6.1.y/6.6.y: fsl-usb fails to build

Hi,

While working on some backport patches for an out-of-tree module and looking for the exact versions containing the disable_delayed_work_sync symbol, I stumbled upon occurrences of that function on longterm v6.1 and v6.6 in drivers/usb/phy/phy-fsl-usb.c despite not being declared or defined anywhere. In fact, disable_delayed_work_sync was introduced by upstream commit 86898fa6b8cd ("workqueue: Implement disable/enable for (delayed) work items") and is only available since v6.10. Neither 6.1.y nor 6.6.y defines it.

Consequently, kernels built with CONFIG_FSL_USB2_OTG fail with:

drivers/usb/phy/phy-fsl-usb.c:990:2: error: call to undeclared function 'disable_delayed_work_sync'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]

  990 |         disable_delayed_work_sync(&fsl_otg_dev->otg_event);
      |         ^

drivers/usb/phy/phy-fsl-usb.c:990:2: note: did you mean 'cancel_delayed_work_sync'? include/linux/workqueue.h:470:13: note: 'cancel_delayed_work_sync' declared here
470 | extern bool cancel_delayed_work_sync(struct delayed_work *dwork);
| ^

I reproduced the failure on PowerPC using corenet64_smp_defconfig with CONFIG_FSL_USB2_OTG=m, building drivers/usb/phy/phy-fsl-usb.o with Clang.

AFAICT, in 6.1.y, the regression was introduced in v6.1.160 by commit 4476c73bbbb0 ("usb: phy: fsl-usb: Fix use-after-free in delayed work during device removal") and remains present in v6.1.185.

In 6.6.y, it was introduced in v6.6.120 by commit 319f7a85b3c4 ("usb:
phy: fsl-usb: Fix use-after-free in delayed work during device removal") and remains present in v6.6.154.

Note that the same problem was pointed out by Ben Hutchings during the 5.10.y review:
https://lore.kernel.org/all/[email protected]/

The corresponding patches were dropped from 5.10.y and 5.15.y, but appear to have been retained in 6.1.y and 6.6.y.

Simply replacing disable_delayed_work_sync with cancel_delayed_work_sync at its current location may not be sufficient: it looks like the delayed work can also be scheduled from fsl_otg_set_host and from the IRQ handler but I'm not familiar with this part of the kernel.

Quoted 37 lines · Ralf Lici

Hi,

While working on some backport patches for an out-of-tree module and looking for the exact versions containing the disable_delayed_work_sync symbol, I stumbled upon occurrences of that function on longterm v6.1 and v6.6 in drivers/usb/phy/phy-fsl-usb.c despite not being declared or defined anywhere. In fact, disable_delayed_work_sync was introduced by upstream commit 86898fa6b8cd ("workqueue: Implement disable/enable for (delayed) work items") and is only available since v6.10. Neither 6.1.y nor 6.6.y defines it.

Consequently, kernels built with CONFIG_FSL_USB2_OTG fail with:

drivers/usb/phy/phy-fsl-usb.c:990:2: error: call to undeclared function 'disable_delayed_work_sync'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]

  990 |         disable_delayed_work_sync(&fsl_otg_dev->otg_event);
      |         ^

drivers/usb/phy/phy-fsl-usb.c:990:2: note: did you mean 'cancel_delayed_work_sync'? include/linux/workqueue.h:470:13: note: 'cancel_delayed_work_sync' declared here
470 | extern bool cancel_delayed_work_sync(struct delayed_work *dwork);
| ^

I reproduced the failure on PowerPC using corenet64_smp_defconfig with CONFIG_FSL_USB2_OTG=m, building drivers/usb/phy/phy-fsl-usb.o with Clang.

AFAICT, in 6.1.y, the regression was introduced in v6.1.160 by commit 4476c73bbbb0 ("usb: phy: fsl-usb: Fix use-after-free in delayed work during device removal") and remains present in v6.1.185.

In 6.6.y, it was introduced in v6.6.120 by commit 319f7a85b3c4 ("usb:
phy: fsl-usb: Fix use-after-free in delayed work during device removal") and remains present in v6.6.154.

Note that the same problem was pointed out by Ben Hutchings during the 5.10.y review:
https://lore.kernel.org/all/[email protected]/

Quoted 7 lines

The corresponding patches were dropped from 5.10.y and 5.15.y, but appear to have been retained in 6.1.y and 6.6.y.

Simply replacing disable_delayed_work_sync with cancel_delayed_work_sync at its current location may not be sufficient: it looks like the delayed work can also be scheduled from fsl_otg_set_host and from the IRQ handler but I'm not familiar with this part of the kernel.

Can you just send patches to revert the offending commits? Or we can just ignore it as it's been this way for a while and no one has noticed which implies to me that no one builds these modules for these arches on these very old kernel trees :)