diff mbox series

[2/2] s390/vfio-ccw: make mdev_types not look like a fake flex array

Message ID 20250217100614.3043620-3-pasic@linux.ibm.com (mailing list archive)
State New
Headers show
Series s390/vfio-*: make mdev_types unlike a fake flex array | expand

Commit Message

Halil Pasic Feb. 17, 2025, 10:06 a.m. UTC 
The vfio-ccw driver and the vfio parent device provided by it (parent)
support just a single mdev_type, and this is not likely to change any
time soon. To match the mdev interfaces nicely initially the choice was
made that mdev_types (which gets passed into mdev_register_parent())
shall be an array of pointers to struct mdev_type with a single element,
and to make things worse it ended up being the last member.

Now the problem with that is that before C99 the usual way to get
something similar to a flexible array member was to use a trailing array
of size 0 or 1. This is what I called fake flex array. For a while now
the community is trying to get rid of fake flex arrays. And while
mdev_types was not a fake flex array but an array of size one, because
it can easily be and probably was mistaken for a fake flex array it got
converted into a real C99 flex array with a compile time known constant
size of one.

As per [1] it was established that "only fake flexible arrays should be
transformed into C99 flex-array members". Since IMHO the entire point of
flex arrays is being flexible about the array size at run time, a C99
flex array is a poor fit for mdev_types.  But an array of a size one is
a poor fit as well for the reason stated above, let us try to get rid of
the flex array without introducing back the one sized array.

So, lets make mdev_types a pointer to struct mdev_type and pass in the
address of that pointer as the 4th formal parameter of
mdev_register_parent().

[1] https://lore.kernel.org/lkml/85863d7a-2d8b-4c1b-b76a-e2f40834a7a8@embeddedor.com/

Signed-off-by: Halil Pasic <pasic@linux.ibm.com>
Reviewed-by: Matthew Rosato <mjrosato@linux.ibm.com>
Tested-by: Eric Farman <farman@linux.ibm.com>

---

I've also considered switching up the order in which the members
mdev_types and mdev_type are defined in struct vfio_ccw_parent but
decided against that because that could look to somebody like
well known mistake that can be made when using fake flex arrays.
---
 drivers/s390/cio/vfio_ccw_drv.c     | 6 +++---
 drivers/s390/cio/vfio_ccw_private.h | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)
diff mbox series

Patch

diff --git a/drivers/s390/cio/vfio_ccw_drv.c b/drivers/s390/cio/vfio_ccw_drv.c
index 914dde041675..6ff5c9cfb7ed 100644
--- a/drivers/s390/cio/vfio_ccw_drv.c
+++ b/drivers/s390/cio/vfio_ccw_drv.c
@@ -171,7 +171,7 @@  static int vfio_ccw_sch_probe(struct subchannel *sch)
 		return -ENODEV;
 	}
 
-	parent = kzalloc(struct_size(parent, mdev_types, 1), GFP_KERNEL);
+	parent = kzalloc(sizeof(*parent), GFP_KERNEL);
 	if (!parent)
 		return -ENOMEM;
 
@@ -186,10 +186,10 @@  static int vfio_ccw_sch_probe(struct subchannel *sch)
 
 	parent->mdev_type.sysfs_name = "io";
 	parent->mdev_type.pretty_name = "I/O subchannel (Non-QDIO)";
-	parent->mdev_types[0] = &parent->mdev_type;
+	parent->mdev_types = &parent->mdev_type;
 	ret = mdev_register_parent(&parent->parent, &sch->dev,
 				   &vfio_ccw_mdev_driver,
-				   parent->mdev_types, 1);
+				   &parent->mdev_types, 1);
 	if (ret)
 		goto out_unreg;
 
diff --git a/drivers/s390/cio/vfio_ccw_private.h b/drivers/s390/cio/vfio_ccw_private.h
index b62bbc5c6376..0501d4bbcdbd 100644
--- a/drivers/s390/cio/vfio_ccw_private.h
+++ b/drivers/s390/cio/vfio_ccw_private.h
@@ -79,7 +79,7 @@  struct vfio_ccw_parent {
 
 	struct mdev_parent	parent;
 	struct mdev_type	mdev_type;
-	struct mdev_type	*mdev_types[];
+	struct mdev_type	*mdev_types;
 };
 
 /**