Tue, 21 Aug 2018 1:54, ronnie sahlberg <ronniesahlberg@xxxxxxxxx>: > > I think this is probably not due to these patches themselves but as > they may make compounding more common they probably uncover bugs > introduced in the previous patch sets of mine > (that did touch a lot of very hairy code I struggled to unwind.) I suspect that in some conditions the following part of compound_send_recv is being executed once we process a part of the compounding chain: 910 out: 911 >---for (i = 0; i < num_rqst; i++) 912 >--->---cifs_delete_mid(midQ[i]); 913 >---add_credits(ses->server, credits, optype); 914 915 >---return rc; 916 } cifs_delete_mid() calls list_del() for a mid. So a subsequent dequeue_mid() may generate the general protection fault reported by Aurelien. Actually cifs_delete_mid() and dequeue_mid() doing the same thing (removing the mid from the MID list) looks scary and this needs to be revised. The easiest fix would be to change list_del() to list_del_init() in cifs_delete_mid() but this may just hide the problem. -- Best regards, Pavel Shilovsky
