On Thu, Nov 05, 2020 at 09:12:25PM -0800, Palmer Dabbelt wrote:
> On Thu, 05 Nov 2020 20:59:02 PST (-0800), natechancellor@xxxxxxxxx wrote:
> > On Thu, Nov 05, 2020 at 08:42:22PM -0800, 'Palmer Dabbelt' via Clang Built Linux wrote:
> > > On Mon, 26 Oct 2020 12:48:08 PDT (-0700), Nick Desaulniers wrote:
> > > > On Fri, Oct 23, 2020 at 10:03 PM 'Palmer Dabbelt' via Clang Built
> > > > Linux <clang-built-linux@xxxxxxxxxxxxxxxx> wrote:
> > > > >
> > > > > We were relying on GNU ld's ability to re-link executable files in order
> > > > > to extract our VDSO symbols. This behavior was deemed a bug as of
> > > > > binutils-2.35 (specifically the binutils-gdb commit a87e1817a4 ("Have
> > > > > the linker fail if any attempt to link in an executable is made."), but as that
> > > > > has been backported to at least Debian's binutils-2.34 in may manifest in other
> > > > > places.
> > > > >
> > > > > The previous version of this was a bit of a mess: we were linking a
> > > > > static executable version of the VDSO, containing only a subset of the
> > > > > input symbols, which we then linked into the kernel. This worked, but
> > > > > certainly wasn't a supported path through the toolchain. Instead this
> > > > > new version parses the textual output of nm to produce a symbol table.
> > > > > Both rely on near-zero addresses being linkable, but as we rely on weak
> > > > > undefined symbols being linkable elsewhere I don't view this as a major
> > > > > issue.
> > > > >
> > > > > Fixes: e2c0cdfba7f6 ("RISC-V: User-facing API")
> > > > > Cc: clang-built-linux@xxxxxxxxxxxxxxxx
> > > > > Cc: stable@xxxxxxxxxxxxxxx
> > > > > Signed-off-by: Palmer Dabbelt <palmerdabbelt@xxxxxxxxxx>
> > > >
> > > > Any way to improve the error message if/when this fails?
> > > > https://travis-ci.com/github/ClangBuiltLinux/continuous-integration/jobs/407165683
> > >
> > > Probably, but I can't get that command to actually run this stuff. I tried
> > > pulling the commands, but I'm getting some weirdness
> > >
> > > $ rm -f arch/riscv/kernel/vdso/vdso-syms.S
> > > $ make ARCH=riscv defconfig
> > > $ make -j2 AR=llvm-ar 'CC=clang' 'HOSTCC=clang' HOSTLD=ld KCFLAGS=-Wno-implicit-fallthrough LD=riscv64-linux-gnu-ld LLVM_IAS=1 NM=llvm-nm OBJCOPY=llvm-objcopy OBJDUMP=llvm-objdump OBJSIZE=llvm-size READELF=llvm-readelf STRIP=llvm-strip ARCH=riscv Image
> >
> > This command is simpler and reproduces it for me locally on next-20201105.
> >
> > $ make -skj"$(nproc)" ARCH=riscv CROSS_COMPILE=riscv64-linux-gnu- LD=riscv64-linux-gnu-ld LLVM=1 distclean defconfig Image
> > clang-12: warning: argument unused during compilation: '-no-pie' [-Wunused-command-line-argument]
> > make[4]: *** [arch/riscv/kernel/vdso/Makefile:53: arch/riscv/kernel/vdso/vdso-syms.S] Error 1
> > make[4]: *** Deleting file 'arch/riscv/kernel/vdso/vdso-syms.S'
> > make[4]: Target '__build' not remade because of errors.
> > make[3]: *** [scripts/Makefile.build:500: arch/riscv/kernel/vdso] Error 2
> > make[3]: Target '__build' not remade because of errors.
> > make[2]: *** [scripts/Makefile.build:500: arch/riscv/kernel] Error 2
> > make[2]: Target '__build' not remade because of errors.
> > make[1]: *** [Makefile:1797: arch/riscv] Error 2
> > make[1]: Target 'Image' not remade because of errors.
> > make: *** [Makefile:335: __build_one_by_one] Error 2
> > make: Target 'distclean' not remade because of errors.
> > make: Target 'defconfig' not remade because of errors.
> > make: Target 'Image' not remade because of errors.
> >
> > <snip>
> >
> > > diff --git a/arch/riscv/kernel/vdso/.gitignore b/arch/riscv/kernel/vdso/.gitignore
> > > index 3a19def868ec..88206dd8b472 100644
> > > --- a/arch/riscv/kernel/vdso/.gitignore
> > > +++ b/arch/riscv/kernel/vdso/.gitignore
> > > @@ -2,3 +2,4 @@
> > > vdso.lds
> > > *.tmp
> > > vdso-syms.S
> > > +vdso-syms.nm
> > > diff --git a/arch/riscv/kernel/vdso/Makefile b/arch/riscv/kernel/vdso/Makefile
> > > index a8ecf102e09b..fe5c969a6bf4 100644
> > > --- a/arch/riscv/kernel/vdso/Makefile
> > > +++ b/arch/riscv/kernel/vdso/Makefile
> > > @@ -49,8 +49,11 @@ SYSCFLAGS_vdso.so.dbg = -shared -s -Wl,-soname=linux-vdso.so.1 \
> > > # We also create a special relocatable object that should mirror the symbol
> > > # table and layout of the linked DSO. With ld --just-symbols we can then
> > > # refer to these symbols in the kernel code rather than hand-coded addresses.
> > > -$(obj)/vdso-syms.S: $(obj)/vdso.so FORCE
> > > - $(call if_changed,so2s)
> > > +$(obj)/vdso-syms.nm: $(obj)/vdso.so
> > > + $(call if_changed,nm_d)
> > > +
> > > +$(obj)/vdso-syms.S: $(obj)/vdso-syms.nm
> > > + $(call if_changed,nm2s)
> > >
> > > # strip rule for the .so file
> > > $(obj)/%.so: OBJCOPYFLAGS := -S
> > > @@ -68,9 +71,13 @@ quiet_cmd_vdsold = VDSOLD $@
> > > $(patsubst %, -G __vdso_%, $(vdso-syms)) $@.tmp $@ && \
> > > rm $@.tmp
> > >
> > > -# Extracts
> > > -quiet_cmd_so2s = SO2S $@
> > > - cmd_so2s = $(NM) -D $< | $(srctree)/$(src)/so2s.sh > $@
> > > +# Extracts symbol offsets from the VDSO, converting them into an assembly file
> > > +# that contains the same symbols at the same offsets.
> > > +quiet_cmd_nm_d = NM -D $@
> > > + cmd_nm_d = $(NM) -D $< > $@
> > > +
> > > +quiet_cmd_nm2s = SYMS2S $@
> > > + cmd_nm2s = cat $< | $(srctree)/$(src)/so2s.sh > $@
> > >
> > > # install commands for the unstripped file
> > > quiet_cmd_vdso_install = INSTALL $@
> > >
> > > For reference, here's the output of nmo for me:
> > >
> > > $ cat arch/riscv/kernel/vdso/vdso-syms.nm 0000000000000000 A LINUX_4.15
> > > 00000000000009e0 T __vdso_clock_getres@@LINUX_4.15
> > > 000000000000080a T __vdso_clock_gettime@@LINUX_4.15
> > > 0000000000000a48 T __vdso_flush_icache@@LINUX_4.15
> > > 0000000000000a3c T __vdso_getcpu@@LINUX_4.15
> > > 0000000000000916 T __vdso_gettimeofday@@LINUX_4.15
> > > 0000000000000800 T __vdso_rt_sigreturn@@LINUX_4.15
> >
> > This diff does not solve the issue for me with the above command.
>
> It wasn't really meant to solve anything, just split the commands up a touch
> more so we could see what's going on.
>
> I just installed Debian's toolchain, which is LLVM 9. IIRC that's pretty
> ancient WRT RISC-V, so my guess is that it's just a long way from building
> Linux. Looks like llvm-nm on my system doesn't put the @@LINUX_4.15 after the
> symbol names, I think this should do it?
>
> diff --git a/arch/riscv/kernel/vdso/so2s.sh b/arch/riscv/kernel/vdso/so2s.sh
> index 3c5b43207658..e64cb6d9440e 100755
> --- a/arch/riscv/kernel/vdso/so2s.sh
> +++ b/arch/riscv/kernel/vdso/so2s.sh
> @@ -2,5 +2,5 @@
> # SPDX-License-Identifier: GPL-2.0+
> # Copyright 2020 Palmer Dabbelt <palmerdabbelt@xxxxxxxxxx>
>
> -sed 's!\([0-9a-f]*\) T \([a-z0-9_]*\)@@LINUX_4.15!.global \2\n.set \2,0x\1!' \
> +sed 's!\([0-9a-f]*\) T \([a-z0-9_]*\)\(@@LINUX_4.15\)*!.global \2\n.set \2,0x\1!' \
> | grep '^\.'
>
> It works for me, at least for this specific problem.
Same here. defconfig minus CONFIG_EFI builds and boots in QEMU with that
diff (plus LLVM_IAS=1 in the make command, I forgot that earlier).
Tested-by: Nathan Chancellor <natechancellor@xxxxxxxxx>
> $ llvm-nm-9 -D arch/riscv/kernel/vdso/vdso.so |
> ./arch/riscv/kernel/vdso/so2s.sh .global __vdso_clock_getres
> .set __vdso_clock_getres,0x00000000000009e0
> .global __vdso_clock_gettime
> .set __vdso_clock_gettime,0x000000000000080a
> .global __vdso_flush_icache
> .set __vdso_flush_icache,0x0000000000000a48
> .global __vdso_getcpu
> .set __vdso_getcpu,0x0000000000000a3c
> .global __vdso_gettimeofday
> .set __vdso_gettimeofday,0x0000000000000916
> .global __vdso_rt_sigreturn
> .set __vdso_rt_sigreturn,0x0000000000000800
> $ nm -D arch/riscv/kernel/vdso/vdso.so | ./arch/riscv/kernel/vdso/so2s.sh
> .global __vdso_clock_getres
> .set __vdso_clock_getres,0x00000000000009e0
> .global __vdso_clock_gettime
> .set __vdso_clock_gettime,0x000000000000080a
> .global __vdso_flush_icache
> .set __vdso_flush_icache,0x0000000000000a48
> .global __vdso_getcpu
> .set __vdso_getcpu,0x0000000000000a3c
> .global __vdso_gettimeofday
> .set __vdso_gettimeofday,0x0000000000000916
> .global __vdso_rt_sigreturn
> .set __vdso_rt_sigreturn,0x0000000000000800
>
> >
> > Cheers,
> > Nathan
