#
#asahi-dev · 2026-08-20 08:44
svenstill amazing we just missed an entire controller for that fix and no one noticed so far
#asahi-dev · 2026-08-20 08:42
jannauI've tagged asahi-7.1.9-1 (bt compatibility fix for bcm4378 and speaker amp shutdown pin handling changes)
#asahi-dev · 2026-08-20 08:38
chaos_princessor, just nfs boot over usb4net :P
#asahi-dev · 2026-08-20 08:38
svensomeone should just implement usb4net in m1n1 so that we can load a kernel with a full rootfs in <1s!
#asahi-dev · 2026-08-20 08:37
chaos_princessit can't, since the other side is pretty often macos
#asahi-dev · 2026-08-20 08:37
jannaustarting with 7.2 usb4stream can serve the same purpose although setup is more complicated
#asahi-dev · 2026-08-20 08:34
chaos_princessi sometimes use it for the purposes a normal person would use a usb stick
#asahi-dev · 2026-08-20 08:33
jannaudoes anyone has an use case for usb4net (except for tests/benchmarks)? it would be convenient if it could be used in parallel with usb serial for the hypervisor
#asahi-dev · 2026-08-20 06:00
sventhe pages from that page_pool can just stay mapped
#asahi-dev · 2026-08-20 06:00
svensomething like drivers/net/ethernet/freescale/fec_main.c, commit 95698ff6177b should work for usb4net as well I think
#asahi-dev · 2026-08-20 05:54
svengonna prototype something this weekend
#asahi-dev · 2026-08-20 05:53
sveni've been reading some network code and i think i can make usb4net re-use dma mappings in the rx path as well and still retain zero copy
#asahi-dev · 2026-08-19 21:16
jannauboth with macos and linux. intel <> t6020/linux iperf throughput is 17.4 / 20.5 Gbits/sec
#asahi-dev · 2026-08-19 21:12
jannauworks with a different cable
#asahi-dev · 2026-08-19 21:02
jannauin turn of events macOS logs "[ACIO1:high_speed_lane.c:289] Gen2/3 link error. lane=1, error=89"
#asahi-dev · 2026-08-19 20:56
svenand then we get to play another game of spot the difference that’s gonna end up in something dumb like the second dart stream 🙃
#asahi-dev · 2026-08-19 20:55
svenand then Intel/Linux where that thing above happens from the Linux side as well
#asahi-dev · 2026-08-19 20:55
svenyes, from the macos side
#asahi-dev · 2026-08-19 20:51
jannauwould an acio trace of intel/macos be helpful?
#asahi-dev · 2026-08-19 20:50
jannauand https://paste.centos.org/view/88067842
#asahi-dev · 2026-08-19 20:49
jannaustill only "RTKit: syslog message: high_speed_lane.c:289: Gen2/3 link error. lane=1, error=89" spam when connecting the intel notebook to t6020/linux
#asahi-dev · 2026-08-19 20:47
jannau"30 / 24 Gbit/s" from the backlog
#asahi-dev · 2026-08-19 20:44
jannaucurrently seeing 18.0 / 21.2 gbps between 13th gen core i7 and t8103 but I think I saw more a couple of days ago
#asahi-dev · 2026-08-19 20:41
chaos_princessactually, ignore me, i can't read, it is on cpu
#asahi-dev · 2026-08-19 20:40
chaos_princessso, 4x4
#asahi-dev · 2026-08-19 20:40
chaos_princessapparently hangs off the chipset
#asahi-dev · 2026-08-19 20:40
chaos_princesshttps://download.gigabyte.com/FileList/Manual/mb_manual_z890-aorus-master_1004_e.pdf?v=79e8398ec40729ae864e1016bb99aca7
#asahi-dev · 2026-08-19 20:34
jannauhasn't intel integrated tbt controller in recent CPUs?
#asahi-dev · 2026-08-19 20:33
chaos_princessisn't it "as fast as pcie 3x4 + overhead"?
#asahi-dev · 2026-08-19 20:30
svenhow fast is this on non-apple machines again? :>
#asahi-dev · 2026-08-19 20:29
sven28gbps even
#asahi-dev · 2026-08-19 20:29
svenwow, 27.5gbps during peaks
#asahi-dev · 2026-08-19 20:26
svenotherwise we should've seen traffic in one direction before we found that secondary stream
#asahi-dev · 2026-08-19 20:26
jannauand the linux transmit case could be faster than 23.5gbps. I'm blaming retransmits new: https://paste.centos.org/view/5d2bd097
#asahi-dev · 2026-08-19 20:26
svenchaos_princess: i think both rx and tx are on stream 2 and just the control ring is on stream 1
#asahi-dev · 2026-08-19 20:21
svenhah, nice!
#asahi-dev · 2026-08-19 20:21
jannau16.0 Gbits/sec / 23.5 Gbits/sec with 13 additional chars: "dma-coherent;"
#asahi-dev · 2026-08-19 20:00
svenah :/
#asahi-dev · 2026-08-19 20:00
chaos_princessso, is tx on stream 1 and rx on stream 2?
#asahi-dev · 2026-08-19 19:59
jannauI think it is the way it is to support zero-copy network packet handling
#asahi-dev · 2026-08-19 19:58
sven*remap
#asahi-dev · 2026-08-19 19:58
sveni guess stream map internally then doesn't adjust the stream id but just mirrors the configuration (and invalidations) :(
#asahi-dev · 2026-08-19 19:57
sveni wonder if we can make the RX path re-use buffers that stay mapped the entire time as well
#asahi-dev · 2026-08-19 19:55
jannauusing DART's stream remap instead of 2 iommus doesn't increase the iperf3 throughput on t6020
#asahi-dev · 2026-08-19 18:47
svennice, thanks for confirming!
#asahi-dev · 2026-08-19 18:47
chaos_princesssven: both work
#asahi-dev · 2026-08-19 18:45
jannauyes, that does it. I thought strict was the default but fedora uses lazy
#asahi-dev · 2026-08-19 18:28
sveni think we need to add a capable function that returns true for IOMMU_CAP_DEFERRED_FLUSH
#asahi-dev · 2026-08-19 18:24
sventhat sysfs write just bypasses that check :D
#asahi-dev · 2026-08-19 18:24
svenlast i checked that won't work because we claim we can't do dma-fq

plus ›