Date:   Mon, 14 Sep 2020 14:09:38 +0100
From:   Will Deacon <will@...nel.org>
To:     David Brazdil <dbrazdil@...gle.com>
Cc:     Marc Zyngier <maz@...nel.org>,
        Catalin Marinas <catalin.marinas@....com>,
        Dennis Zhou <dennis@...nel.org>, Tejun Heo <tj@...nel.org>,
        Christoph Lameter <cl@...ux.com>,
        Arnd Bergmann <arnd@...db.de>,
        James Morse <james.morse@....com>,
        Julien Thierry <julien.thierry.kdev@...il.com>,
        Suzuki K Poulose <suzuki.poulose@....com>,
        linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
        kvmarm@...ts.cs.columbia.edu, linux-arch@...r.kernel.org,
        kernel-team@...roid.com
Subject: Re: [PATCH v2 02/10] kvm: arm64: Partially link nVHE hyp code,
 simplify HYPCOPY

On Thu, Sep 03, 2020 at 11:17:04AM +0200, David Brazdil wrote:
> Previous series introduced custom build rules for nVHE hyp code, using
> objcopy to prefix ELF section and symbol names to separate nVHE code
> into its own "namespace". This approach was limited by the expressiveness
> of objcopy's command line interface, eg. missing support for wildcards.

nit: "Previous series" isn't a lot of use here or in the git log. You can
just say something like:

  "Relying on objcopy to prefix the ELF section names of the nVHE hyp code
   is brittle and prevents us from using wildcards to match specific
   section names."

and then go on to explain what the change is doing (see
Documentation/process/submitting-patches.rst for more help here)

Also, given that this is independent of the other patches, please can you
move it right to the start of the series? I'm a bit worried about the
potential for regressions given the changes to the way in which we link,
so the sooner we can get this patch some more exposure, the better.

> Improve the build rules by partially linking all '.hyp.o' files and
> prefixing their ELF section names using a linker script. Continue using
> objcopy for prefixing ELF symbol names.
> 
> One immediate advantage of this approach is that all subsections
> matching a pattern can be merged into a single prefixed section, eg.
> .text and .text.* can be linked into a single '.hyp.text'. This removes
> the need for -fno-reorder-functions on GCC and will be useful in the
> future too: LTO builds use .text subsections, compilers routinely
> generate .rodata subsections, etc.
> 
> Partially linking all hyp code into a single object file also makes it
> easier to analyze.
> 
> Signed-off-by: David Brazdil <dbrazdil@...gle.com>
> ---
>  arch/arm64/kvm/hyp/nvhe/Makefile  | 56 ++++++++++++++++---------------
>  arch/arm64/kvm/hyp/nvhe/hyp.lds.S | 14 ++++++++
>  2 files changed, 43 insertions(+), 27 deletions(-)
>  create mode 100644 arch/arm64/kvm/hyp/nvhe/hyp.lds.S
> 
> diff --git a/arch/arm64/kvm/hyp/nvhe/Makefile b/arch/arm64/kvm/hyp/nvhe/Makefile
> index aef76487edc2..1b2fbb19f3e8 100644
> --- a/arch/arm64/kvm/hyp/nvhe/Makefile
> +++ b/arch/arm64/kvm/hyp/nvhe/Makefile
> @@ -10,40 +10,42 @@ obj-y := timer-sr.o sysreg-sr.o debug-sr.o switch.o tlb.o hyp-init.o
>  obj-y += ../vgic-v3-sr.o ../aarch32.o ../vgic-v2-cpuif-proxy.o ../entry.o \
>  	 ../fpsimd.o ../hyp-entry.o
>  
> -obj-y := $(patsubst %.o,%.hyp.o,$(obj-y))
> -extra-y := $(patsubst %.hyp.o,%.hyp.tmp.o,$(obj-y))
> +##
> +## Build rules for compiling nVHE hyp code
> +## Output of this folder is `hyp.o`, a partially linked object file containing
> +## all nVHE hyp code and data.
> +##
>  
> -$(obj)/%.hyp.tmp.o: $(src)/%.c FORCE
> +hyp-obj := $(patsubst %.o,%.hyp.o,$(obj-y))
> +obj-y := hyp.o
> +extra-y := $(hyp-obj) hyp.tmp.o hyp.lds
> +
> +# 1) Compile all source files to `.hyp.o` object files. The file extension
> +#    avoids file name clashes for files shared with VHE.
> +$(obj)/%.hyp.o: $(src)/%.c FORCE
>  	$(call if_changed_rule,cc_o_c)
> -$(obj)/%.hyp.tmp.o: $(src)/%.S FORCE
> +$(obj)/%.hyp.o: $(src)/%.S FORCE
>  	$(call if_changed_rule,as_o_S)
> -$(obj)/%.hyp.o: $(obj)/%.hyp.tmp.o FORCE
> -	$(call if_changed,hypcopy)
>  
> -# Disable reordering functions by GCC (enabled at -O2).
> -# This pass puts functions into '.text.*' sections to aid the linker
> -# in optimizing ELF layout. See HYPCOPY comment below for more info.
> -ccflags-y += $(call cc-option,-fno-reorder-functions)
> +# 2) Compile linker script.
> +$(obj)/hyp.lds: $(src)/hyp.lds.S FORCE
> +	$(call if_changed_dep,cpp_lds_S)

Why is it not sufficient just to list the linker script as a target, like
we do for vmlinux.lds in extra-y?

> +# 3) Partially link all '.hyp.o' files and apply the linker script.
> +#    Prefixes names of ELF sections with '.hyp', eg. '.hyp.text'.
> +LDFLAGS_hyp.tmp.o := -r -T $(obj)/hyp.lds
> +$(obj)/hyp.tmp.o: $(addprefix $(obj)/,$(hyp-obj)) $(obj)/hyp.lds FORCE
> +	$(call if_changed,ld)
> +
> +# 4) Produce the final 'hyp.o', ready to be linked into 'vmlinux'.
> +#    Prefixes names of ELF symbols with '__kvm_nvhe_'.
> +$(obj)/hyp.o: $(obj)/hyp.tmp.o FORCE
> +	$(call if_changed,hypcopy)
>  
>  # The HYPCOPY command uses `objcopy` to prefix all ELF symbol names
> -# and relevant ELF section names to avoid clashes with VHE code/data.
> -#
> -# Hyp code is assumed to be in the '.text' section of the input object
> -# files (with the exception of specialized sections such as
> -# '.hyp.idmap.text'). This assumption may be broken by a compiler that
> -# divides code into sections like '.text.unlikely' so as to optimize
> -# ELF layout. HYPCOPY checks that no such sections exist in the input
> -# using `objdump`, otherwise they would be linked together with other
> -# kernel code and not memory-mapped correctly at runtime.
> +# to avoid clashes with VHE code/data.
>  quiet_cmd_hypcopy = HYPCOPY $@
> -      cmd_hypcopy =							\
> -	if $(OBJDUMP) -h $< | grep -F '.text.'; then			\
> -		echo "$@: function reordering not supported in nVHE hyp code" >&2; \
> -		/bin/false;						\
> -	fi;								\
> -	$(OBJCOPY) --prefix-symbols=__kvm_nvhe_				\
> -		   --rename-section=.text=.hyp.text			\
> -		   $< $@
> +      cmd_hypcopy = $(OBJCOPY) --prefix-symbols=__kvm_nvhe_ $< $@
>  
>  # Remove ftrace and Shadow Call Stack CFLAGS.
>  # This is equivalent to the 'notrace' and '__noscs' annotations.
> diff --git a/arch/arm64/kvm/hyp/nvhe/hyp.lds.S b/arch/arm64/kvm/hyp/nvhe/hyp.lds.S
> new file mode 100644
> index 000000000000..aaa0ce133a32
> --- /dev/null
> +++ b/arch/arm64/kvm/hyp/nvhe/hyp.lds.S
> @@ -0,0 +1,14 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +/*
> + * Linker script used during partial linking of nVHE EL2 object files.
> + * Written by David Brazdil <dbrazdil@...gle.com>
> + */
> +
> +/*
> + * Defines an ELF hyp section from input section @NAME and its subsections.
> + */
> +#define HYP_SECTION(NAME) .hyp##NAME : { *(NAME NAME##.[0-9a-zA-Z_]*) }

Is 'NAME##.*' likely to cause a problem here?

Will

Powered by blists - more mailing lists