On 29.04.20 18:07, Dave Hansen wrote: > On 4/28/20 3:50 PM, Claudio Imbrenda wrote: >> If a page is inaccesible and it is used for things like sendfile, then >> the content of the page is not always touched, and can be passed >> directly to a driver, causing issues. >> >> This patch fixes the issue by adding a call to arch_make_page_accessible >> in page_cache_pipe_buf_confirm; this fixes the issue. > > I spent about 5 minutes putting together a patch: > > https://sr71.net/~dave/intel/accessible.patch You only set the page flag for compound pages. that of course leaves a big pile of pages marked a not accessible, thus explaining the sendto trace and all kind of other random traces. What do you see when you also do the SetPageAccessible(page); in the else page of prep_new_page (order == 0). (I do get > 10000 of these non compound page allocs just during boot). > > It adds a page flag ("daccess") which starts out set. It clears the > flag it when the page is added to the page cache or mapped as anonymous. > This are presumably the the two mostly likely kinds of pages to be > problematic. It re-sets the flag when it hits the new hook for s390: > arch_make_page_accessible(). > > It then patches the DMA mapping API. If a page gets to the DMA mapping > API without being accessible, it hits a tracepoint. > > It goes boom shortly after hitting userspace underneath a sys_sendto(). > That code uses lib/iov_iter.c which does get_user_pages_fast() and > apparently does not set FOLL_PIN, so never hits the s390 arch hooks. > > I hacked out the FOLL_PIN check and just universally call the hook for > all gup_pte_range() calls. I think you'll need to do that as well. I > don't think the assumptions about FOLL_PIN always preceding I/O is true > universally. Hacking out FOLL_PIN quiets down the warning spew quite a > bit, but it still hits a few of them. > > Here's one example: > > 0) sd-reso-410 | | /* mm_accessible_error: ... > sd-resolve-410 [000] .... 212.918838: <stack trace> > => trace_event_raw_event_mm_accessible_error > => check_page_accessible > => e1000_xmit_frame > => dev_hard_start_xmit > => sch_direct_xmit > => __qdisc_run > => __dev_queue_xmit > => ip_finish_output2 > => ip_output > => ip_send_skb > => udp_send_skb.isra.59 > => udp_sendmsg > => ____sys_sendmsg > => ___sys_sendmsg > => __sys_sendmmsg > => __x64_sys_sendmmsg > => do_syscall_64 > => entry_SYSCALL_64_after_hwframe > > This is just from booting and sitting on an idle Ubuntu 16.04.6 system. > I think the process in question here is the systemd resolver. >
