Status:
Reset:
Errors: _________________ ASUS GT-BE98 PRO Main: Fiber 5gbps up/down
ASUS AXE16000: AI Mesh node
2 X ASUS RT-AX89X: AI Mesh nodes
QNAP QSW-1208-8C 12-Port 10GbE Switch
XS712T ProSafe 12-Port 10GbE Switch
3 X R9000 DD-WRT Mesh
@Labo- No activity at all on the WAN? Is nothing plugged in? Are you using the 60Ghz- it shows active? Looks like BS definitely was playing with the IRQ's and messed up with core allocation; If you have WAN plugged in...
Router Model TP-Link TL-WR841ND v11
Firmware Version DD-WRT v3.0-r39956 std (06/06/19)
Kernel Version Linux 3.18.140 #42926 Thu Jun 6 03:27:04 CEST 2019 mips
Uptime 3:14
Update via telnet from r39944
Reset: No
Working good as AP, NG-mixed, HT40, QOS.
Problems: none.
Joined: 30 Jan 2015 Posts: 677 Location: Texas, USA
Posted: Fri Jun 07, 2019 0:29 Post subject:
msoengineer wrote:
@Labo- No activity at all on the WAN? Is nothing plugged in? Are you using the 60Ghz- it shows active? Looks like BS definitely was playing with the IRQ's and messed up with core allocation; If you have WAN plugged in...
This is just mesh WDS bridge, no WAN, mesh connected using 5GHz radio which is not connecting after update.
Update:
The issue may be beyond the interrupts.
But why do we need to reassign so many interrupts ?
All we need is WAN and 5GHz radio to be on separate cores from the original assignments.
Alpine chip so complex it has two built-in 10G MAC and cryptodev engine. Not sure if we are using all the hardware features in our dd-wrt firmware.
Also, the original irq257 and 258 assignments may be intended.
Original firmware supports Link Aggregation on two lan ports not sure if these were meant for it. _________________ ASUS GT-BE98 PRO Main: Fiber 5gbps up/down
ASUS AXE16000: AI Mesh node
2 X ASUS RT-AX89X: AI Mesh nodes
QNAP QSW-1208-8C 12-Port 10GbE Switch
XS712T ProSafe 12-Port 10GbE Switch
3 X R9000 DD-WRT Mesh
Joined: 21 Jan 2017 Posts: 1797 Location: Illinois Moderator
Posted: Fri Jun 07, 2019 13:56 Post subject:
labo wrote:
Update:
The issue may be beyond the interrupts.
But why do we need to reassign so many interrupts ?
All we need is WAN and 5GHz radio to be on separate cores from the original assignments. Alpine chip so complex it has two built-in 10G MAC and cryptodev engine. Not sure if we are using all the hardware features in our dd-wrt firmware.
Also, the original irq 257 and 258 assignments may be intended. Original firmware supports Link Aggregation on two lan ports not sure if these were meant for it.
The wireless mesh is not related to IRQ's being re-assigned- those are different Gremlins...
When you run "lspci" you will see there are 3 switches. The goofy xxxx:xx:xx.x format is the PCI address of the particular device. So when you run cat /proc/interrupts you have a ton of IRQ's for each switch. For whatever reason, there are 4-RX and 4-TX PCI assignments per each switch. This leads me to believe that Netgear split up the RX and TX across the 4 cores (there's 6 LAN ports, so they're not splitting each port per core...) When you look at BS Build 39944 cat /proc/interrupts you will see that the LAN, on pci:0000:00:01.0,has all IRQ's split nicely and evenly across the cores:
Posted: Fri Jun 07, 2019 15:05 Post subject: Upgrades - Edited
PIE has arrived on several routers now. I am still testing it. Yes, even on the ancient 841ND v8, not the DIR-615.
Router/Version: TP-Link Archer C7 v2 x2
Firmware: 39956
Previous: 39855
Mode/Status: WDS AP, WDS Station
Reset: No
Issues/Errors: 2.4GHz wifi was unstable. It would drop the connection after ~24 hours.
Router/Version: D-Link DIR-862L A1 x2
Firmware: 39956
Previous: 39855
Mode/Status: WDS Station
Reset: No
Issues/Errors: None, so far. No WPA3.
Router/Version: D-Link DIR-825 B1
Firmware: 39956
Previous: 39855
Mode/Status: WDS Station
Reset: No
Issues/Errors: None, so far.
Router/Version: TP-Link WR841ND v8
Firmware: 39956
Previous: 39855, 39944
Mode/Status: WDS Station, WDS AP
Reset: No
Issues/Errors: None, so far. No WPA3.
Router/Version: TP-Link WDR3600 v1
Firmware: 39956
Previous: 39855
Mode/Status: WDS Station
Reset: No
Issues/Errors: None, so far. No WPA3.
Router/Version: D-Link DIR-615 C1
Firmware: 39956
Previous: 39944
Mode/Status: WDS Station
Reset: No
Issues/Errors: None, so far. No WPA3.
Edited, had to downgrade all routers to 39944. My Main Archer C7 v2 went back to 39855. 2.4GHz wifi was unstable. It would drop the connection after ~24 hours. _________________ Before asking a question on the forums, update dd-wrt: Where do I download firmware? I suggest reading it all.
QCA Best WiFi Settings
Some dd-wrt wiki pages are up to date, others are not. PM me if you find an old one.
Atheros:
Netgear R7800 x5 - WDS AP / station, gateway, QoS
TP-Link Archer C7 v2 x2 - WDS Station
D-Link 615 C1/E3/I1 x 7 - 1 WDS station
D-Link 825 B1 - Not used
D-Link 859 - WDS Station
D-Link 862L A1 x2 - WDS Station
UBNT 5AC loco x2 - airOS
UBNT loco M2 x2 - airOS
Broadcom
Linksys EA6400 - Gateway, QoS
Asus N66U - AP
Netgear R6400 - AP
MediaTek
UBNT EdgeRouter X - switch
Netgear R6350 - AP
Last edited by ian5142 on Mon Jun 10, 2019 11:59; edited 1 time in total
Joined: 30 Jan 2015 Posts: 677 Location: Texas, USA
Posted: Fri Jun 07, 2019 15:40 Post subject:
msoengineer wrote:
labo wrote:
Update:
The issue may be beyond the interrupts.
But why do we need to reassign so many interrupts ?
All we need is WAN and 5GHz radio to be on separate cores from the original assignments. Alpine chip so complex it has two built-in 10G MAC and cryptodev engine. Not sure if we are using all the hardware features in our dd-wrt firmware.
Also, the original irq 257 and 258 assignments may be intended. Original firmware supports Link Aggregation on two lan ports not sure if these were meant for it.
The wireless mesh is not related to IRQ's being re-assigned- those are different Gremlins...
When you run "lspci" you will see there are 3 switches. The goofy xxxx:xx:xx.x format is the PCI address of the particular device. So when you run cat /proc/interrupts you have a ton of IRQ's for each switch. For whatever reason, there are 4-RX and 4-TX PCI assignments per each switch. This leads me to believe that Netgear split up the RX and TX across the 4 cores (there's 6 LAN ports, so they're not splitting each port per core...) When you look at BS Build 39944 cat /proc/interrupts you will see that the LAN, on pci:0000:00:01.0,has all IRQ's split nicely and evenly across the cores:
Well, on the switch IRQ assignments, only 6 out of 8 ports are used.
So in one of the switch only two ports are used then the IRQ assignment are meaningless on the unused 2 ports.
Also one switch with two used only might be meant for Link Aggregation of two ports which is not supported by dd-wrt.
Though it logically makes sense on the distribution of switch IRQs, All I'm saying is, let's not commit changes in hurry.
I'm already concerned on the level of attention given by the Devs on this router.
I have two R9000's mash connected each having 10G SFP+ connected server/raid5 all working OK.
The real major issue that I'm seeing with this router is the 5Ghz radio performance which needs our immediate attention.
Any efforts to fix these issues will be appreciated.
BTW, I'm not discouraging the efforts and appreciate the insights being enlightened.
We just need to more prudent. _________________ ASUS GT-BE98 PRO Main: Fiber 5gbps up/down
ASUS AXE16000: AI Mesh node
2 X ASUS RT-AX89X: AI Mesh nodes
QNAP QSW-1208-8C 12-Port 10GbE Switch
XS712T ProSafe 12-Port 10GbE Switch
3 X R9000 DD-WRT Mesh
Joined: 21 Jan 2017 Posts: 1797 Location: Illinois Moderator
Posted: Fri Jun 07, 2019 16:20 Post subject:
labo wrote:
Well, on the switch IRQ assignments, only 6 out of 8 ports are used. So in one of the switch only two ports are used then the IRQ assignment are meaningless on the unused 2 ports.
Also one switch with two used only might be meant for Link Aggregation of two ports which is not supported by dd-wrt.
I don't use the SFP myself, Since you do, I would need to see your cat /proc/interrupts dump of build R39944 and R39956 to see where what cores the SFP is mapped to. The PCI for the SFP is 0000:00:00.0. (Mine doesn't show up since it's not active)
labo wrote:
Though it logically makes sense on the distribution of switch IRQs, All I'm saying is, let's not commit changes in hurry. I'm already concerned on the level of attention given by the Devs on this router.
In a good or bad way?
labo wrote:
I have two R9000's mash connected each having 10G SFP+ connected server/raid5 all working OK.
The real major issue that I'm seeing with this router is the 5Ghz radio performance which needs our immediate attention.
What type of issues are you seeing? Once I moved to core 2 things run smooth.
labo wrote:
Any efforts to fix these issues will be appreciated. BTW, I'm not discouraging the efforts and appreciate the insights being enlightened.
We just need to more prudent.
I would give BS the benefit of the doubt about moving the cores around, I agree this is not a simplistic change; but having the same PCI run across 4 cores introduces latency; I've been able to clearly see that playing around when doing speed tests. I haven't delved into running iperf yet to be able to offer more evidence/data to prove one way or another. However, since you need mesh, and 39956 is has mesh gremlins, I can't fault you. _________________ FORUM RULES
