| Message ID | 20240411235623.1260061-3-saravanak@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | fw_devlink overlay fix | expand |
On Thu, Apr 11, 2024 at 6:56 PM Saravana Kannan <saravanak@google.com> wrote: > > When an overlay is applied, if the target device has already probed > successfully and bound to a device, then some of the fw_devlink logic > that ran when the device was probed needs to be rerun. This allows newly > created dangling consumers of the overlayed device tree nodes to be > moved to become consumers of the target device. > > Fixes: 1a50d9403fb9 ("treewide: Fix probing of devices in DT overlays") > Reported-by: Herve Codina <herve.codina@bootlin.com> > Closes: https://lore.kernel.org/lkml/CAMuHMdXEnSD4rRJ-o90x4OprUacN_rJgyo8x6=9F9rZ+-KzjOg@mail.gmail.com/ > Closes: https://lore.kernel.org/all/20240221095137.616d2aaa@bootlin.com/ > Closes: https://lore.kernel.org/lkml/20240312151835.29ef62a0@bootlin.com/ > Signed-off-by: Saravana Kannan <saravanak@google.com> > --- > drivers/base/core.c | 76 +++++++++++++++++++++++++++++++++++++----- > drivers/of/overlay.c | 15 +++++++++ > include/linux/fwnode.h | 1 + > 3 files changed, 83 insertions(+), 9 deletions(-) > > diff --git a/drivers/base/core.c b/drivers/base/core.c > index 5f4e03336e68..1a646f393dd7 100644 > --- a/drivers/base/core.c > +++ b/drivers/base/core.c > @@ -46,6 +46,8 @@ static bool fw_devlink_drv_reg_done; > static bool fw_devlink_best_effort; > static struct workqueue_struct *device_link_wq; > > +#define get_dev_from_fwnode(fwnode) get_device((fwnode)->dev) I think it is better to not have this wrapper. We want it to be clear when we're acquiring a ref. I know get_device() does that, but I have to look up what get_dev_from_fwnode() does exactly. Side note: I didn't know fwnode has a ptr to the struct device. I wonder if we can kill off of_find_device_by_node() using that. That's for platform devices though. Rob
On Fri, Apr 12, 2024 at 07:54:32AM -0500, Rob Herring wrote: > On Thu, Apr 11, 2024 at 6:56 PM Saravana Kannan <saravanak@google.com> wrote: > I think it is better to not have this wrapper. We want it to be clear > when we're acquiring a ref. I know get_device() does that, but I have > to look up what get_dev_from_fwnode() does exactly. > > Side note: I didn't know fwnode has a ptr to the struct device. I > wonder if we can kill off of_find_device_by_node() using that. That's > for platform devices though. I don't like the idea because we already have a big design issue with fwnode that is used in struct device. Ideally, fwnode has to be a node in the linked list, head of which is provided by the user (struct device, for example). When it's done, it will be easy to handle. Have you read the comment in the struct fwnode_handle definition?
On Fri, Apr 12, 2024 at 07:54:32AM -0500, Rob Herring wrote: > On Thu, Apr 11, 2024 at 6:56 PM Saravana Kannan <saravanak@google.com> wrote: > > +#define get_dev_from_fwnode(fwnode) get_device((fwnode)->dev) > I think it is better to not have this wrapper. We want it to be clear > when we're acquiring a ref. I know get_device() does that, but I have > to look up what get_dev_from_fwnode() does exactly. Or perhaps calling it get_device_from_fwnode() would make it more obvious that it is a get_device() variant?
On Sun, Apr 14, 2024 at 6:06 PM Mark Brown <broonie@kernel.org> wrote: > > On Fri, Apr 12, 2024 at 07:54:32AM -0500, Rob Herring wrote: > > On Thu, Apr 11, 2024 at 6:56 PM Saravana Kannan <saravanak@google.com> wrote: > > > > +#define get_dev_from_fwnode(fwnode) get_device((fwnode)->dev) > > > I think it is better to not have this wrapper. We want it to be clear > > when we're acquiring a ref. I know get_device() does that, but I have > > to look up what get_dev_from_fwnode() does exactly. > > Or perhaps calling it get_device_from_fwnode() would make it more > obvious that it is a get_device() variant? Ack, I'll do this in my next version. Right now I'm waiting for Herve and Geert to confirm this series fixes their issues. -Saravana
diff --git a/drivers/base/core.c b/drivers/base/core.c index 5f4e03336e68..1a646f393dd7 100644 --- a/drivers/base/core.c +++ b/drivers/base/core.c @@ -46,6 +46,8 @@ static bool fw_devlink_drv_reg_done; static bool fw_devlink_best_effort; static struct workqueue_struct *device_link_wq; +#define get_dev_from_fwnode(fwnode) get_device((fwnode)->dev) + /** * __fwnode_link_add - Create a link between two fwnode_handles. * @con: Consumer end of the link. @@ -237,6 +239,70 @@ static void __fw_devlink_pickup_dangling_consumers(struct fwnode_handle *fwnode, __fw_devlink_pickup_dangling_consumers(child, new_sup); } +static void fw_devlink_pickup_dangling_consumers(struct device *dev) +{ + struct fwnode_handle *child; + + mutex_lock(&fwnode_link_lock); + fwnode_for_each_available_child_node(dev->fwnode, child) + __fw_devlink_pickup_dangling_consumers(child, dev->fwnode); + __fw_devlink_link_to_consumers(dev); + mutex_unlock(&fwnode_link_lock); +} + +/** + * fw_devlink_refresh_fwnode - Recheck the tree under this firmware node + * @fwnode: The fwnode under which the fwnode tree has changed + * + * This function is mainly meant to adjust the supplier/consumer dependencies + * after a fwnode tree overlay has occurred. + */ +void fw_devlink_refresh_fwnode(struct fwnode_handle *fwnode) +{ + struct device *dev; + + /* + * Find the closest ancestor fwnode that has been converted to a device + * that can bind to a driver (bus device). + */ + fwnode_handle_get(fwnode); + do { + if (fwnode->flags & FWNODE_FLAG_NOT_DEVICE) + continue; + + dev = get_dev_from_fwnode(fwnode); + if (!dev) + continue; + + if (dev->bus) + break; + + put_device(dev); + } while ((fwnode = fwnode_get_next_parent(fwnode))); + + /* + * If none of the ancestor fwnodes have (yet) been converted to a device + * that can bind to a driver, there's nothing to fix up. + */ + if (!fwnode) + return; + + WARN(device_is_bound(dev) && dev->links.status != DL_DEV_DRIVER_BOUND, + "Don't multithread overlaying and probing the same device!\n"); + + /* + * If the device has already bound to a driver, then we need to redo + * some of the work that was done after the device was bound to a + * driver. If the device hasn't bound to a driver, running thing too + * soon would incorrectly pick up consumers that it shouldn't. + */ + if (dev->links.status == DL_DEV_DRIVER_BOUND) + fw_devlink_pickup_dangling_consumers(dev); + + put_device(dev); + fwnode_handle_put(fwnode); +} + static DEFINE_MUTEX(device_links_lock); DEFINE_STATIC_SRCU(device_links_srcu); @@ -1322,14 +1388,8 @@ void device_links_driver_bound(struct device *dev) * child firmware node. */ if (dev->fwnode && dev->fwnode->dev == dev) { - struct fwnode_handle *child; fwnode_links_purge_suppliers(dev->fwnode); - mutex_lock(&fwnode_link_lock); - fwnode_for_each_available_child_node(dev->fwnode, child) - __fw_devlink_pickup_dangling_consumers(child, - dev->fwnode); - __fw_devlink_link_to_consumers(dev); - mutex_unlock(&fwnode_link_lock); + fw_devlink_pickup_dangling_consumers(dev); } device_remove_file(dev, &dev_attr_waiting_for_supplier); @@ -1888,8 +1948,6 @@ static void fw_devlink_unblock_consumers(struct device *dev) device_links_write_unlock(); } -#define get_dev_from_fwnode(fwnode) get_device((fwnode)->dev) - static bool fwnode_init_without_drv(struct fwnode_handle *fwnode) { struct device *dev; diff --git a/drivers/of/overlay.c b/drivers/of/overlay.c index 2ae7e9d24a64..7b2396c53127 100644 --- a/drivers/of/overlay.c +++ b/drivers/of/overlay.c @@ -179,6 +179,15 @@ static int overlay_notify(struct overlay_changeset *ovcs, return 0; } +static void overlay_fw_devlink_refresh(struct overlay_changeset *ovcs) +{ + for (int i = 0; i < ovcs->count; i++) { + struct device_node *np = ovcs->fragments[i].target; + + fw_devlink_refresh_fwnode(of_fwnode_handle(np)); + } +} + /* * The values of properties in the "/__symbols__" node are paths in * the ovcs->overlay_root. When duplicating the properties, the paths @@ -953,6 +962,12 @@ static int of_overlay_apply(struct overlay_changeset *ovcs, pr_err("overlay apply changeset entry notify error %d\n", ret); /* notify failure is not fatal, continue */ + /* + * Needs to happen after changset notify to give the listeners a chance + * to finish creating all the devices they need to create. + */ + overlay_fw_devlink_refresh(ovcs); + ret_tmp = overlay_notify(ovcs, OF_OVERLAY_POST_APPLY); if (ret_tmp) if (!ret) diff --git a/include/linux/fwnode.h b/include/linux/fwnode.h index 0d79070c5a70..95a78b87777a 100644 --- a/include/linux/fwnode.h +++ b/include/linux/fwnode.h @@ -220,6 +220,7 @@ int fwnode_link_add(struct fwnode_handle *con, struct fwnode_handle *sup, u8 flags); void fwnode_links_purge(struct fwnode_handle *fwnode); void fw_devlink_purge_absent_suppliers(struct fwnode_handle *fwnode); +void fw_devlink_refresh_fwnode(struct fwnode_handle *fwnode); bool fw_devlink_is_strict(void); #endif
When an overlay is applied, if the target device has already probed successfully and bound to a device, then some of the fw_devlink logic that ran when the device was probed needs to be rerun. This allows newly created dangling consumers of the overlayed device tree nodes to be moved to become consumers of the target device. Fixes: 1a50d9403fb9 ("treewide: Fix probing of devices in DT overlays") Reported-by: Herve Codina <herve.codina@bootlin.com> Closes: https://lore.kernel.org/lkml/CAMuHMdXEnSD4rRJ-o90x4OprUacN_rJgyo8x6=9F9rZ+-KzjOg@mail.gmail.com/ Closes: https://lore.kernel.org/all/20240221095137.616d2aaa@bootlin.com/ Closes: https://lore.kernel.org/lkml/20240312151835.29ef62a0@bootlin.com/ Signed-off-by: Saravana Kannan <saravanak@google.com> --- drivers/base/core.c | 76 +++++++++++++++++++++++++++++++++++++----- drivers/of/overlay.c | 15 +++++++++ include/linux/fwnode.h | 1 + 3 files changed, 83 insertions(+), 9 deletions(-)