Date: Sun, 1 Nov 2015 02:20:43 +0700 From: Eugene Grosbein <eugen@grosbein.net> To: Adrian Chadd <adrian.chadd@gmail.com> Cc: "freebsd-mips@freebsd.org" <freebsd-mips@freebsd.org> Subject: Re: CPU underload Message-ID: <5635148B.2070307@grosbein.net> In-Reply-To: <CAJ-Vmo=EpmG6OJxq_v_mqKMF48h_B3K7vKqnnaKAi3x1s-uaWQ@mail.gmail.com> References: <56348063.3090508@grosbein.net> <56348239.3050701@grosbein.net> <CAJ-Vmok%2BgKXcpzpxNCBhxX8sYSh%2BQO7OaeojB4MCufO%2BA4zceg@mail.gmail.com> <563500FC.8020201@grosbein.net> <CAJ-Vmo=EpmG6OJxq_v_mqKMF48h_B3K7vKqnnaKAi3x1s-uaWQ@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On 01.11.2015 01:11, Adrian Chadd wrote: > Hi, > > I don't see any DELAY in my testing, nor MipsUserIntr, so yeah, I must > be doing something different. I had compiled the system using CFLAGS+=-Os trying to minimize code size. It seems, that was not wise. Now I rebuilt it without -Os and got better numbers: 60MByte/s under same inter-vlan briding test. However, this time "interrupt storm" stroke again but that was not same bug but just really higher numbers of interrupts on arge0 due to higher speed. I raised hw.intr_storm_threshold from default 1000 upto 10000 and "storm" is no more. "Unaligned Load Word" messages in dmesg I've mentioned before are generated by running pmcstat. If I do not run it, dmesg is not flooded and its last line is: "Nov 1 01:43:06 tl-wdr3600 login: ROOT LOGIN (root) ON ttyu0". No port flapping. CPU utilization has raised too, but idle part is still pretty large and it is very unstable: # vmstat 1 procs memory page disks faults cpu r b w avm fre flt re pi po fr sr fl0 md0 in sy cs us sy id 0 0 0 58M 93M 297 0 3 0 217 2 0 0 1826 8680 184 5 63 32 0 0 0 58M 93M 0 0 0 0 0 2 0 0 3773 105 138 0 78 22 0 0 0 58M 93M 0 0 0 0 0 2 0 0 2316 105 116 0 62 38 0 0 0 58M 93M 0 0 0 0 0 5 0 0 5786 105 167 0 80 20 0 0 0 58M 93M 0 0 0 0 0 2 0 0 4177 105 95 0 79 21 0 0 0 58M 93M 0 0 0 0 0 2 0 0 3032 105 85 0 81 19 0 0 0 58M 93M 0 0 0 0 0 2 0 0 2831 105 152 0 66 34 0 0 0 58M 93M 0 0 0 0 0 1 0 0 4436 105 147 0 75 25 0 0 0 58M 93M 0 0 0 0 0 2 0 0 2943 105 84 0 75 25 0 0 0 58M 93M 0 0 0 0 0 2 0 0 2886 105 141 0 73 27 0 0 0 58M 93M 0 0 0 0 0 3 0 0 5727 105 50 0 95 5 0 0 0 58M 93M 0 0 0 0 0 2 0 0 3021 105 141 0 64 36 0 0 0 58M 93M 0 0 0 0 0 2 0 0 2824 105 127 0 76 24 0 0 0 58M 93M 0 0 0 0 0 2 0 0 2168 105 153 0 54 46 0 0 0 58M 93M 0 0 0 0 0 2 0 0 4879 105 116 0 83 17 0 0 0 58M 93M 0 0 0 0 0 2 0 0 4341 105 210 0 81 19 0 0 0 58M 93M 0 0 0 0 0 2 0 0 2294 105 127 0 54 46 0 0 0 58M 93M 0 0 0 0 0 2 0 0 4178 109 126 1 79 20 0 0 0 58M 93M 0 0 0 0 0 5 0 0 5048 105 106 0 72 28 pmcstat still shows some DELAY samples (but much less now) and MipsUserIntr is still here: PMC: [INSTR_EXECUTED] Samples: 14535 (100.0%) , 0 unresolved %SAMP IMAGE FUNCTION CALLERS 9.1 kernel MipsUserIntr 6.1 kernel if_inc_counter vlan_transmit:1.1 bridge_forward:1.1 ... 3.8 kernel ether_nh_input netisr_dispatch_src 3.7 kernel bridge_input ether_nh_input 3.5 kernel _bus_dmamap_sync arge_start_locked:1.7 arge_intr:1.0 ... 3.3 kernel _bus_dmamap_load_buf _bus_dmamap_load_mbuf_sg 3.1 kernel critical_enter 3.1 kernel arge_intr intr_event_execute_handlers 2.9 kernel bridge_rtnode_lookup bridge_rtlookup:1.5 ... 2.9 kernel memcpy 2.8 kernel arge_start_locked arge_start 2.6 libc.so.7 bsearch pmcstat_symbol_search 2.2 kernel critical_exit if_inc_counter 2.1 kernel bridge_forward bridge_input 1.9 kernel bridge_rtnode_addr_c 1.9 kernel bus_dmamap_sync_buf _bus_dmamap_sync 1.9 kernel DELAY 1.8 kernel arge_newbuf arge_intr 1.7 kernel netisr_dispatch_src netisr_dispatch 1.6 kernel mipsNN_pdcache_inv_r CONVERSION STATISTICS:g_read pmcstat_process_log #exec/elf 2 #samples/total 261959 #samples/unknown-function 8887
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5635148B.2070307>