init:main.c: start 'usermodehelper_enable()' a little early

commit d5767c53535ac79758084773418e0ad186aba4a2 move 'usermodehelper_enable()' to end of rest_init, then I get failed to let uvesafb work on my computer, and lose the splash boot.So maybe we could start usermodehelper_enable a little early to make some task work that need eary init with the help of user mode.

---
 init/main.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

index 23702bb..03b408d 100644
@@ -730,8 +730,8 @@ static void __init do_basic_setup(void)
 	driver_init();
 	init_irq_proc();
 	do_ctors();
-	do_initcalls();
 	usermodehelper_enable();
+	do_initcalls();
 }
 
 static void __init do_pre_smp_initcalls(void)

1.7.3.4

Grr. Initcalls *really* shouldn't depend on usermode helpers.

That said, I'm not entirely shocked if they do. But could you try to describe *which* initcall it is? The call trace should tell you?

I'm not saying the patch is wrong (and it may be what we have to do at this point), but I do think it would be even better if we could fix the real problem of initcalls calling usermode helpers instead...

Linus

Yes, I also want to fix the *really* problem,and this is the call trace:

[    0.542894] msgmni has been set to 1740
[    0.543128] Block layer SCSI generic (bsg) driver version 0.4
loaded (major 253)
[    0.543138] io scheduler noop registered
[    0.543140] io scheduler deadline registered
[    0.543151] io scheduler cfq registered (default)
[    0.543488] Pid: 1, comm: swapper Not tainted 3.1.0-rc8NX3+ #13
[    0.543490] Call Trace:
[    0.543499]  [<c11728dd>] uvesafb_exec+0x12d/0x238
[    0.543504]  [<c12aaeff>] uvesafb_probe+0x7e/0xb1e
[    0.543510]  [<c11be53f>] platform_drv_probe+0xc/0xe
[    0.543513]  [<c11bd75d>] driver_probe_device+0x81/0xfd
[    0.543516]  [<c11bd862>] __device_attach+0x2a/0x2e
[    0.543520]  [<c11bcd64>] bus_for_each_drv+0x3d/0x67
[    0.543523]  [<c11bd8d5>] device_attach+0x50/0x6b
[    0.543526]  [<c11bd838>] ? __driver_attach+0x5f/0x5f
[    0.543529]  [<c11bcbef>] bus_probe_device+0x18/0x2d
[    0.543531]  [<c11bc0fc>] device_add+0x37a/0x4b2
[    0.543535]  [<c13ff1e0>] ? parse_early_options+0x1c/0x1c
[    0.543538]  [<c11bb26d>] ? dev_set_name+0x14/0x16
[    0.543541]  [<c11beab2>] platform_device_add+0xe4/0x12b
[    0.543544]  [<c13ff1e0>] ? parse_early_options+0x1c/0x1c
[    0.543547]  [<c12aaadd>] uvesafb_init+0x307/0x354
[    0.543550]  [<c13ff1e0>] ? parse_early_options+0x1c/0x1c
[    0.543554]  [<c1001159>] do_one_initcall+0x71/0x113
[    0.543557]  [<c12aa7d6>] ? uvesafb_is_valid_mode+0x56/0x56
[    0.543560]  [<c13ff1e0>] ? parse_early_options+0x1c/0x1c
[    0.543562]  [<c13ff264>] kernel_init+0x84/0xfd
[    0.543565]  [<c12bc4b6>] kernel_thread_helper+0x6/0xd
[    0.543709] v86d used greatest stack depth: 7028 bytes left
[    0.583239] uvesafb: NVIDIA Corporation, GT216 Board - 0696a290,
Chip Rev   , OEM: NVIDIA, VBE v3.0
[    0.606852] uvesafb: protected mode interface info at c000:d350
[    0.606854] uvesafb: pmi: set display start = c00cd3b3, set palette
= c00cd40e
[    0.606856] uvesafb: pmi: ports = 3b4 3b5 3ba 3c0 3c1 3c4 3c5 3c6
3c7 3c8 3c9 3cc 3ce 3cf 3d0 3d1 3d2 3d3 3d4 3d5 3da
[    0.610376] uvesafb: VBIOS/hardware doesn't support DDC transfers
[    0.610378] uvesafb: no monitor limits have been set, default
refresh rate will be used
[    0.610528] uvesafb: scrolling: ypan using protected mode
interface, yres_virtual=4915
[    0.739820] Console: switching to colour frame buffer device 80x30
[    0.740634] uvesafb: framebuffer at 0xcf000000, mapped to
0xf8480000, using 6144k, total 14336k
[    0.740636] fb0: VESA VGA frame buffer device

maybe uvesafb has some specific thing that it requires a userspace helper application called 'v86d' to work!
I will try to make uvesafb workaround it, but i don't know whether it will work, and splash boot *really* need
early init.

Ok. I guess we cannot avoid this. I really would have preferred to run the kernel init routines without usermodehelper enabled, but it looks like reality doesn't care about my wishes.

Your "move it up to before all initcalls" works, and seems to be the simplest solution. I considered doing this when we mount the root filesystem, but depending on config options that is in multiple places. We could do the usermode helper enable as a rootfs_initcall()..

I'm nervous about this. Enabling the usermode helpers too early can re-introduce the bug where we do it *much* too early before everything is initialized. But yes, uvesafb clearly does depend on being able to execute v86d from the root filesystem.

So I think I'll just have to apply your patch. The alternatives look uglier.

Thanks for the report and the debug help,

Linus

It is my glory, Linus.Thanks