drivers/slipdev: make use of chunked ringbuffer

[benpicco]

Contribution description

Just a test to see if this can increase performance with high data ingress.

But turns out this also solves two nasty SLIP bugs:

• trivial solution for frames stuck in RX queue
• fixes pktbuf corruption (drivers/slipdev: fix off-by-one error in _recv() #18229) by removing the offending code

Testing procedure

SLIP should work as before, ideally a bit better.

Issues/PRs references

alternative to #18826, #18229

[miri64]

Just a test to see if this can increase performance with high data ingress.

Any results yet?

[benpicco]

Any results yet?

So far no improvements, but also no worse than master. (I tested examples/benchmark_udp and for i in {10..20}; do echo $i | nc -u -6 -w0 2001:dbc::e4ea:aff8:af5d:f8e5 1234; done and counted the RX frames)
But I didn't yet tests conditions with higher CPU load yet.

[benpicco]

Uh so this turns out to fix a gnarly pktbuf corruption issue with SLIP

[kfessel]

From my POV (non-user of this driver):

• the change improves it:
  reading in chunks enables the networkstack to check the size of the next chunck prior to requesting it making loss of packages less likely
• and the readability (there is only one place the unescaping is done)

[kfessel]

@benpicco: you may squash and make this ready for review

[benpicco]

All squashed now.

Seems like my worries were misplaced.

[riot-ci]

Murdock results

✔️ PASSED

1a888fe drivers/slipdev: make use of chunked ringbuffer

Success	Failures	Total	Runtime
8100	0	8100	10m:06s

Artifacts

• Documentation preview

[fabian18]

Even if it may not be a performance increase,I can understand the new code better than the old code.

[fabian18]

BOARD=same54-xpro UPLINK=slip make flash term (gnrc_border_router)

> ping -c 10 -s 1000 fe80::1%6 
ping -c 10 -s 1000 fe80::1%6
1008 bytes from fe80::1%6: icmp_seq=0 ttl=64 time=185.192 ms
ps
1008 bytes from fe80::1%6: icmp_seq=1 ttl=64 time=185.191 ms
1008 bytes from fe80::1%6: icmp_seq=2 ttl=64 time=185.160 ms
1008 bytes from fe80::1%6: icmp_seq=3 ttl=64 time=185.192 ms
1008 bytes from fe80::1%6: icmp_seq=4 ttl=64 time=185.189 ms
1008 bytes from fe80::1%6: icmp_seq=5 ttl=64 time=185.169 ms
1008 bytes from fe80::1%6: icmp_seq=6 ttl=64 time=185.157 ms
1008 bytes from fe80::1%6: icmp_seq=7 ttl=64 time=185.172 ms
1008 bytes from fe80::1%6: icmp_seq=8 ttl=64 time=185.264 ms
1008 bytes from fe80::1%6: icmp_seq=9 ttl=64 time=185.176 ms

--- fe80::1%6 PING statistics ---
10 packets transmitted, 10 packets received, 0% packet loss
round-trip min/avg/max = 185.157/185.186/185.264 ms
> ps
	pid | name                 | state    Q | pri | stack  ( used) ( free) | base addr  | current     
	  - | isr_stack            | -        - |   - |    512 (  256) (  256) | 0x20000000 | 0x200001c8
	  1 | main                 | running  Q |   7 |   1536 (  936) (  600) | 0x20000320 | 0x20000624 
	  2 | 6lo                  | bl rx    _ |   3 |   1024 (  256) (  768) | 0x20004664 | 0x20004964 
	  3 | ipv6                 | bl rx    _ |   4 |   1024 (  496) (  528) | 0x200009fc | 0x20000c6c 
	  4 | udp                  | bl rx    _ |   5 |    512 (  256) (  256) | 0x20004a64 | 0x20004b64 
	  5 | sam0_eth             | bl anyfl _ |   2 |    896 (  296) (  600) | 0x20001988 | 0x20001c4c 
	  6 | slipdev              | bl anyfl _ |   2 |    896 (  280) (  616) | 0x20001e90 | 0x20002154 
	    | SUM                  |            |     |   6400 ( 2776) ( 3624)

• works fine IMO
• code got cleaner
• PR has been open for a long time now and we also have been using it internally for that time

so let's have it merged