Hi Miquel,
On 11/10/2023 08:38, Miquel Raynal wrote:
Hi Srinivas,
srinivas.kandagatla@xxxxxxxxxx wrote on Mon, 9 Oct 2023 10:44:45 +0100:
On 05/10/2023 16:59, Miquel Raynal wrote:
Current layout support was initially written without modules support in
mind. When the requirement for module support rose, the existing base
was improved to adopt modularization support, but kind of a design flaw
was introduced. With the existing implementation, when a storage device
registers into NVMEM, the core tries to hook a layout (if any) and
populates its cells immediately. This means, if the hardware description
expects a layout to be hooked up, but no driver was provided for that,
the storage medium will fail to probe and try later from
scratch. Technically, the layouts are more like a "plus" and, even we
This is not true, As layouts are kind of resources for nvmem providers, Ideally the provider driver should defer if there is no matching layout available.
That is not possible as layouts are now devices, the device will be
populated but you cannot know when it will be actually probed?
Expressing this as a weak dependency is going to be an issue,
1. With creating the sysfs entries and user notifications
For me, this is not an issue. Greg?
2. nvmem consumers will be in a confused state with provider registered but without cells added yet.
Wow, I feel like we are moving backwards.
Consumers don't know about the nvmem devices, they just care about a
cell. If the cell isn't there, the consumer decides what it wants
to do with that.
We initially discussed that we would not EPROBE_DEFER if the layouts
were not yet available because the NVMEM device may be created from a
device that is the main storage and while you don't have your rootfs,
Does it not sound like we are not expressing the dependencies between
nvmem provider and layout drivers correctly?
you don't have access to your modules. And anyway it's probably a bad
idea to allow endless probe deferrals on your main storage device.
If the cells are not available at that time, it's not a huge deal? The
consumers will have to wait a bit more (or take any other action, this
is device dependent).
In this case the nvmem consumers will get an -ENOENT error, which is
very confusing TBH.
thanks,
Srini
--srini
consider that the hardware description shall be correct, we could still
probe the storage device (especially if it contains the rootfs).
One way to overcome this situation is to consider the layouts as
devices, and leverage the existing notifier mechanism. When a new NVMEM
device is registered, we can:
- populate its nvmem-layout child, if any
- try to modprobe the relevant driver, if relevant
- try to hook the NVMEM device with a layout in the notifier
And when a new layout is registered:
- try to hook all the existing NVMEM devices which are not yet hooked to
a layout with the new layout
This way, there is no strong order to enforce, any NVMEM device creation
or NVMEM layout driver insertion will be observed as a new event which
may lead to the creation of additional cells, without disturbing the
probes with costly (and sometimes endless) deferrals.
In order to achieve that goal we need:
* To keep track of all nvmem devices
* To create a new bus for the nvmem-layouts with minimal logic to match
nvmem-layout devices with nvmem-layout drivers.
All this infrastructure code is created in the layouts.c file.
Signed-off-by: Miquel Raynal <miquel.raynal@xxxxxxxxxxx>
Tested-by: Rafał Miłecki <rafal@xxxxxxxxxx>
---
Thanks,
Miquèl
