| Summary: | DAHLIN-00269: dahdi timing issues with recent kernels and intel_idle driver | ||||||
| Reporter: | frawd (frawd) | Labels: | |||||
| Date Opened: | 2011-12-09 05:50:06.000-0600 | Date Closed: | 2019-05-31 09:45:14 | ||||
| Priority: | Minor | Regression? | |||||
| Status: | Closed/Complete | Components: | wcte11xp | ||||
| Versions: | 2.5.0.2 | Frequency of Occurrence | Constant | ||||
| Related Issues: |
| ||||||
| Environment: | Attachments: | ||||||
| Description: | When the intel_idle driver is active (default in recent linux kernels like 3.0), i see lots of timing issues rendering PCI and PCI-E PRI cards useless (HDLC Abort, etc.). dahdi_test drops to 90% or less. Workaround is disabling intel_idle by booting with intel_idle.max_cstate=0 (which makes use of acpi_idle instead). This affects all digium cards i could test so far on HP servers with modern Intel CPU (core iX CPUs and recent Xeon) on recent kernels. Does not affect older Intel CPUs (core duo). For info: Sangoma hardware works perfectly in the same server. | ||||||
| Comments: | By: Shaun Ruffell (sruffell) 2011-12-09 10:47:13.605-0600 Which card are you using? It's one of the older single span cards (really using wcte11xp driver?) By: frawd (frawd) 2011-12-09 11:11:22.135-0600 Yeah I guess those are old single span cards. I only do remote maintenance and the customer tried with a PCI and a PCI-E version of the PRI card in three different HP servers. The server with a "core duo" didn't suffer any issue, that's when i thought of the intel_idle thing (I had just upgraded kernel from 2.6.30 to 3.0). The PCI-E card uses the wcte11xp driver and reports in lspci as: Network controller: Tiger Jet Network Inc. Tiger3XX Modem/ISDN interface I had a few other dahdi_test issues in other Intel machines with old Digium and Openvox analog cards, I'll report here if those are related problems. Trying with equivalent Sangoma cards solved the issue, I don't know why (maybe those cards are more tolerant with small interrupt delays?) By: Shaun Ruffell (sruffell) 2011-12-09 12:21:29.138-0600 If I provided a patch to the wcte11xp driver could you test it easily? While that device isn't officially supported, it might be made to work on your system by using the pm_qos interfaces. By: Shaun Ruffell (sruffell) 2011-12-09 14:02:34.131-0600 Also note that Digium's TE122 single span card and 5th generation dual and quad span cards are tolerant of these types of latency delays. It's just the old TE110 cards (and X100P and TDM400) cards that are sensitive to 1ms interrupt latencies. By: frawd (frawd) 2011-12-11 08:18:17.557-0600 Thanks for the effort, I'll make sure to test a patch if you provide it. Also nice to know newer cards are more robust to these problems. I just saw the EOL notices for these old cards, it appears the I may be able to send those back for RMA and have them replaced with newer version. I'll see with Digium tech support. Ref: http://www.digium.com/en/products/eol.php By: Shaun Ruffell (sruffell) 2012-04-06 18:04:24.666-0500 There is a slight chance that this is related to the problem reported in ASTERISK-18742. | ||||||