On 12/18/18 2:04 PM, Dominik Adamski wrote:
On 12/14/18 9:15 PM, Heinrich Schuchardt wrote:
On 12/14/18 3:58 PM, Dominik Adamski wrote:
On 12/12/18 7:11 PM, Heinrich Schuchardt wrote:
On 12/12/18 3:02 PM, Dominik Adamski wrote:
Hello, I think that I have found a bug in U-Boot UEFI implementation for BeagleBone Black board.
I have tested U-Boot UEFI implementation for BeagleBone Black. I am able to load Linux successfully via GRUB bootloader with U-Boot v2018.09-rc2 . U-Boot v2018.09-rc3 and above versions cause that the platform restarts over and over again. It starts booting, then it loads GRUB and when GRUB finishes its work an error occurs and the board restarts.
I have looked through U-Boot repository and I have found the commit, which breaks booting. It's name is as follows: "efi_loader: update runtime services table crc32" (commit ID: a39f39cdd8be5cd3e7a8b696a463b621e3d827e0 ) . I have figured out, that when I comment out function call: efi_update_table_header_crc32(&efi_runtime_services.hdr); in function efi_runtime_detach then I am able to launch Linux successfully.
My setup of U-Boot, GRUB and Linux was as follows: https://github.com/DominikAdamski/Beaglebone_GRUB_Manual . For U-Boot v2018.09 I get following log when booting:
U-Boot 2018.09 (Dec 11 2018 - 11:11:16 +0100)
CPU : AM335X-GP rev 2.1 Model: TI AM335x BeagleBone Black DRAM: 512 MiB NAND: 0 MiB MMC: OMAP SD/MMC: 0, OMAP SD/MMC: 1 Loading Environment from FAT... *** Warning - bad CRC, using default environment
No USB device found <ethaddr> not set. Validating first E-fuse MAC Net: eth0: ethernet@4a100000 Hit any key to stop autoboot: 2 ^H^H^H 1 ^H^H^H 0 switch to partitions #0, OK mmc0 is current device SD/MMC found on device 0 ** Unable to read file boot.scr ** 85 bytes read in 1 ms (83 KiB/s) Loaded env from uEnv.txt Importing environment from mmc0 ... Running uenvcmd ... 403968 bytes read in 28 ms (13.8 MiB/s) ^[[18tScanning disks on usb... Disk usb0 not ready Disk usb1 not ready Disk usb2 not ready Disk usb3 not ready Scanning disks on mmc... MMC Device 2 not found MMC Device 3 not found Found 4 disks WARNING: booting without device tree ## Starting EFI application at 82000000 ... ^[[?25h^[[0;30;47mWelcome to GRUB! ^M ^M^[[0;37;40m^[[0;37;40m^[[0;37;40mEFI stub: Booting Linux Kernel...^M EFI stub: Using DTB from configuration table^M EFI stub: Exiting boot services and installing virtual address map...^M U-Boot EFI: Relocation at 9ff4a8a8 is out of range (2008214f) data abort pc : [<9ff4a7cc>] lr : [<9ff4a8b9>] reloc pc : [<808007cc>] lr : [<808008b9>] sp : 9df1ebac ip : 00000020 fp : 00000000 r10: 00001000 r9 : 9df29eb8 r8 : 00000028 r7 : 9ff4aa10 r6 : 9ff4aec4 r5 : 00000003 r4 : 9ff4aa70 r3 : 2002aac0 r2 : 00000050 r1 : 9ff4aa70 r0 : ffffffff Flags: NzCv IRQs off FIQs on Mode SVC_32 Code: 61204621 f876f000 bd106120 b5f04b36 (b31c681c) UEFI image [0x9ceb1000:0x9cf139ff] '/\grub.efi' UEFI image [0x94728000:0x9510bfff] Resetting CPU ...
resetting ...
Please let me know, if I should add more detailed logs.
Dominik Adamski
Hello Dominik,
you refer to an outdated release candidate v2018.09-rc2.
The following patch is missing in the release candidate:
commit 483dbab9f9318149e5ea97daacbfae320f53e35a Author: Heinrich Schuchardt xypron.glpk@gmx.de Date: Sun Jul 29 09:49:03 2018 +0200
lib: crc32: mark function crc32() as __efi_runtime
It is contained in the v2018.09 release. Please, update your page https://github.com/DominikAdamski/Beaglebone_GRUB_Manual accordingly.
The latest release is v2018.11. Release v2018.11 will require a recent GRUB (later than patch d0c070179d4d). Or apply the following patch http://git.denx.de/?p=u-boot.git;a=commit;h=f31239acff61f7def88a06eef1f091fc...
Revert "efi_loader: remove efi_exit_caches()"
Best regards
Heinrich
Hello Heinrich,
I am resending my mail, because I haven't noticed that I should answer below your response. I am also adding some more information just to be sure, that everything is clear.
I have build the newest GRUB from master branch and I have checked out U-Boot with this patch:
http://git.denx.de/?p=u-boot.git;a=commit;h=f31239acff61f7def88a06eef1f091fc...
Revert "efi_loader: remove efi_exit_caches()"
and I still get an error while Linux booting :
EFI stub: Booting Linux Kernel... EFI stub: Using DTB from configuration table EFI stub: Exiting boot services and installing virtual address map... efi_runtime_relocate: Relocation at 9ff57a20 is out of range (20081092) data abort pc : [<9ff57944>] lr : [<9ff57a31>] reloc pc : [<80800944>] lr : [<80800a31>] sp : 9df2b594 ip : 1f757000 fp : 00000000 r10: 00002000 r9 : 9df36eb8 r8 : 00000028 r7 : 9ff57bc8 r6 : 9ff5807c r5 : 00000003 r4 : 9ff57c28 r3 : 20027c78 r2 : 00000050 r1 : 9ff57c28 r0 : ffffffff Flags: NzCv IRQs off FIQs on Mode SVC_32 Code: 46216120 f876f000 bd106120 4b36b5f0 (b31c681c) UEFI image [0x9cebf000:0x9cf21bff] '/\grub.efi' UEFI image [0x94732000:0x95115fff] Resetting CPU ...
resetting ...
I can boot Linux only when I remove line: efi_update_table_header_crc32(&efi_runtime_services.hdr); (file lib/efi_loader/efi_runtime.c line: 344 ) .
In my manual I referred to version v2018.09-rc2 because it is the last tagged version which works fine on BeagleBone with GRUB. v2018.09 version does not cooperate with GRUB for BB. It restarts BB over and over again. The problem is caused by this patch:
commit a39f39cdd8be5cd3e7a8b696a463b621e3d827e0 (HEAD) Author: Heinrich Schuchardt xypron.glpk@gmx.de Date: Sun Jul 29 09:49:04 2018 +0200
efi_loader: update runtime services table crc32
This patch is a child of the patch, which you have mentioned:
commit 483dbab9f9318149e5ea97daacbfae320f53e35a Author: Heinrich Schuchardt xypron.glpk@gmx.de Date: Sun Jul 29 09:49:03 2018 +0200
lib: crc32: mark function crc32() as __efi_runtime
The revert patch which you have mentioned: http://git.denx.de/?p=u-boot.git;a=commit;h=f31239acff61f7def88a06eef1f091fc...
Revert "efi_loader: remove efi_exit_caches()" does not solve the problem. I can boot Linux only if I revert this patch:
commit a39f39cdd8be5cd3e7a8b696a463b621e3d827e0 (HEAD) Author: Heinrich Schuchardt xypron.glpk@gmx.de Date: Sun Jul 29 09:49:04 2018 +0200
efi_loader: update runtime services table crc32
When I revert the wrong patch ( a39f39cdd8be5cd3e7a8b696a463b621e3d827e0 ), then I can boot Linux with U-Boot sources younger than v2018.09-rc2 .
Best regards, Dominik
Hello Dominik,
the message efi_runtime_relocate: Relocation at 9ff57a20 is out of range (20081092) looks like a runtime function trying to call a non-runtime function.
Actual addresses depend on compiler versions.
Please, supply the following files to Alex and me for further analysis:
.config u-boot.map u-boot lib/crc32.o lib/efi_loader/efi_runtime.o
which match the U-Boot version you use together with the matching error output. Please, also provide the output of the U-Boot command line tool bdinfo.
Do I understand you right in that the code you compiled is commit f31239acff61f7def88a06eef1f091fce74ecd61 with no further patches applied and the .config file is generated with am335x_evm_defconfig?
Best regards
Heinrich
Hello Heinrich, Yes, my code is commit f31239acff61f7def88a06eef1f091fce74ecd61 with no further patches applied and the .config file is generated with am335x_evm_defconfig .
Here is output of bdinfo command:
=> bdinfo arch_number = 0x00000e05 boot_params = 0x80000100 DRAM bank = 0x00000000 -> start = 0x80000000 -> size = 0x20000000 baudrate = 115200 bps TLB addr = 0x9fff0000 relocaddr = 0x9ff57000 reloc off = 0x1f757000 irq_sp = 0x9df2bb00 sp start = 0x9df2baf0 Early malloc usage: 1fc / 1000 fdt_blob = 0x9df2bb18 =>
I attach logs from booting and other files which you have requested.
Please let me know if you don't receive binary files.
Best regards, Dominik
Hello Dominik,
thanks for the files.
am335x_evm_defconfig is one of the boards with CONFIG_DYNAMIC_CRC_TABLE=y.
Lets analyze the log:
efi_runtime_relocate: Relocation at 9ff57a20 is out of range (20081092) data abort pc : [<9ff57944>] lr : [<9ff57a31>] reloc pc : [<80800944>] lr : [<80800a31>]
9ff57a20 - pc + reloc pc = 80800A20.
Looking at u-boot.map this is 0xE0 after the start of crc32_to_comp().
In `objdump -S arm-linux-gnueabihf-objdump -S crc32.o' we find these lines:
8080094a: 2601 movs r6, #1 for (n = 0; n < sizeof(p)/sizeof(Byte); n++) 8080094c: 463d mov r5, r7 8080094e: f8df c0d0 ldr.w ip, [pc, #208] ; 80800a20 <crc32_no_comp+0xe0> poly |= 1L << (31 - p[n]); 80800952: f815 400c ldrb.w r4, [r5, ip] 80800956: 3501 adds r5, #1
So the optimization in gcc has inlined function make_crc_table() as it is only used once.
80800a20 holds the pointer to: static const Byte p[] = {0,1,2,4,5,7,8,10,11,12,16,22,23,26};
Please, observe that it is static but not marked as __efi_runtime_data.
By chance pointer p contains an odd address and we get a data abort.
I will send out a patch now. Could you, please, test it.
Thanks for your patience and support to analyze the problem.
Best regards
Heinrich