Hi Tom and Simon,
On 2024-04-18 20:13, Tom Rini wrote:
On Sun, Apr 14, 2024 at 06:58:03PM +0200, Jonas Karlman wrote:
Hi Tom and Simon,
On 2024-04-11 03:45, Tom Rini wrote:
On Thu, 14 Dec 2023 21:18:58 -0700, Simon Glass wrote:
This series is the culmanation of the current line of refactoring series. It adjusts pxe to call the booting functionality directly rather than going through the command-line interface.
With this is is possible to boot using the extlinux bootmeth without the command line enabled.
[...]
Applied to u-boot/master, thanks!
This series is causing boot issues using extlinux in bootm_run_states():
ERROR: booting os 'Invalid OS' (0) is not supported
Following extlinux.conf was used:
label linux kernel /Image.gz initrd /initramfs.cpio.gz
Before this series booting works, bootm_run_states() is first called with states=0x1 (BOOTM_STATE_START):
Scanning bootdev 'mmc@fe2b0000.bootdev': 1 extlinux ready mmc 1 mmc@fe2b0000.bootdev.part /extlinux/extlinux.conf ** Booting bootflow 'mmc@fe2b0000.bootdev.part_1' with extlinux 1: linux Retrieving file: /Image.gz Retrieving file: /initramfs.cpio.gz bootm_run_states(): images->state: 0, states: 1 bootm_run_states(): images->os.os: 0 bootm_run_states(): images->os.arch: 0 bootm_run_states(): boot_fn: 0000000000000000, need_boot_fn: 0 Uncompressing Kernel Image to 0 ## Flattened Device Tree blob at edef8410 Booting using the fdt blob at 0xedef8410 Working FDT set to edef8410 bootm_run_states(): images->state: 1, states: 1710 Loading Ramdisk to ecdfd000, end eceb274d ... OK bootm_run_states(): images->os.os: 5 bootm_run_states(): images->os.arch: 16 boot_fn: 00000000eff2b83c, need_boot_fn: 0 Loading Device Tree to 00000000ecde8000, end 00000000ecdfc97f ... OK Working FDT set to ecde8000
After this series booting fails, bootm_run_states() is first called with states=0x1710.
Scanning bootdev 'mmc@fe2b0000.bootdev': 1 extlinux ready mmc 1 mmc@fe2b0000.bootdev.part /extlinux/extlinux.conf ** Booting bootflow 'mmc@fe2b0000.bootdev.part_1' with extlinux 1: linux Retrieving file: /Image.gz Retrieving file: /initramfs.cpio.gz bootm_run_states(): images->state: 0, states: 1710 bootm_run_states(): images->os.os: 0 bootm_run_states(): images->os.arch: 0 bootm_run_states(): boot_fn: 0000000000000000, need_boot_fn: 0 ERROR: booting os 'Invalid OS' (0) is not supported Boot failed (err=-14)
Looks like booti_start() -> bootm_run_states(bmi, BOOTM_STATE_START) is no longer called due to changes in this series.
I think the problem is with: commit 6d803ec9cc757bda0c48f2fd419cb6878eab92c4 Author: Simon Glass sjg@chromium.org Date: Thu Dec 14 21:19:12 2023 -0700
pxe: Allow booting without CMDLINE for bootm methods Use bootm_run() and booti_run() to boot rather than the command line. This allows extlinux to be used without CMDLINE being enabled. Collect any error but do not return it, to match the existing code. Signed-off-by: Simon Glass <sjg@chromium.org>diff --git a/boot/pxe_utils.c b/boot/pxe_utils.c [snip] @@ -641,23 +644,18 @@ static int label_run_boot(struct pxe_context *ctx, struct pxe_label *label, if (!fdt_addr && genimg_get_format(buf) != IMAGE_FORMAT_FIT) fdt_addr = env_get("fdtcontroladdr");
- if (fdt_addr) {
if (!bootm_argv[2])bootm_argv[2] = "-";bootm_argc = 4;- }
- bootm_argv[3] = (char *)fdt_addr;
bmi.conf_fdt = fdt_addr;
/* Try bootm for legacy and FIT format image */ if (genimg_get_format(buf) != IMAGE_FORMAT_INVALID &&
IS_ENABLED(CONFIG_CMD_BOOTM))do_bootm(ctx->cmdtp, 0, bootm_argc, bootm_argv);
IS_ENABLED(CONFIG_BOOTM))ret = bootm_run(&bmi);
- else if (IS_ENABLED(CONFIG_CMD_BOOTI))
do_booti(ctx->cmdtp, 0, bootm_argc, bootm_argv);
- else if (IS_ENABLED(CONFIG_BOOTM))
ret = booti_run(&bmi);
- else if (IS_ENABLED(CONFIG_CMD_BOOTZ))
do_bootz(ctx->cmdtp, 0, bootm_argc, bootm_argv);
- else if (IS_ENABLED(CONFIG_BOOTM))
ret = bootz_run(&bmi);And this doesn't seem equivalent really. The logic used to be checking if we had bootm/booti/bootz and now it's always checking for CONFIG_BOOTM. Jonas, can you please share the defconfig you used here as well? But I think for now reverting the series is the best path forward, unfortunately.
I have re-tested following on rockpro64-rk3399_defconfig using:
- Image.gz: plain Linux kernel v6.9-rc4, ARCH=arm64 defconfig (without module support) - initramfs.cpio.gz: minimal busybox system - extlinux/extlinux.conf: minimal extlinux file
""" label linux kernel /Image.gz initrd /initramfs.cpio.gz """
And following diff to add debug prints:
""" diff --git a/boot/bootm.c b/boot/bootm.c index 032f5a4a1605..e106cde7076b 100644 --- a/boot/bootm.c +++ b/boot/bootm.c @@ -998,6 +998,7 @@ int bootm_run_states(struct bootm_info *bmi, int states) ulong iflag = 0; int ret = 0, need_boot_fn;
+ printf("%s(): images->state: %x, states: %x\n", __func__, images->state, states); images->state |= states;
/* @@ -1060,10 +1061,13 @@ int bootm_run_states(struct bootm_info *bmi, int states) /* From now on, we need the OS boot function */ if (ret) return ret; + printf("%s(): images->os.os: %x\n", __func__, images->os.os); + printf("%s(): images->os.arch: %x\n", __func__, images->os.arch); boot_fn = bootm_os_get_boot_func(images->os.os); need_boot_fn = states & (BOOTM_STATE_OS_CMDLINE | BOOTM_STATE_OS_BD_T | BOOTM_STATE_OS_PREP | BOOTM_STATE_OS_FAKE_GO | BOOTM_STATE_OS_GO); + printf("%s(): boot_fn: %p, need_boot_fn: %x\n", __func__, boot_fn, need_boot_fn); if (boot_fn == NULL && need_boot_fn) { if (iflag) enable_interrupts(); """
Result for the U-Boot prior to you revert using commit 3434b88d2c2f ("Merge branch 'master-fdt' of https://source.denx.de/u-boot/custodians/u-boot-sh"):
Scanning bootdev 'mmc@fe320000.bootdev': 1 extlinux ready mmc 1 mmc@fe320000.bootdev.part /extlinux/extlinux.conf ** Booting bootflow 'mmc@fe320000.bootdev.part_1' with extlinux 1: linux Retrieving file: /Image.gz Retrieving file: /initramfs.cpio.gz bootm_run_states(): images->state: 0, states: 1710 bootm_run_states(): images->os.os: 0 bootm_run_states(): images->os.arch: 0 bootm_run_states(): boot_fn: 0000000000000000, need_boot_fn: 700 ERROR: booting os 'Invalid OS' (0) is not supported Boot failed (err=-14)
And using an uncompressed /Image does not make any difference.
Result for the U-Boot after to your revert using commit cdd20e3f66fe ("Revert "Merge patch series "pxe: Allow extlinux booting without CMDLINE enabled"""):
Scanning bootdev 'mmc@fe320000.bootdev': 1 extlinux ready mmc 1 mmc@fe320000.bootdev.part /extlinux/extlinux.conf ** Booting bootflow 'mmc@fe320000.bootdev.part_1' with extlinux 1: linux Retrieving file: /Image.gz Retrieving file: /initramfs.cpio.gz bootm_run_states(): images->state: 0, states: 1 bootm_run_states(): images->os.os: 0 bootm_run_states(): images->os.arch: 0 bootm_run_states(): boot_fn: 0000000000000000, need_boot_fn: 0 Uncompressing Kernel Image to 0 Moving Image from 0x2080000 to 0x2200000, end=6df0000 ERROR: RD image overlaps OS image (OS=2200000..6df0000) Boot failed (err=-14)
Changing the RD image addr fixes such overlap (a separate issue not present on my initial testing board, generic-rk3568_defconfig):
=> env set ramdisk_addr_r 10000000 => boot
Scanning bootdev 'mmc@fe320000.bootdev': 1 extlinux ready mmc 1 mmc@fe320000.bootdev.part /extlinux/extlinux.conf ** Booting bootflow 'mmc@fe320000.bootdev.part_1' with extlinux 1: linux Retrieving file: /Image.gz Retrieving file: /initramfs.cpio.gz bootm_run_states(): images->state: 1, states: 1 bootm_run_states(): images->os.os: 0 bootm_run_states(): images->os.arch: 0 bootm_run_states(): boot_fn: 0000000000000000, need_boot_fn: 0 Uncompressing Kernel Image to 0 Moving Image from 0x2080000 to 0x2200000, end=6df0000 ## Flattened Device Tree blob at f1ef7c70 Booting using the fdt blob at 0xf1ef7c70 Working FDT set to f1ef7c70 bootm_run_states(): images->state: 1, states: 1710 Loading Ramdisk to f0dfb000, end f0eb074d ... OK bootm_run_states(): images->os.os: 5 bootm_run_states(): images->os.arch: 16 bootm_run_states(): boot_fn: 00000000f3f13828, need_boot_fn: 700 Loading Device Tree to 00000000f0de6000, end 00000000f0dfa0e7 ... OK Working FDT set to f0de6000 Starting kernel ...
I hope this can give some more details on how to reproduce the issue.
Regards, Jonas