   	}
> > -	if (atomic_read(&csdev->refcnt) == 0) {
> > +	if (csdev->refcnt == 0) {
> >   		ret = tmc_etf_enable_hw(drvdata);
> >   		if (!ret) {
> >   			drvdata->mode = CS_MODE_SYSFS;
> > @@ -379,7 +380,7 @@ static int tmc_enable_etf_link(struct coresight_device *csdev,
> >   		}
> >   	}
> >   	if (!ret)
> > -		atomic_inc(&csdev->refcnt);
> > +		csdev->refcnt++;
> >   	spin_unlock_irqrestore(&drvdata->spinlock, flags);
> >   	if (first_enable)
> > @@ -401,7 +402,8 @@ static void tmc_disable_etf_link(struct coresight_device *csdev,
> >   		return;
> >   	}
> > -	if (atomic_dec_return(&csdev->refcnt) == 0) {
> > +	csdev->refcnt--;
> > +	if (csdev->refcnt == 0) {
> >   		tmc_etf_disable_hw(drvdata);
> >   		drvdata->mode = CS_MODE_DISABLED;
> >   		last_disable = true;
> > @@ -489,7 +491,7 @@ static unsigned long tmc_update_etf_buffer(struct coresight_device *csdev,
> >   	spin_lock_irqsave(&drvdata->spinlock, flags);
> >   	/* Don't do anything if another tracer is using this sink */
> > -	if (atomic_read(&csdev->refcnt) != 1)
> > +	if (csdev->refcnt != 1)
> >   		goto out;
> >   	CS_UNLOCK(drvdata->base);
> > diff --git a/drivers/hwtracing/coresight/coresight-tmc-etr.c b/drivers/hwtracing/coresight/coresight-tmc-etr.c
> > index 52c55e8172ee..b1c55e7d3d3b 100644
> > --- a/drivers/hwtracing/coresight/coresight-tmc-etr.c
> > +++ b/drivers/hwtracing/coresight/coresight-tmc-etr.c
> > @@ -1231,14 +1231,14 @@ static int tmc_enable_etr_sink_sysfs(struct coresight_device *csdev)
> >   	 * touched, even if the buffer size has changed.
> >   	 */
> >   	if (drvdata->mode == CS_MODE_SYSFS) {
> > -		atomic_inc(&csdev->refcnt);
> > +		csdev->refcnt++;
> >   		goto out;
> >   	}
> >   	ret = tmc_etr_enable_hw(drvdata, sysfs_buf);
> >   	if (!ret) {
> >   		drvdata->mode = CS_MODE_SYSFS;
> > -		atomic_inc(&csdev->refcnt);
> > +		csdev->refcnt++;
> >   	}
> >   out:
> > @@ -1564,7 +1564,7 @@ tmc_update_etr_buffer(struct coresight_device *csdev,
> >   	spin_lock_irqsave(&drvdata->spinlock, flags);
> >   	/* Don't do anything if another tracer is using this sink */
> > -	if (atomic_read(&csdev->refcnt) != 1) {
> > +	if (csdev->refcnt != 1) {
> >   		spin_unlock_irqrestore(&drvdata->spinlock, flags);
> >   		goto out;
> >   	}
> > @@ -1676,7 +1676,7 @@ static int tmc_enable_etr_sink_perf(struct coresight_device *csdev, void *data)
> >   	 * use for this session.
> >   	 */
> >   	if (drvdata->pid == pid) {
> > -		atomic_inc(&csdev->refcnt);
> > +		csdev->refcnt++;
> >   		goto unlock_out;
> >   	}
> > @@ -1686,7 +1686,7 @@ static int tmc_enable_etr_sink_perf(struct coresight_device *csdev, void *data)
> >   		drvdata->pid = pid;
> >   		drvdata->mode = CS_MODE_PERF;
> >   		drvdata->perf_buf = etr_perf->etr_buf;
> > -		atomic_inc(&csdev->refcnt);
> > +		csdev->refcnt++;
> >   	}
> >   unlock_out:
> > @@ -1719,7 +1719,8 @@ static int tmc_disable_etr_sink(struct coresight_device *csdev)
> >   		return -EBUSY;
> >   	}
> > -	if (atomic_dec_return(&csdev->refcnt)) {
> > +	csdev->refcnt--;
> > +	if (csdev->refcnt) {
> >   		spin_unlock_irqrestore(&drvdata->spinlock, flags);
> >   		return -EBUSY;
> >   	}
> > diff --git a/drivers/hwtracing/coresight/coresight-tpiu.c b/drivers/hwtracing/coresight/coresight-tpiu.c
> > index 59eac93fd6bb..c23a6b9b41fe 100644
> > --- a/drivers/hwtracing/coresight/coresight-tpiu.c
> > +++ b/drivers/hwtracing/coresight/coresight-tpiu.c
> > @@ -58,6 +58,7 @@ struct tpiu_drvdata {
> >   	void __iomem		*base;
> >   	struct clk		*atclk;
> >   	struct coresight_device	*csdev;
> > +	spinlock_t		spinlock;
> >   };
> >   static void tpiu_enable_hw(struct csdev_access *csa)
> > @@ -72,8 +73,11 @@ static void tpiu_enable_hw(struct csdev_access *csa)
> >   static int tpiu_enable(struct coresight_device *csdev, enum cs_mode mode,
> >   		       void *__unused)
> >   {
> > +	struct tpiu_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
> > +
> > +	guard(spinlock)(&drvdata->spinlock);
> >   	tpiu_enable_hw(&csdev->access);
> > -	atomic_inc(&csdev->refcnt);
> > +	csdev->refcnt++;
> >   	dev_dbg(&csdev->dev, "TPIU enabled\n");
> >   	return 0;
> >   }
> > @@ -96,7 +100,11 @@ static void tpiu_disable_hw(struct csdev_access *csa)
> >   static int tpiu_disable(struct coresight_device *csdev)
> >   {
> > -	if (atomic_dec_return(&csdev->refcnt))
> > +	struct tpiu_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
> > +
> > +	guard(spinlock)(&drvdata->spinlock);
> > +	csdev->refcnt--;
> > +	if (csdev->refcnt)
> >   		return -EBUSY;
> >   	tpiu_disable_hw(&csdev->access);
> > @@ -132,6 +140,8 @@ static int tpiu_probe(struct amba_device *adev, const struct amba_id *id)
> >   	if (!drvdata)
> >   		return -ENOMEM;
> > +	spin_lock_init(&drvdata->spinlock);
> > +
> >   	drvdata->atclk = devm_clk_get(&adev->dev, "atclk"); /* optional */
> >   	if (!IS_ERR(drvdata->atclk)) {
> >   		ret = clk_prepare_enable(drvdata->atclk);
> > diff --git a/drivers/hwtracing/coresight/ultrasoc-smb.c b/drivers/hwtracing/coresight/ultrasoc-smb.c
> > index 10e886455b8b..2c1227fd701b 100644
> > --- a/drivers/hwtracing/coresight/ultrasoc-smb.c
> > +++ b/drivers/hwtracing/coresight/ultrasoc-smb.c
> > @@ -103,7 +103,7 @@ static int smb_open(struct inode *inode, struct file *file)
> >   	if (drvdata->reading)
> >   		return -EBUSY;
> > -	if (atomic_read(&drvdata->csdev->refcnt))
> > +	if (drvdata->csdev->refcnt)
> >   		return -EBUSY;
> >   	smb_update_data_size(drvdata);
> > @@ -270,7 +270,7 @@ static int smb_enable(struct coresight_device *csdev, enum cs_mode mode,
> >   	if (ret)
> >   		return ret;
> > -	atomic_inc(&csdev->refcnt);
> > +	csdev->refcnt++;
> >   	dev_dbg(&csdev->dev, "Ultrasoc SMB enabled\n");
> >   	return ret;
> > @@ -285,7 +285,8 @@ static int smb_disable(struct coresight_device *csdev)
> >   	if (drvdata->reading)
> >   		return -EBUSY;
> > -	if (atomic_dec_return(&csdev->refcnt))
> > +	csdev->refcnt--;
> > +	if (csdev->refcnt)
> >   		return -EBUSY;
> >   	/* Complain if we (somehow) got out of sync */
> > @@ -380,7 +381,7 @@ static unsigned long smb_update_buffer(struct coresight_device *csdev,
> >   	guard(spinlock)(&drvdata->spinlock);
> >   	/* Don't do anything if another tracer is using this sink. */
> > -	if (atomic_read(&csdev->refcnt) != 1)
> > +	if (csdev->refcnt != 1)
> >   		return 0;
> >   	smb_disable_hw(drvdata);
> > diff --git a/include/linux/coresight.h b/include/linux/coresight.h
> > index 4c9ff95c2c87..7a908907a652 100644
> > --- a/include/linux/coresight.h
> > +++ b/include/linux/coresight.h
> > @@ -226,7 +226,10 @@ struct coresight_sysfs_link {
> >    *		by @coresight_ops.
> >    * @access:	Device i/o access abstraction for this device.
> >    * @dev:	The device entity associated to this component.
> > - * @refcnt:	keep track of what is in use.
> > + * @refcnt:	keep track of what is in use. Only access this outside of the
> > + *		device's spinlock when the coresight_mutex held and mode ==
> > + *		CS_MODE_SYSFS. Otherwise it must be accessed from inside the
> > + *		spinlock.
> >    * @orphan:	true if the component has connections that haven't been linked.
> >    * @enable:	'true' if component is currently part of an active path.
> >    * @activated:	'true' only if a _sink_ has been activated.  A sink can be
> > @@ -250,7 +253,7 @@ struct coresight_device {
> >   	const struct coresight_ops *ops;
> >   	struct csdev_access access;
> >   	struct device dev;
> > -	atomic_t refcnt;
> > +	int refcnt;
> >   	bool orphan;
> >   	bool enable;	/* true only if configured as part of a path */
> >   	/* sink specific fields */
>



More information about the kernel-team mailing list