Hi Peter, Thank you for your prompt reply.
I ran a git bisect between v2025.07 and v2025.10. The first bad commit is: bab54f5942c428be698216224fd10b91d974d4da net: phy: Do not do CL22 phy reset before ethernet phy driver probe Author: SkyLake.Huang [email protected] On Raspberry Pi 4 Model B Rev 1.5 / board revision d03115: - v2025.07 works - after this commit, DHCP still works - small TFTP transfers still work - larger TFTP transfers, for example start4.elf, stall with repeated “T” The bisect test was: setenv autoload no dhcp tftpboot $loadaddr boot/vg3/boot/start4.elf # MY LARGE FILE FOR TESTING The same test succeeds on good commits and fails on bad commits. Thank you. ### BITSECT LOG ### git bisect start # status: waiting for both good and bad commits # bad: [e50b1e8715011def8aff1588081a2649a2c6cd47] Prepare v2025.10 git bisect bad e50b1e8715011def8aff1588081a2649a2c6cd47 # status: waiting for good commit(s), bad commit known # good: [e37de002fac3895e8d0b60ae2015e17bb33e2b5b] Prepare v2025.07 git bisect good e37de002fac3895e8d0b60ae2015e17bb33e2b5b # bad: [a7d1214d2c9a38af3a9a1271dd8388068cba904c] bootstd: rauc: Fix potential memory leak git bisect bad a7d1214d2c9a38af3a9a1271dd8388068cba904c # bad: [d1b6489ef11bc89145e9fcf1d0f05d89a934e410] kbuild: do not overwrite .gitignore in output directory git bisect bad d1b6489ef11bc89145e9fcf1d0f05d89a934e410 # bad: [8aa384d01a079ec2d816f7051a7208a337727338] expo: Rename x and y in struct scene_obj_bbox git bisect bad 8aa384d01a079ec2d816f7051a7208a337727338 # bad: [24e7625085ee14a55275e2fbdaa5e595af1d632f] patman: Add a helper for managing cseries git bisect bad 24e7625085ee14a55275e2fbdaa5e595af1d632f # bad: [24776ab2766787d9629e7250486a0a662871b070] patman: Move Patch and Review to patchwork module git bisect bad 24776ab2766787d9629e7250486a0a662871b070 # good: [da24eb55327978410f5559f44ce1434ae8d8932d] Merge patch series "BOOTP/DHCPv4 enhancements" git bisect good da24eb55327978410f5559f44ce1434ae8d8932d # bad: [f74c6f9a81de1c0c04c578b0e6e2d8b1303d6d43] u_boot_pylib: Improve ANSI-colour output with backgrounds git bisect bad f74c6f9a81de1c0c04c578b0e6e2d8b1303d6d43 # bad: [96b0a77da0e3877308effd3d916c779fb3deb9df] binman: Read the skip-at-start value on startup git bisect bad 96b0a77da0e3877308effd3d916c779fb3deb9df # bad: [bab54f5942c428be698216224fd10b91d974d4da] net: phy: Do not do CL22 phy reset before ethernet phy driver probe git bisect bad bab54f5942c428be698216224fd10b91d974d4da # good: [44340456475da953a30c5cc4fbe8a9bc013c8c0f] net: tftp: don't call stop callback from UDP handler git bisect good 44340456475da953a30c5cc4fbe8a9bc013c8c0f # first bad commit: [bab54f5942c428be698216224fd10b91d974d4da] net: phy: Do not do CL22 phy reset before ethernet phy driver probe On Mon, May 4, 2026 at 12:55 PM Peter Robinson <[email protected]> wrote: > Hi Matteo, > > > TFTP downloads in U-Boot >= 2025.07 on Raspberry Pi 4 Model B Revision > 1.5 > > stall during transfer (continuous T output), while the same setup works > > correctly with U-Boot 2025.04. > > > > Please note that U-Boot starts correctly, DHCP works, and small TFTP > still > > works, but large TFTP stalls (e.g., kernel or initramfs files). > > > > On the contrary, Raspberry Pi 4 Model B Revision 1.4 does not have the > > stall issue and can download files correctly up to the latest stable > > version of U-Boot (2026.04). > > As far as I can tell from PCN [1] the only difference is the dual RAM > chips so I'm not sure what difference there may be and I only have rev > 1.1 and 1.4 of the 4B. Any chance you could do a git bisect between > 2025.04 and 2025.07 to work out what change causes the issue? > > Peter > > [1] https://pip.raspberrypi.com/categories/560-pcn >
