usb: dwc2: gadget: kill requests with 'force' in s3c_hsotg_udc_stop()
This makes us sure that all requests are completed before we unbind gadget. There are assumptions in gadget API that all requests have to be completed and leak of complete can break some usb function drivers. For example unbind of ECM function can cause NULL pointer dereference: [ 26.396595] configfs-gadget gadget: unbind function 'cdc_ethernet'/e79c4c00 [ 26.414999] Unable to handle kernel NULL pointer dereference at virtual address 00000000 (...) [ 26.452223] PC is at ecm_unbind+0x6c/0x9c [ 26.456209] LR is at ecm_unbind+0x68/0x9c (...) [ 26.603696] [<c033fdb4>] (ecm_unbind) from [<c033661c>] (purge_configs_funcs+0x94/0xd8) [ 26.611674] [<c033661c>] (purge_configs_funcs) from [<c0336674>] (configfs_composite_unbind+0x14/0x34) [ 26.620961] [<c0336674>] (configfs_composite_unbind) from [<c0337124>] (usb_gadget_remove_driver+0x68/0x9c) [ 26.630683] [<c0337124>] (usb_gadget_remove_driver) from [<c03376c8>] (usb_gadget_unregister_driver+0x64/0x94) [ 26.640664] [<c03376c8>] (usb_gadget_unregister_driver) from [<c0336be8>] (unregister_gadget+0x20/0x3c) [ 26.650038] [<c0336be8>] (unregister_gadget) from [<c0336c84>] (gadget_dev_desc_UDC_store+0x80/0xb8) [ 26.659152] [<c0336c84>] (gadget_dev_desc_UDC_store) from [<c0335120>] (gadget_info_attr_store+0x1c/0x28) [ 26.668703] [<c0335120>] (gadget_info_attr_store) from [<c012135c>] (configfs_write_file+0xe8/0x148) [ 26.677818] [<c012135c>] (configfs_write_file) from [<c00c8dd4>] (vfs_write+0xb0/0x1a0) [ 26.685801] [<c00c8dd4>] (vfs_write) from [<c00c91b8>] (SyS_write+0x44/0x84) [ 26.692834] [<c00c91b8>] (SyS_write) from [<c000e560>] (ret_fast_syscall+0x0/0x30) [ 26.700381] Code: e30409f8 e34c0069 eb07b88d e59430a8 (e5930000) [ 26.706485] ---[ end trace f62a082b323838a2 ]--- It's because in some cases request is still running on endpoint during unbind and kill_all_requests() called from s3c_hsotg_udc_stop() function doesn't cause call of complete() of request. Missing complete() call causes ecm->notify_req equals NULL in ecm_unbind() function, and this is reason of this bug. Similar breaks can be observed in another usb function drivers. This patch fixes this bug forcing usb request completion in when s3c_hsotg_ep_disable() is called from s3c_hsotg_udc_stop(). Acked-by: Paul Zimmerman <paulz@synopsys.com> Signed-off-by: Robert Baldyga <r.baldyga@samsung.com> Signed-off-by: Felipe Balbi <balbi@ti.com>
Showing
Please register or sign in to comment