On Wed, May 28, 2025 at 5:24 AM Jerome Forissier <jerome.foriss...@linaro.org> wrote: > > > > On 5/27/25 18:35, Tim Harvey wrote: > > On Mon, May 26, 2025 at 5:17 AM Jerome Forissier > > <jerome.foriss...@linaro.org> wrote: > >> > >> Hi Tim, > >> > >> On 5/23/25 20:31, Tim Harvey wrote: > >>> Hi Jerome, > >>> > >>> I've enabled LWIP on imx8mm_venice to see how it's doing and while > >>> dhcp/dns/ping/wget appear to work great, tftp is broken. > >>> > >>> Enabling LWIP_DEBUG shows the following: > >>> ip4_input: packet accepted on interface et > >>> IP packet dropped since it was fragmented (0x2000) (while IP_REASSEMBLY > >>> == 0). > >> > >> Hmmmmm IP_REASSEMBLY should probably default to 1. > >>> > >>> I tried enabling IP_REASSEMBLY in lwip but that seems to change the issue > >>> to: > >>> ip4_input: packet accepted on interface et > >>> ip4_input: > >>> IP header: > >>> +-------------------------------+ > >>> | 4 | 5 | 0xc0 | 108 | (v, hl, tos, len) > >>> +-------------------------------+ > >>> | 10568 |000| 0 | (id, flags, offset) > >>> +-------------------------------+ > >>> | 64 | 1 | 0xe25b | (ttl, proto, chksum) > >>> +-------------------------------+ > >>> | 172 | 24 | 0 | 2 | (src) > >>> +-------------------------------+ > >>> | 172 | 24 | 21 | 251 | (dest) > >>> +-------------------------------+ > >>> ip4_input: p->len 108 p->tot_len 108 > >>> Unsupported transport protocol 1 > >>> > >>> Any ideas what the issue is here? > >> > >> Not really :-/ Does the transfer work with a smaller block size? > >> (CONFIG_TFTP_BLOCKSIZE, default 1468, try 500 for instance). You could > >> also enable CONFIG_LWIP_DEBUG_RXTX in order to get a full hex dump of > >> the messages going into/out of the lwIP stack. > > > > Hi Jerome, > > > > My config was with 4096, setting it to 1468 works and anything above > > that seems to cause this failure. > > Good to know. > > > Some sort of fragmentation issue? > > Quite possibly, yes. In any case it does not make much sense to use a > clock size that causes the packet to be larger than the PMTU (it would > be sub-optimal) but it should definitely work, especially since I don't > believe PMTU discovery is implemented in lwIP. I add this to my to-do > list but feel free to investigate further :) >
Hi Jerome, I'll take a look at it as larger than MTU packets are definitely not suboptimal for TFTP. A 4096B block size will ack once per 4K so 4x less ACK's and in my testing the 4K blocksize gives a 3x performance boost on large files. Best Regards, Tim
