On Thu, Nov 30, 2017 at 02:50:16PM +0100, Michal Hocko wrote:
On Thu 30-11-17 21:38:40, Wu Fengguang wrote:Hello,It looks like a regression in 4.15.0-rc1 -- the test case simply run a set of parallel dd's and there seems no reason to run into memory problem. It occurs in 1 out of 4 tests.This is an atomic allocations. So the failure really depends on the state of the free memory and that can vary between runs depending on timing I guess. So I am not really sure this is a regression. But maybe there is something reclaim related going on here.
Yes, it does depend on how the drivers rely on atomic allocations. I just wonder if any changes make the pressure more tight than before. It may not even be a MM change -- in theory drivers might also use atomic allocations more aggressively than before.
[...][ 71.088242] dd: page allocation failure: order:0, mode:0x1080020(GFP_ATOMIC), nodemask=(null) [ 71.098654] dd cpuset=/ mems_allowed=0-1 [ 71.104460] CPU: 0 PID: 6016 Comm: dd Tainted: G O 4.15.0-rc1 #1 [ 71.113553] Call Trace: [ 71.117886] <IRQ> [ 71.121749] dump_stack+0x5c/0x7b: dump_stack at lib/dump_stack.c:55 [ 71.126785] warn_alloc+0xbe/0x150: preempt_count at arch/x86/include/asm/preempt.h:23 (inlined by) should_suppress_show_mem at mm/page_alloc.c:3244 (inlined by) warn_alloc_show_mem at mm/page_alloc.c:3254 (inlined by) warn_alloc at mm/page_alloc.c:3293 [ 71.131939] __alloc_pages_slowpath+0xda7/0xdf0: __alloc_pages_slowpath at mm/page_alloc.c:4151 [ 71.138110] ? xhci_urb_enqueue+0x23d/0x580: xhci_urb_enqueue at drivers/usb/host/xhci.c:1389 [ 71.143941] __alloc_pages_nodemask+0x269/0x280: __alloc_pages_nodemask at mm/page_alloc.c:4245 [ 71.150167] page_frag_alloc+0x11c/0x150: __page_frag_cache_refill at mm/page_alloc.c:4335 (inlined by) page_frag_alloc at mm/page_alloc.c:4364 [ 71.155668] __netdev_alloc_skb+0xa0/0x110: __netdev_alloc_skb at net/core/skbuff.c:415 [ 71.161386] rx_submit+0x3b/0x2e0: rx_submit at drivers/net/usb/usbnet.c:488 [ 71.166232] rx_complete+0x196/0x2d0: rx_complete at drivers/net/usb/usbnet.c:659 [ 71.171354] __usb_hcd_giveback_urb+0x86/0x100: arch_local_irq_restore at arch/x86/include/asm/paravirt.h:777 (inlined by) __usb_hcd_giveback_urb at drivers/usb/core/hcd.c:1769 [ 71.177281] xhci_giveback_urb_in_irq+0x86/0x100 [ 71.184107] xhci_td_cleanup+0xe7/0x170: xhci_td_cleanup at drivers/usb/host/xhci-ring.c:1924 [ 71.189457] handle_tx_event+0x297/0x1190: process_bulk_intr_td at drivers/usb/host/xhci-ring.c:2267 (inlined by) handle_tx_event at drivers/usb/host/xhci-ring.c:2598 [ 71.194905] ? reweight_entity+0x145/0x180: enqueue_runnable_load_avg at kernel/sched/fair.c:2742 (inlined by) reweight_entity at kernel/sched/fair.c:2810 [ 71.200466] xhci_irq+0x300/0xb80: xhci_handle_event at drivers/usb/host/xhci-ring.c:2676 (inlined by) xhci_irq at drivers/usb/host/xhci-ring.c:2777 [ 71.205195] ? scheduler_tick+0xb2/0xe0: rq_last_tick_reset at kernel/sched/sched.h:1643 (inlined by) scheduler_tick at kernel/sched/core.c:3036 [ 71.210407] ? run_timer_softirq+0x73/0x460: __collect_expired_timers at kernel/time/timer.c:1375 (inlined by) collect_expired_timers at kernel/time/timer.c:1609 (inlined by) __run_timers at kernel/time/timer.c:1656 (inlined by) run_timer_softirq at kernel/time/timer.c:1688 [ 71.215905] __handle_irq_event_percpu+0x3a/0x1a0: __handle_irq_event_percpu at kernel/irq/handle.c:147 [ 71.221975] handle_irq_event_percpu+0x20/0x50: handle_irq_event_percpu at kernel/irq/handle.c:189 [ 71.227641] handle_irq_event+0x3d/0x60: handle_irq_event at kernel/irq/handle.c:206 [ 71.232682] handle_edge_irq+0x71/0x190: handle_edge_irq at kernel/irq/chip.c:796 [ 71.237715] handle_irq+0xa5/0x100: handle_irq at arch/x86/kernel/irq_64.c:78 [ 71.242326] do_IRQ+0x41/0xc0: do_IRQ at arch/x86/kernel/irq.c:241 [ 71.246472] common_interrupt+0x96/0x96: ret_from_intr at arch/x86/entry/entry_64.S:611 [ 71.251509] </IRQ>Ugh, this looks unreadable... Inlining information can be helpful sometime, alright but I find the below much more readable.
Heh, agreed.
[ 78.848629] dd: page allocation failure: order:0, mode:0x1080020(GFP_ATOMIC), nodemask=(null) [ 78.857841] dd cpuset=/ mems_allowed=0-1 [ 78.862502] CPU: 0 PID: 6131 Comm: dd Tainted: G O 4.15.0-rc1 #1 [ 78.870437] Call Trace: [ 78.873610] <IRQ> [ 78.876342] dump_stack+0x5c/0x7b [ 78.880414] warn_alloc+0xbe/0x150 [ 78.884550] __alloc_pages_slowpath+0xda7/0xdf0 [ 78.889822] ? xhci_urb_enqueue+0x23d/0x580 [ 78.894713] __alloc_pages_nodemask+0x269/0x280 [ 78.899891] page_frag_alloc+0x11c/0x150 [ 78.904471] __netdev_alloc_skb+0xa0/0x110 [ 78.909277] rx_submit+0x3b/0x2e0 [ 78.913256] rx_complete+0x196/0x2d0 [ 78.917560] __usb_hcd_giveback_urb+0x86/0x100 [ 78.922681] xhci_giveback_urb_in_irq+0x86/0x100 [ 78.928769] ? ip_rcv+0x261/0x390 [ 78.932739] xhci_td_cleanup+0xe7/0x170 [ 78.937308] handle_tx_event+0x297/0x1190 [ 78.941990] xhci_irq+0x300/0xb80 [ 78.945968] ? pciehp_isr+0x46/0x320 [ 78.950870] __handle_irq_event_percpu+0x3a/0x1a0 [ 78.956311] handle_irq_event_percpu+0x20/0x50 [ 78.961466] handle_irq_event+0x3d/0x60 [ 78.965962] handle_edge_irq+0x71/0x190 [ 78.970480] handle_irq+0xa5/0x100 [ 78.974565] do_IRQ+0x41/0xc0 [ 78.978206] ? pagevec_move_tail_fn+0x350/0x350 [ 78.983412] common_interrupt+0x96/0x96Unfortunatelly we are missing the most imporatant information, the meminfo. We cannot tell much without it. Maybe collecting /proc/vmstat during the test will tell us more.
Attached the JSON format per-second vmstat records. It feels more readable than the raw dumps. Thanks, Fengguang
proc-vmstat.json.gz
Description: application/gzip
