jim-wilson
commented
6 years ago I tried looking at this. Running spike with the -l option, I see
core 0: 0x0000000080003434 (0x00066703) lwu a4, 0(a2)
core 0: exception trap_load_address_misaligned, epc 0x0000000080003434
core 0: badaddr 0xffffffe0007a807e
Looking at the linux kernel System.map file I see
ffffffe0007a8056 t next_event
ffffffe0007a8098 T timer_riscv_init
so the badaddr is inside the next_event function. Disassembling this function I see
ffffffe0007a8056
It looks like most of the SDK was built with the default rv64imafdc, and only riscv-pk got the -march=rv64imafd option. I don't know why this didn't fail earlier. I can fix the kernel by adding CONFIG_RISCV_ISA_C=n to the top of the conf/linux_defconfig file. Forcing the rebuild of the kernel I can now boot to the login prompt.
However, I see that buildroot busybox still has compressed instructions in it, though I'm not seeing failures inside spike. The -march option should be passed down to buildroot somehow, perhaps via CFLAGS. And you also need to pass the -march option down to spike somehow also. I see that spike has a --isa option, so if I change the toplevel sim rule to $(spike) --isa=rv64imafd -p4 $(bbl) that seems to do the right thing, but now I get a linux kernel panic, because init has compressed instructions in it. And those compressed instructions apparently come from ld.so from glibc. I think the reason things mostly worked even with compressed instructions is because spike was silently executing the compressed instructions by default. When the spike compressed support is turned off, things break badly, but it shows where the build problems are.
I think you need to pass the -march option down into the toolchain build also to get ld.so built correctly without compressed instructions. This needs to be passed to the toolchain configure command.
Clearly, this is not something supported out of the fox, but it looks doable with a bit more work.
Vasanvs
richardxia
palmer-dabbelt
We are trying to build the tool chain without the support of compression ISA set.
We did the following changes in the different files for this purpose: a) In the Makefile @ freedom-u-sdk directory, we changed the rule for building bbl as follows: $(bbl): $(pk_srcdir) $(vmlinux_stripped) rm -rf $(pk_wrkdir) mkdir -p $(pk_wrkdir) cd $(pk_wrkdir) && $</configure \ --host=$(target) \ --with-payload=$(vmlinux_stripped) \ --enable-logo \ --with-logo=$(abspath conf/sifive_logo.txt) CFLAGS="-mabi=lp64d -march=rv64imafd" $(MAKE) -C $(pk_wrkdir)
b) In the riscv-pk/configure script, made the change in the CFLAGS and LDFLAGS for a 64-bit architecture as follows:
case "${BUILD_32BIT}" in yes|default) echo "Building 32-bit pk" CFLAGS="$default_CFLAGS -march=rv32imafdc -mabi=ilp32d" LDFLAGS="-march=rv32imafdc -mabi=ilp32d" install_subdir="riscv32-unknown-elf" ;; *) CFLAGS="$default_CFLAGS -march=rv64imafd -mabi=lp64d" LDFLAGS="-march=rv64imafd -mabi=lp64d" install_subdir="riscv64-unknown-elf" ;; esac
After making the above changes in the files, did a 'make clean' and then issued a fresh make and then subsequently make sim. Verified that there was no reference of architecture as rv64imafdc and reference of architecture was only rv64imafd - in the entire make build output - thereby proving that the architecture was never getting set as rv64imafdc through out the entire repository build.
When the bbl has got built and appended with the vmlinux as payload as part of this make, and then spike started this bbl loading, there is an error message that is coming in as part of the load process as below:
[ 0.000000] OF: fdt: Ignoring memory range 0x80000000 - 0x80200000 [ 0.000000] Linux version 4.14.0-rc7-00028-g160fa84-dirty (admin1@admin) (gcc version 7.2.0 (GCC)) #1 SMP Mon Nov 13 14:19:43 IST 2017 [ 0.000000] bootconsole [early0] enabled [ 0.000000] Initial ramdisk at: 0xffffffe00001e200 (6598656 bytes) [ 0.000000] Zone ranges: [ 0.000000] DMA [mem 0x0000000080200000-0x00000000ffffffff] [ 0.000000] Normal empty [ 0.000000] Movable zone start for each node [ 0.000000] Early memory node ranges [ 0.000000] node 0: [mem 0x0000000080200000-0x00000000ffffffff] [ 0.000000] Initmem setup node 0 [mem 0x0000000080200000-0x00000000ffffffff] [ 0.000000] elf_hwcap is 0x112d [ 0.000000] percpu: Embedded 14 pages/cpu @ffffffe07e1ee000 s28376 r0 d28968 u57344 [ 0.000000] Built 1 zonelists, mobility grouping on. Total pages: 516615 [ 0.000000] Kernel command line: earlyprintk [ 0.000000] PID hash table entries: 4096 (order: 3, 32768 bytes) [ 0.000000] Dentry cache hash table entries: 262144 (order: 9, 2097152 bytes) [ 0.000000] Inode-cache hash table entries: 131072 (order: 8, 1048576 bytes) [ 0.000000] Sorting __ex_table... [ 0.000000] Memory: 2052044K/2095104K available (2826K kernel code, 239K rwdata, 758K rodata, 6599K init, 774K bss, 43060K reserved, 0K cma-reserved) [ 0.000000] SLUB: HWalign=64, Order=0-3, MinObjects=0, CPUs=1, Nodes=1 [ 0.000000] Hierarchical RCU implementation. [ 0.000000] RCU event tracing is enabled. [ 0.000000] RCU restricting CPUs from NR_CPUS=8 to nr_cpu_ids=1. [ 0.000000] RCU: Adjusting geometry for rcu_fanout_leaf=16, nr_cpu_ids=1 [ 0.000000] NR_IRQS: 0, nr_irqs: 0, preallocated irqs: 0 [ 0.000000] riscv,cpu_intc,0: 64 local interrupts mapped [ 0.000000] clocksource: riscv_clocksource: mask: 0xffffffffffffffff max_cycles: 0x24e6a1710, max_idle_ns: 440795202120 ns /home/admin1/Desktop/riscv/new-freedom-u-sdk/freedom-u-sdk/riscv-pk/machine/mtrap.c:18: machine mode: unhandlable trap 4 @ 0x0000000080003434
On the disassembled bbl executable, 80003434 is having: 80003434: 00066703 lwu a4,0(a2) # 1000 <_ftext-0x7ffff000>
For having the support of RV64IMAFD and NOT having the support of RV64IMAFDC ISA, should there be anything else that is done for a successful load up in spike till the login prompt?
Thanks in advance for your inputs and help,
Vasan