[PATCH] Revert "net: eth_bootdev_hunt() should not run DHCP"

[Heiko Stübner]

Am Dienstag, 1. April 2025, 18:13:35 MESZ schrieb Heinrich Schuchardt:
> On 01.04.25 17:51, Simon Glass wrote:
> > On Tue, 1 Apr 2025 at 21:38, Heiko Stuebner <heiko at sntech.de> wrote:
> >>
> >> This reverts commit 1f68057e03206e6597ca8b2be8bb1c49d4bd47d0.
> >>
> >> Commit 1f68057e0320 ("net: eth_bootdev_hunt() should not run DHCP")
> >> aims to reduce EFI boot times by disabling the dhcp_run when
> >> checking ethernet bootdevices, by preventing it from running double,
> >> with the reasoning
> >>
> >>      We need to call eth_bootdev_hunt() when setting up the EFI sub-system to
> >>      supply the simple network protocol. We don't need an IP address set up.
> >>
> >> That might by true for EFI, but not for everything else, because when
> >> running distro-boot and for example the PXE method in it, nothing will
> >> set up an IP address now.
> 
> The removed call was dhcp_run(addr, NULL, true);
> 
> We have:
> 
> distro_efi_read_bootflow_net():
> boot/bootmeth_efi.c:205:        ret = dhcp_run(addr, NULL, true);
> 
> script_read_bootflow_net():
> boot/bootmeth_script.c:132:     ret = dhcp_run(addr, fname, true);
> 
> extlinux_pxe_read_bootflow() seems to be lacking the call.
> 
> So instead of reverting
> 1f68057e0320 ("net: eth_bootdev_hunt() should not run DHCP")
> we should add the missing call in extlinux_pxe_read_bootflow().

doing

----- 8< -----
diff --git a/boot/bootmeth_pxe.c b/boot/bootmeth_pxe.c
index b91e61bcbc4..6e5e0f99ea4 100644
--- a/boot/bootmeth_pxe.c
+++ b/boot/bootmeth_pxe.c
@@ -73,6 +73,10 @@ static int extlinux_pxe_read_bootflow(struct udevice *dev,
                return log_msg_ret("pxeb", -EPERM);
        addr = simple_strtoul(addr_str, NULL, 16);
 
+       ret = dhcp_run(addr, NULL, false);
+       if (ret)
+               return log_msg_ret("dhc", ret);
+
        log_debug("calling pxe_get()\n");
        ret = pxe_get(addr, &bootdir, &size, false);
        log_debug("pxe_get() returned %d\n", ret);
----- 8< -----

does seem to work in _my_ usecase and gets me loading stuff over pxe
again.

I guess this whole thing now becomes a strategy decision ;-) :

- it's very late in the release, do we revert to the known working state
  or hope the above fixes all issues
- Simon's wish of not sprinkling dhcp_run calls in multiple places

I guess I'm fine with both way, as either fixes my use-case ;-)

Heiko