Hi Heinrich,
On Mon, 27 Mar 2023 at 19:48, Heinrich Schuchardt heinrich.schuchardt@canonical.com wrote:
On 3/27/23 06:00, Simon Glass wrote:
Hi Heinrich,
On Mon, 27 Mar 2023 at 06:27, Heinrich Schuchardt heinrich.schuchardt@canonical.com wrote:
Currently the device paths don't match the dm tree. We should create a device path node per dm tree node.
Signed-off-by: Heinrich Schuchardt heinrich.schuchardt@canonical.com
include/dm/uclass.h | 5 +++++ 1 file changed, 5 insertions(+)
This affects very few uclasses but adds a field to all of them. I think an event might be best for this. We can add an event spy in each of the affected uclasses.
EVENT_SPY does not allow to return information to the emitter of the event.
You just add it into unio event_data
event_notfify() does not allow to select a single recipient.
Do you need that? You can check whether the uclass matches, for example.
In struct uclass_driver there are many other fields that are rarely used:
int (*post_bind)(struct udevice *dev); int (*pre_unbind)(struct udevice *dev); int (*pre_probe)(struct udevice *dev); int (*post_probe)(struct udevice *dev); int (*pre_remove)(struct udevice *dev); int (*child_post_bind)(struct udevice *dev); int (*child_pre_probe)(struct udevice *dev); int (*child_post_probe)(struct udevice *dev);If we wanted to save memory in linker generated lists we would convert structures with pointers to variable size arrays e.g.
struct ops * = { { POST_BIND, post_bind }, { CHILD_POST_BIND, child_post_bind}, { END, NULL }, }
Yes. We have the same problem with devices. We have a way to resolve this today, using tags. We use DM_TAG_EFI and there are tags for the devices but support for using them to reduce the struct sizes is not yet added.
We might have a problem with efficiency, since direct pointers are faster, but of course we can address that using a suitable data structure.
Regards, Simon