i suspect a usb kernel config issue - but no idea what there are so many to find.

hardware: using a semi-recent PC running an icore5

grub2 is on the usb thumb and loads kernel fine, kernel has static usb stuff (see below)

VFS: Cannot open root device "sdb2" ...

The kernel loads usbcore ... but the messages go by so quick i can't see them. my scroll lock doesn't seem to work on logitech to read it. with an iphone i photo'ed and saw usb-storage is registered before (scsi).

Next the kernel loads scsi/sata and says it finds "sda" (the pc hardrive which doesn't have linux on it yet, but not sdb,sdc or what. If i am runtime older pc and ranran 'fdisk -l' it'd list both sda and sdb with the usb plugged in at any time (without mounting).

based on some docs i tried kernel param: rootdelay=15. i even tried unplugging the USB and plugging it back in during that 15 seconds. didn't help, still not sdX arrived.

for root= I tried: sdb,sdc,sdd, and UUIDs. (note kernel never mounts never sees fstab)

# obvious. here's what i'd load if module, but it's compiled into kernel being used
usbcore
uhci-hcd # not required for usb-storage
usb-storage

I do realize it's possible to make an initrd or initramfs that would be read off USB at the time the kernel is. I haven't done so: takes time to prepare one. My reasoning is there's no module and no binary I know of that an initrd has to load. It should be all done at some point during boot using kernel i got.

I can't find ANY previous linuxquestion thread on the matter of when and how usb-storage memstick become available when plugged (at what stage in whole boot / runlevel thing). I just know if I add an HDD with linux and it booted that plugged memsticks would be detected as sdb,sdc,sdd. But i'd have to buy gear since my old pc hd is ide not sata.

All I know is allot of people make USB boot disks so there is a recipe not too well shared ?!

i'm just about to try a 71MB initrd shell to see if linux won't choke on it. but again i see no reason, read no reason, why it would help. i already know that merely entering initrd with a trivial shell "doesn't help", same VFS error after exit

tl;dr
what are you trying to boot?
and how did you create the bootable usb?

You claim you don't require initrd to boot the system, yet the system won't boot when you don't have one. Sometimes the system will enumerate the USB drive as /dev/sda when you're booting off it.

no and yes.

no it's not even close to true a system needs an initrd to boot. if you built modules into the kernel you need to boot you do not need any initrd at all. the initrd is for solving the "mainstream linux OS" problem of making a kernel for everyone (people who don't built their own kernel). go to www.kernel.org and try it.

yes linux assigns sdX in a bus order manner that can change. however i mentioned about that isn't the issue and that UUID is being used.

linux kernel 2.6.3x with modules "built-in" to kernel i need to boot is being used on old and hopefully new system. userland is gcc-4.4.5-ish and i'd use linux kernel 4.16.12 if i knew of a firm reason to do so. but upgrading kernel when it's not what's wrong - well many have done it only to find out they still have more to do. i was thinking of upgrading kernel after i got the thing working actually (because the newer pc compiles faster and has more hd space)

the grub-2.02 itself was easy to create. i merely compiled grub-2.02 from source and followed the directions of it's manual.

what about the drivers for the usb host controller ehci/xhci?

i skimmed a few kernel changelogs by eye and see USB has had an extremum of problemed releses. several serious and non-serious usb fixes per changelog.

"the industry" promoted USB as cheaper and less complicated than Firewire like it did ide versus scsi. Looking at the record on usb being hacked up to speed 3.0 and the resulting "changes" and reliance on kernel drivers having to service the weak protocols: i wonder if that's true - that it's cheaper and simpler.

I've never been able to configure a kernel to boot without initrd. I know it's possible. But it isn't easy.

It's very easy if your system is on the hard drive! I've been doing it for years. All you have to do is make sure the pata/sata driver and the root filesystem driver are included in the kernel, not built as modules.

I've heard of people having problems with systems on memory sticks. It's usually due to timing errors. But OP says he's used rootdelay and it didn't help.

so you have grub on sdb? and where is the kernel sdb?/boot.
can you post your grub.cfg that you are trying to use.
I take it you can boot from a usb device "bios" allowing

they are in kernel, static, no need to "load modules"

So you have your usb storage built in plus sata driver plus ext4 filesystem driver.

btw the modern sata driver (libata) handles ide disks too, so you could use your hard drive if you had to.

grub2 is not an issue: it already loads kernel and initrd, kernel starts

kernel parameters may be an issue. but it was said sdb and UUID were tried / used. the question on parameters is if there is some magic parameter for making usb-storage appear before mounting root (before the VFS message)

i don't use ext4. but your right and wrong both: i could risk moving the old ide drive into the new pc. but it would require $purchases and risk of (ie, of dropping it, of twisting a pin, whatever)

it'd be beautiful if these PCs were more network friendly. they aren't!

here's a for example i'm wondering about

years ago i made a bootable CD. if i remember, the initrd or linuxrc had to have a "second root" because the bios would not let go of the original drive: one had to load a ramdisk root, unmount, remount, the mount the "real root". but cdrom was always a backwards thing: ie, it couldn't be formatted ext2 with a boot partition - they needed to use el torito or something.

i have ZERO idea if those kinds of issues are involved for usb flash. i'm rather hoping it's more like booting from a floppy drive. (that the bios and kernel both have rigorous treatment and require not games being played)