On Fri, Dec 26, 2025 at 08:55:40PM +0100, Michael Nazzareno Trimarchi wrote: > Hi > > On Fri, Dec 26, 2025 at 7:57 PM Tom Rini <[email protected]> wrote: > > > > On Fri, Dec 26, 2025 at 07:28:43PM +0100, Michael Nazzareno Trimarchi wrote: > > > Hi Tom > > > > > > On Fri, Dec 19, 2025 at 9:21 PM Tom Rini <[email protected]> wrote: > > > > > > > > The devm alloc functions that we have may follow the Linux kernel model > > > > where allocations are (almost always) automatically free()'d. However, > > > > quite often we don't enable, in full U-Boot, the corresponding automatic > > > > free. This in turn leads to memory leaks. Rather than manually tracking > > > > allocations and implementing frees, rework things so that we follow > > > > expectations now and enable the DEVRES functionality to manage frees. > > > > > > > > This turns DEVRES from a prompted symbol to a symbol that must be > > > > select'd, and we now remove our non-managed alloc/free functions from > > > > outside of xPL builds. > > [snip] > > > > @@ -279,8 +280,8 @@ config DEVRES > > > > released whether initialization fails half-way or the device > > > > gets > > > > detached. > > > > > > > > - If this option is disabled, devres functions fall back to > > > > - non-managed variants. For example, devres_alloc() to > > > > kzalloc(), > > > > + This option is disabled in xPL phases anddevres functions > > > > fall back > > > > > > Fix this typo > > > > OK. > > > > [snip] > > > Can you please in the commit message a concrete example on what this > > > commit takes care of? > > > In short expand here: > > > " in full U-Boot, the corresponding automatic free. This in turn leads > > > to memory leaks." > > > > Sure. So for example: > > commit 00e1fed93c8c3e4c9037741ea1b70a9e693a6e65 > > Author: Francois Berder <[email protected]> > > Date: Tue Nov 11 11:30:19 2025 +0100 > > > > firmware: ti_sci: Fix memory leaks in devm_ti_sci_get_of_resource > > > > - Fix temp memory leak > > - Free memory during error handling > > > > Signed-off-by: Francois Berder <[email protected]> > > > > Is needed (I'd have to re-read the threads to see if in whole or in > > part) because DEVRES is optional today. And barring > > drivers/core/devres.c doesn't work the way it's documented and expected, > > the normal unwind path will eventually say "Ah, and devm_kfree these > > devm_kmalloc'd areas". This was part of either the discussion for that > > patch, or another one of Francois' patches adding devm_kfree() calls to > > devm_kmalloc()'d areas. > > As far as I understand the devres_release_probe should then free the resources > allocated to dm_device.
Right. And without DEVRES enabled we don't have this match between alloc and free, which is unexpected by most users of devm_kmalloc/etc as that path is not optional in the kernel. -- Tom
signature.asc
Description: PGP signature
