From: Boris Brezillon <boris.brezillon@xxxxxxxxxxxxx>
Date: Thu, Aug 29, 2019 at 15:35:20
> On Thu, 29 Aug 2019 13:53:24 +0000
> Vitor Soares <Vitor.Soares@xxxxxxxxxxxx> wrote:
>
> > Hi Boris,
> >
> > From: Boris Brezillon <boris.brezillon@xxxxxxxxxxxxx>
> > Date: Thu, Aug 29, 2019 at 11:41:15
> >
> > > +Przemek
> > >
> > > Please try to Cc active I3C contributors so they get a chance to
> > > comment on your patches.
> >
> > I can do that next time.
> >
> > >
> > > On Thu, 29 Aug 2019 12:19:32 +0200
> > > Vitor Soares <Vitor.Soares@xxxxxxxxxxxx> wrote:
> > >
> > > > On pre_assing_dyn_addr() the devices that fail:
> > > > i3c_master_setdasa_locked()
> > > > i3c_master_reattach_i3c_dev()
> > > > i3c_master_retrieve_dev_info()
> > > >
> > > > are kept in memory and master->bus.devs list. This makes the i3c devices
> > > > without a dynamic address are sent on DEFSLVS CCC command. Fix this by
> > > > detaching and freeing the devices that fail on pre_assign_dyn_addr().
> > >
> > > I don't think removing those entries is a good strategy, as one might
> > > want to try to use a different dynamic address if the requested one
> > > is not available.
> >
> > Do you mean same 'assigned-address' attribute in DT?
>
> Yes, or say it's another device that got the address we want and this
> device doesn't want to release the address (I'm assuming the !SA case).
>
> >
> > If so, it is checked here:
> >
> > static int i3c_master_bus_init(struct i3c_master_controller *master)
> > ...
> > list_for_each_entry(i3cboardinfo, &master->boardinfo.i3c, node) {
> > struct i3c_device_info info = {
> > .static_addr = i3cboardinfo->static_addr,
> > };
> >
> > if (i3cboardinfo->init_dyn_addr) {
> > status = i3c_bus_get_addr_slot_status(&master->bus,
> > ^
> > i3cboardinfo->init_dyn_addr);
> > if (status != I3C_ADDR_SLOT_FREE) {
> > ret = -EBUSY;
> > goto err_detach_devs;
> > }
> > }
> >
> > i3cdev = i3c_master_alloc_i3c_dev(master, &info);
> > if (IS_ERR(i3cdev)) {
> > ret = PTR_ERR(i3cdev);
> > goto err_detach_devs;
> > }
> >
> > i3cdev->boardinfo = i3cboardinfo;
> >
> > ret = i3c_master_attach_i3c_dev(master, i3cdev);
> > if (ret) {
> > i3c_master_free_i3c_dev(i3cdev);
> > goto err_detach_devs;
> > }
> > }
> > ...
> >
> > and later if it fails i3c_master_pre_assign_dyn_addr(), the device can
> > participate in Enter Dynamic Address Assignment process.
> > I may need to check the boardinfo->init_dyn_addr status on
> > i3c_master_add_i3c_dev_locked before i3c_master_setnewda_locked().
>
> I need to double check but I thought we were already handling that case
> properly.
Yes, it is handled in the code above.
>
> >
> > > Why not simply skipping entries that have ->dyn_addr
> > > set to 0 when preparing a DEFSLVS frame
> >
> > I considered that solution too but if the device isn't enumerated why
> > should it be attached and kept in memory?
>
> Might be a device that supports HJ, and in that case we might want the
> controller to reserve a slot in its device table for that device.
> Anyway, it doesn't hurt to have it around as long as we don't pass the
> device through DEFSLVS if it doesn't have a dynamic address. I really
> prefer to keep the logic unchanged and fix it if it needs to be fixed.
Well, we aren't reserving a slot because we need another one to attach
the device when it is enumerated and hence a device may be using 2 slots
in the controller.
It may cause problems in HC with reduced slots and it is another reason
why I think we should detach device without dynamic address after the
enumeration phase.
>
> > Anyway we have i3c_boardinfo
> > with DT information.
> >
> > >
> > > >
> > > > Signed-off-by: Vitor Soares <vitor.soares@xxxxxxxxxxxx>
> > > > ---
> > > > drivers/i3c/master.c | 11 ++++++++---
> > > > 1 file changed, 8 insertions(+), 3 deletions(-)
> > > >
> > > > diff --git a/drivers/i3c/master.c b/drivers/i3c/master.c
> > > > index 5f4bd52..4d29e1f 100644
> > > > --- a/drivers/i3c/master.c
> > > > +++ b/drivers/i3c/master.c
> > > > @@ -1438,7 +1438,7 @@ static void i3c_master_pre_assign_dyn_addr(struct i3c_dev_desc *dev)
> > > > ret = i3c_master_setdasa_locked(master, dev->info.static_addr,
> > > > dev->boardinfo->init_dyn_addr);
> > > > if (ret)
> > > > - return;
> > > > + goto err_detach_dev;
> > > >
> > > > dev->info.dyn_addr = dev->boardinfo->init_dyn_addr;
> > > > ret = i3c_master_reattach_i3c_dev(dev, 0);
> > > > @@ -1453,6 +1453,10 @@ static void i3c_master_pre_assign_dyn_addr(struct i3c_dev_desc *dev)
> > > >
> > > > err_rstdaa:
> > > > i3c_master_rstdaa_locked(master, dev->boardinfo->init_dyn_addr);
> > > > +
> > > > +err_detach_dev:
> > > > + i3c_master_detach_i3c_dev(dev);
> > > > + i3c_master_free_i3c_dev(dev);
> > >
> > > We certainly shouldn't detach/free the i3c_dev_desc from here. If it
> > > has to be done somewhere (which I'd like to avoid), it should be done
> > > in i3c_master_bus_init() (i3c_master_pre_assign_dyn_addr() should be
> > > converted to return an int in that case).
> >
> > I can change it to return an error.
> >
> > >
> > > > }
> > > >
> > > > static void
> > > > @@ -1647,7 +1651,7 @@ static int i3c_master_bus_init(struct i3c_master_controller *master)
> > > > enum i3c_addr_slot_status status;
> > > > struct i2c_dev_boardinfo *i2cboardinfo;
> > > > struct i3c_dev_boardinfo *i3cboardinfo;
> > > > - struct i3c_dev_desc *i3cdev;
> > > > + struct i3c_dev_desc *i3cdev, *i3ctmp;
> > > > struct i2c_dev_desc *i2cdev;
> > > > int ret;
> > > >
> > > > @@ -1746,7 +1750,8 @@ static int i3c_master_bus_init(struct i3c_master_controller *master)
> > > > * Pre-assign dynamic address and retrieve device information if
> > > > * needed.
> > > > */
> > > > - i3c_bus_for_each_i3cdev(&master->bus, i3cdev)
> > > > + list_for_each_entry_safe(i3cdev, i3ctmp, &master->bus.devs.i3c,
> > > > + common.node)
> > > > i3c_master_pre_assign_dyn_addr(i3cdev);
> > > >
> > > > ret = i3c_master_do_daa(master);
> >
> > Thank for your feedback.
> >
> > Best regards,
> > Vitor Soares
