From: sashiko-bot@kernel.org
To: alucerop@amd.com, linux-cxl@vger.kernel.org
Subject: Re: [RFC 1/3] cxl: add tracepoints to the unwinding code path
Date: Wed, 18 Nov 2026 18:00:30 +0000 [thread overview]
Message-ID: <sashiko-outbox-221716@kernel.org> (raw)
In-Reply-To: <20261118191431.21781-2-alucerop@amd.com>
Thank you for your contribution! Sashiko AI review found 3 potential
issues to consider:
- [High] cxl: use-after-free in unwinding tracepoints after kfree() and
put_device()
- [Low] cxl: unmatched start tracepoint on early return in
__cxl_decoder_detach()
- [Low] cxl: missing multi-read guards and convention defects in
tracepoint headers
--
commit a8dc1dbd6b765af2c082df04014f6b4732355967
Author: Alejandro Lucero <alucerop@amd.com>
cxl: add tracepoints to the unwinding code path
This commit adds start and done tracepoints under the cxl_unwinding trace
system across CXL memdev, port, decoder, and region teardown paths.
> diff --git a/drivers/cxl/core/memdev.c b/drivers/cxl/core/memdev.c
> index 787988e92965a..7333fe7594bd9 100644
> --- a/drivers/cxl/core/memdev.c
> +++ b/drivers/cxl/core/memdev.c
[ ... ]
> @@ -27,9 +30,11 @@ static void cxl_memdev_release(struct device *dev)
> struct cxl_memdev *cxlmd = to_cxl_memdev(dev);
> struct device *parent = dev->parent;
>
> + trace_cxl_memdev_release(cxlmd, "start");
> ida_free(&cxl_memdev_ida, cxlmd->id);
> kfree(cxlmd);
> put_device(parent);
> + trace_cxl_memdev_release(cxlmd, "done");
> }
>
> static char *cxl_memdev_devnode(const struct device *dev, umode_t *mode, kuid_t *uid,
> @@ -752,9 +757,11 @@ static void cxl_memdev_unregister(void *_cxlmd)
> struct cxl_memdev *cxlmd = _cxlmd;
> struct device *dev = &cxlmd->dev;
>
> + trace_cxl_memdev_unregister(cxlmd, "start");
> cdev_device_del(&cxlmd->cdev, dev);
> cxl_memdev_shutdown(dev);
> put_device(dev);
> + trace_cxl_memdev_unregister(cxlmd, "done");
> }
>
> static void detach_memdev(struct work_struct *work)
> @@ -763,6 +770,7 @@ static void detach_memdev(struct work_struct *work)
>
> cxlmd = container_of(work, typeof(*cxlmd), detach_work);
>
> + trace_cxl_memdev_detach(cxlmd, "start");
> /*
> * When the creator of @cxlmd sets ->attach it indicates CXL operation
> * is required. In that case, @cxlmd detach escalates to parent device
> @@ -772,7 +780,9 @@ static void detach_memdev(struct work_struct *work)
> device_release_driver(cxlmd->dev.parent);
> else
> device_release_driver(&cxlmd->dev);
> +
> put_device(&cxlmd->dev);
> + trace_cxl_memdev_detach(cxlmd, "done");
> }
>
> static struct lock_class_key cxl_memdev_key;
> diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> index f3705ea44db77..9ccb0d1859047 100644
> --- a/drivers/cxl/core/port.c
> +++ b/drivers/cxl/core/port.c
[ ... ]
> @@ -540,8 +544,10 @@ EXPORT_SYMBOL_NS_GPL(to_cxl_switch_decoder, "CXL");
>
> static void cxl_ep_release(struct cxl_ep *ep)
> {
> + trace_cxl_ep_release(ep->ep, "start");
> put_device(ep->ep);
> kfree(ep);
> + trace_cxl_ep_release(ep->ep, "done");
> }
>
> static void cxl_ep_remove(struct cxl_port *port, struct cxl_ep *ep)
> @@ -558,6 +564,7 @@ static void cxl_port_release(struct device *dev)
> unsigned long index;
> struct cxl_ep *ep;
>
> + trace_cxl_port_release(dev, "start");
> xa_for_each(&port->endpoints, index, ep)
> cxl_ep_remove(port, ep);
> xa_destroy(&port->endpoints);
> @@ -571,6 +578,7 @@ static void cxl_port_release(struct device *dev)
> put_device(dev->parent);
> kfree(port);
> }
> + trace_cxl_port_release(dev, "done");
> }
>
> static ssize_t decoders_committed_show(struct device *dev,
> @@ -1478,6 +1486,8 @@ static void delete_endpoint(void *data)
> struct cxl_port *endpoint = cxlmd->endpoint;
> struct device *host = port_to_host(endpoint);
>
> + trace_cxl_delete_endpoint(&endpoint->dev, "start");
> +
> scoped_guard(device, host) {
> if (host->driver && !endpoint->dead) {
> devm_release_action(host, cxl_unlink_parent_dport, endpoint);
> @@ -1488,6 +1498,7 @@ static void delete_endpoint(void *data)
> }
> put_device(&endpoint->dev);
> put_device(host);
> + trace_cxl_delete_endpoint(&endpoint->dev, "done");
> }
[ ... ]
> @@ -2283,6 +2296,7 @@ EXPORT_SYMBOL_NS_GPL(cxl_decoder_add, "CXL");
>
> static void cxld_unregister(void *dev)
> {
> + trace_cxld_unregister(dev, "start");
> if (is_endpoint_decoder(dev))
> cxl_decoder_detach(NULL, to_cxl_endpoint_decoder(dev), -1,
> DETACH_INVALIDATE);
> @@ -2290,6 +2304,7 @@ static void cxld_unregister(void *dev)
> kill_regions(to_cxl_root_decoder(dev));
>
> device_unregister(dev);
> + trace_cxld_unregister(dev, "done");
> }
>
> int cxl_decoder_autoremove(struct device *host, struct cxl_decoder *cxld)
[ ... ]
> diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
> index a048bd5c06cb6..efbd53a37405d 100644
> --- a/drivers/cxl/core/region.c
> +++ b/drivers/cxl/core/region.c
[ ... ]
> @@ -2275,6 +2278,7 @@ __cxl_decoder_detach(struct cxl_region *cxlr,
> lockdep_assert_held_write(&cxl_rwsem.region);
>
> if (!cxled) {
> + trace_cxl_decoder_detach(&cxlr->dev, "start");
> p = &cxlr->params;
>
> if (pos >= p->interleave_ways) {
[Severity: Low]
Does emitting trace_cxl_decoder_detach(&cxlr->dev, "start") before
checking pos and p->targets[pos] leave an unmatched "start" trace event
when __cxl_decoder_detach() returns NULL early?
When unregister_region() loops over p->interleave_ways calling
detach_target(cxlr, i), any target already detached by cxld_unregister()
has p->targets[pos] == NULL and returns without a matching "done" event
(as does the dev_WARN_ONCE() return later in __cxl_decoder_detach()).
> @@ -2521,6 +2528,7 @@ static void cxl_region_release(struct device *dev)
> struct cxl_root_decoder *cxlrd = cxlr->cxlrd;
> int id = atomic_read(&cxlrd->region_id);
>
> + trace_cxl_region_release(dev, "start");
> /*
> * Try to reuse the recently idled id rather than the cached
> * next id to prevent the region id space from increasing
> @@ -2536,6 +2544,7 @@ static void cxl_region_release(struct device *dev)
> out:
> put_device(dev->parent);
> kfree(cxlr);
> + trace_cxl_region_release(dev, "done");
> }
>
> const struct device_type cxl_region_type = {
> @@ -2565,6 +2574,7 @@ static void unregister_region(struct cxl_region *cxlr)
> struct cxl_region_params *p = &cxlr->params;
> int i;
>
> + trace_cxl_unregister_region(&cxlr->dev, "start");
> xa_erase(&cxlrd->regions, cxlr->id);
> device_del(&cxlr->dev);
>
> @@ -2580,6 +2590,7 @@ static void unregister_region(struct cxl_region *cxlr)
>
> cxl_region_iomem_release(cxlr);
> put_device(&cxlr->dev);
> + trace_cxl_unregister_region(&cxlr->dev, "done");
> }
>
> static void endpoint_unregister_region(void *_cxlr)
> @@ -2587,10 +2598,12 @@ static void endpoint_unregister_region(void *_cxlr)
> struct cxl_region *cxlr = _cxlr;
> struct cxl_root_decoder *cxlrd = to_cxl_root_decoder(cxlr->dev.parent);
>
> + trace_cxl_endpoint_unregister_region(&cxlr->dev, "start");
> guard(mutex)(&cxlrd->regions_lock);
> if (xa_load(&cxlrd->regions, cxlr->id))
> unregister_region(cxlr);
> put_device(&cxlr->dev);
> + trace_cxl_endpoint_unregister_region(&cxlr->dev, "done");
> }
[Severity: High]
Do the "done" tracepoints in these release and unregister callbacks trigger
a use-after-free?
Each of these tracepoints reads dev->kobj.name (or cxlmd->dev.kobj.name),
but the "done" call runs after kfree(), put_device(), or
device_unregister() has already freed the object and its kobject name:
- cxl_ep_release() calls kfree(ep) after put_device(ep->ep), and then
trace_cxl_ep_release(ep->ep, "done") evaluates ep->ep from the freed
struct cxl_ep even when tracing is disabled.
- cxl_memdev_release(), cxl_port_release(), and cxl_region_release() call
kfree() on the enclosing struct before the "done" tracepoint reads
dev->kobj.name.
- cxl_memdev_unregister(), detach_memdev(), delete_endpoint(),
cxld_unregister(), unregister_region(), and
endpoint_unregister_region() drop the final device reference via
put_device() or device_unregister() before calling the "done"
tracepoint.
> diff --git a/drivers/cxl/cxl_memdev_tracepoints.h b/drivers/cxl/cxl_memdev_tracepoints.h
> new file mode 100644
> index 0000000000000..6e8206cf178a0
> --- /dev/null
> +++ b/drivers/cxl/cxl_memdev_tracepoints.h
> @@ -0,0 +1,58 @@
[ ... ]
> +#undef TRACE_SYSTEM
> +#define TRACE_SYSTEM cxl_unwinding
> +
> +#include <linux/tracepoint.h>
> +#include "cxlmem.h"
> +
> +TRACE_EVENT(cxl_memdev_release,
> + TP_PROTO(struct cxl_memdev *cxlmd, char *action),
> + TP_ARGS(cxlmd, action),
> + TP_STRUCT__entry(__string(devname, cxlmd->dev.kobj.name)
> + __string(action, action)
> + ),
> + TP_fast_assign(__assign_str(devname);
> + __assign_str(action);
> + ),
> + TP_printk("%s, %s", __get_str(devname), __get_str(action))
> +);
[ ... ]
> +#undef TRACE_INCLUDE_PATH
> +#define TRACE_INCLUDE_PATH .
> +#undef TRACE_INCLUDE_FILE
> +#define TRACE_INCLUDE_FILE cxl_memdev_tracepoints
> +
> +#include <trace/define_trace.h>
> diff --git a/drivers/cxl/cxl_port_tracepoints.h b/drivers/cxl/cxl_port_tracepoints.h
> new file mode 100644
> index 0000000000000..faefa418d9684
> --- /dev/null
> +++ b/drivers/cxl/cxl_port_tracepoints.h
> @@ -0,0 +1,91 @@
[ ... ]
> +TRACE_EVENT(schedule_cxl_memdev_detach,
> + TP_PROTO(struct cxl_memdev *cxlmd),
> + TP_ARGS(cxlmd),
> + TP_STRUCT__entry(__string(devname, cxlmd->dev.kobj.name)),
> + TP_fast_assign(__assign_str(devname);
> + ),
> + TP_printk("%s", __get_str(devname))
> +);
[ ... ]
> +TRACE_EVENT(cxld_unregister,
> + TP_PROTO(struct device *dev, char *action),
[ ... ]
> diff --git a/drivers/cxl/cxl_region_tracepoints.h b/drivers/cxl/cxl_region_tracepoints.h
> new file mode 100644
> index 0000000000000..977eb45b6aad8
> --- /dev/null
> +++ b/drivers/cxl/cxl_region_tracepoints.h
> @@ -0,0 +1,78 @@
[ ... ]
> +#undef TRACE_SYSTEM
> +#define TRACE_SYSTEM cxl_unwinding
> +
> +#include <linux/tracepoint.h>
> +
> +TRACE_EVENT(cxl_decoder_detach,
> + TP_PROTO(struct device *dev, char *action),
[Severity: Low]
Should these new tracepoint headers wrap their definitions in a
"#if !defined(...) || defined(TRACE_HEADER_MULTI_READ)" guard with #endif
placed before #include <trace/define_trace.h>, like
drivers/cxl/core/trace.h?
Also, schedule_cxl_memdev_detach and cxld_unregister do not start with the
cxl_ subsystem prefix, the events read dev->kobj.name directly instead of
using dev_name(dev) (and take char *action instead of const char *action),
and the twelve identical (dev, action) TRACE_EVENT definitions could share
a DECLARE_EVENT_CLASS in drivers/cxl/core/trace.h.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261118191431.21781-2-alucerop@amd.com?part=1
next prev parent reply other threads:[~2026-11-18 18:00 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-18 19:14 [RFC 0/3] Type2 unwinding ordering change alucerop
2026-11-18 19:14 ` [RFC 1/3] cxl: add tracepoints to the unwinding code path alucerop
2026-11-18 18:00 ` sashiko-bot [this message]
2026-11-18 19:14 ` [RFC 2/3] cxl: make explicit removal of endpoint and decoder for Type2 alucerop
2026-11-18 18:09 ` sashiko-bot
2026-11-18 19:14 ` [RFC 3/3] Documentation/cxl: add document about Type2 binding and unbinding alucerop
2026-11-18 18:04 ` sashiko-bot
2026-11-18 23:39 ` [RFC 0/3] Type2 unwinding ordering change Dave Jiang
2026-11-19 8:13 ` Lucero Palau, Alejandro
2026-11-19 10:00 ` Lucero Palau, Alejandro
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=sashiko-outbox-221716@kernel.org \
--to=sashiko-bot@kernel.org \
--cc=alucerop@amd.com \
--cc=linux-cxl@vger.kernel.org \
--cc=sashiko-reviews@lists.linux.dev \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox