Genode application binary interface (ABI)

This patch decouples the kernel-specific implementation of the dynamic
linker from its kernel-agnostic binary interface. The name of the
kernel-specific dynamic linker binary now corresponds to the kernel,
e.g., 'ld-linux.lib.so' or 'ld-nova.lib.so'. Applications are no longer
linked directly against a concrete instance of the dynamic linker but
against a shallow stub called 'ld.lib.so'. This stub contains nothing
but the symbols provided by the dynamic linker. It thereby represents
the Genode ABI.

At system-integration time, the kernel-specific run/boot_dir back ends
integrate the matching the kernel-specific variant of the dynamic linker
as 'ld.lib.so' into the boot image.

The ABI symbol file for the dynamic linker is located at
'base/lib/symbols/ld'. It contains the joint ABI of all supported
architectures. The new utility 'tool/abi_symbols' eases the creation of
such an ABI symbol file for a given shared library. Its result should be
manually inspected and edited as needed.

The patch removes the 'syscall' library from 'base_libs.mk' to avoid
polluting the kernel-agnostic ABI with kernel-specific interfaces.

Issue #2190
Issue #2195

tool/abi_symbols

@@ -0,0 +1,98 @@
+#!/usr/bin/tclsh
+
+#
+# \brief  Generate ABI symbol list for a shared library
+# \author Norman Feske
+# \date   2016-12-07
+#
+# The tool takes the shared library as argument and writes the output to
+# standard output. Each line of the resulting output contains the symbol name,
+# type, and size
+#
+
+# obtain symbol information via 'nm'
+set symbols [exec nm --format posix --dynamic $argv]
+
+# obtain demangled input via c++filt
+set demangled [exec c++filt << $symbols]
+
+set demangled_lines [split $demangled "\n"]
+
+set i 0
+set output_lines {}
+foreach line [split $symbols "\n"] {
+
+	set size_hex 0
+	set type "U"
+
+	# match undefined symbol
+	regexp {^(\w+) U\s*$} $line dummy name
+
+	# match defined symbol, which does not always feature a size value
+	regexp {^(\w+) (\w) \w+ ?(\w*)$} $line dummy name type size_hex
+
+	# decimal symbol size
+	set size_dec [expr 0x0$size_hex]
+
+	set demangled_name $name
+	regexp {^(.+) \w \w+( \w+)?$} [lindex $demangled_lines $i] dummy demangled_name
+
+	set keep 1
+
+	#
+	# Ignore undefined symbols
+	#
+	if {$type == "U"} { set keep 0 }
+
+	#
+	# Check if the unmangled line contains a template-argument delimiter ('<')
+	# prior the first opening parenthesis. This way, template methods match but
+	# signatures of non-template functions/methods that take templated
+	# arguments won't. We can discard symbols for the former but not the
+	# latter.
+	#
+	if {[regexp {^[^(]+<} $demangled_name dummy]} { set keep 0 }
+
+	#
+	# Drop weak vtable and typeinfo symbols
+	#
+	# Those symbols are solely used to merge vtables between the executable and
+	# shared library. When linking object files that use the same types, the
+	# merging saves a lot space. However, the benefit at the coarse granularity
+	# of shared libraries is rather small compared to the huge inflation of the
+	# ABI size caused by such symbols.
+	#
+	if {($type == "V") && ([regexp {^(typeinfo |vtable )} $demangled_name dummy])} {
+		set keep 0 }
+
+	#
+	# Drop weak C++ symbols
+	#
+	# In contrast to weak C symbols, which are rarely and always deliberately
+	# created, weak C++ symbols are implicitly created by the compiler for
+	# inline functions.
+	#
+	if {[regexp {W|V} $type] && ($name != $demangled_name)} {
+		set keep 0 }
+
+	# write result
+	if {$keep} {
+
+		#
+		# Annotate the size for global data symbols where the size is needed to
+		# create copy relations (ARM EABI) in the binary that is linked against
+		# the shared library. For each global read-only data symbol found in
+		# the shared object, the linker reserves space (according to the
+		# symbol's size) in the binary's BSS segment. At runtime, the dynamic
+		# linker copies the data from the shared library's symbol into the
+		# binary's BSS.
+		#
+		if {$type == "D"} {
+			puts "$name $type $size_dec"
+		} else {
+			puts "$name $type"
+		}
+	}
+	incr i
+}
+

tool/run/boot_dir/fiasco

@@ -1,3 +1,6 @@
+proc binary_name_ld_lib_so { } { return "ld-fiasco.lib.so" }
+
+
 ##
 # Read the location of the Fiasco user directory from 'etc/fiasco.conf'
 #
@@ -23,15 +26,6 @@ proc l4_dir { } {
 }
 
 
-##
-# Return whether the l4-buid-directory is provided from the outside
-#
-proc l4_dir_external { } {
-	if {[l4_dir] == "[pwd]/l4"} { return 0 }
-	return 1
-}
-
-
 ##
 # Return the location of the Fiasco kernel
 #
@@ -40,15 +34,6 @@ proc fiasco { } {
 }
 
 
-##
-# Return whether fiasco kernel is provided from the outside
-#
-proc fiasco_external { } {
-	if {[fiasco] == "[pwd]/kernel/fiasco/fiasco"} { return 0 }
-	return 1
-}
-
-
 proc bin_dir { } {
 	if {[have_spec x86_32]}  { return "[l4_dir]/bin/x86_586" }
 
@@ -71,15 +56,14 @@ proc core_link_address { } { return "0x01000000" }
 #
 proc run_boot_dir {binaries} {
 
+	build { lib/ld/fiasco kernel bootstrap sigma0 }
+
 	build_core_image $binaries
 
 	global fiasco_serial_esc_arg
 
 	exec mkdir -p [run_dir]/fiasco
 
-	if {![fiasco_external]} { build { kernel } }
-	if {![l4_dir_external]} { build { bootstrap sigma0 } }
-
 	# assert existence of the L4 build directory
 	l4_dir
 

tool/run/boot_dir/foc

@@ -6,6 +6,9 @@
 # This file is meant to be used as '--include' argument for 'tool/run'.
 #
 
+proc binary_name_ld_lib_so { } { return "ld-foc.lib.so" }
+
+
 ##
 # Return the location of the Fiasco.OC user directory
 #
@@ -102,11 +105,12 @@ proc core_link_address { } {
 }
 
 proc run_boot_dir_x86 {binaries} {
+
 	global fiasco_serial_esc_arg
 
 	exec mkdir -p [run_dir]/fiasco
 
-	set foc_targets { }
+	set foc_targets { lib/ld/foc }
 	if {![fiasco_external] && ![file exists kernel]} { lappend foc_targets kernel }
 	if {![l4_dir_external]} {
 		if {![file exists bootstrap]} { lappend foc_targets bootstrap }
@@ -114,6 +118,8 @@ proc run_boot_dir_x86 {binaries} {
 	}
 	if {[llength $foc_targets] > 0} { build $foc_targets }
 
+	build_core_image $binaries
+
 	# assert existence of the L4 build directory
 	l4_dir
 
@@ -185,7 +191,9 @@ proc run_boot_dir_arm {binaries} {
 	global run_target
 	global fiasco_serial_esc_arg
 
-	build "kernel sigma0 bootstrap"
+	build { lib/ld/foc kernel sigma0 bootstrap }
+
+	build_core_image $binaries
 
 	#
 	# Generate bootstrap config
@@ -239,7 +247,6 @@ proc run_boot_string { } {
 # Populate boot directory with binaries on fiasco.OC
 #
 proc run_boot_dir {binaries} {
-	build_core_image $binaries
 
 	if {[have_spec x86]} { return [run_boot_dir_x86 $binaries] }
 	if {[have_spec arm]} { return [run_boot_dir_arm $binaries] }

tool/run/boot_dir/hw

@@ -1,6 +1,7 @@
-proc run_boot_string { } {
-	return "\nkernel initialized"
-}
+proc binary_name_ld_lib_so { } { return "ld-hw.lib.so" }
+
+
+proc run_boot_string { } { return "\nkernel initialized" }
 
 
 proc core_link_address { } {
@@ -33,6 +34,8 @@ proc run_boot_dir {binaries} {
 		close $fh
 	}
 
+	build { lib/ld/hw }
+
 	build_core_image $binaries
 
 	if {[have_include "image/iso"] || [have_include "image/disk"]} {

tool/run/boot_dir/linux

@@ -1,7 +1,14 @@
+proc binary_name_ld_lib_so { } { return "ld-linux.lib.so" }
+
+
 ##
 # Populate boot directory with binaries on Linux
 #
 proc run_boot_dir {binaries} {
+
+	build { lib/ld/linux }
+
 	foreach binary $binaries {
-		exec ln -sf ../../../../bin/$binary [run_dir]/genode }
+		set src_binary_path "../../../../bin/[kernel_specific_binary $binary]"
+		exec ln -sf $src_binary_path [run_dir]/genode/$binary }
 }

tool/run/boot_dir/nova

@@ -1,3 +1,6 @@
+proc binary_name_ld_lib_so { } { return "ld-nova.lib.so" }
+
+
 ##
 # Read the location of the NOVA kernel directory from 'etc/nova.conf'
 #
@@ -43,13 +46,17 @@ proc core_ld_opts { } {
 #
 proc run_boot_dir {binaries} {
 
+	if {![nova_external]} {
+		build { lib/ld/nova kernel }
+	} else {
+		build { lib/ld/nova }
+	}
+
 	#
 	# Collect contents of the ISO image
 	#
 	build_core_image $binaries
 
-	if {![nova_external]} { build { kernel } }
-
 	puts "using NOVA kernel at [nova_kernel]"
 	exec [cross_dev_prefix]objcopy -O elf32-i386 [nova_kernel] [run_dir]/hypervisor
 	exec [cross_dev_prefix]strip [run_dir]/hypervisor

tool/run/boot_dir/okl4

@@ -1,3 +1,6 @@
+proc binary_name_ld_lib_so { } { return "ld-okl4.lib.so" }
+
+
 ##
 # Get the base-okl4 repository
 #
@@ -113,10 +116,8 @@ proc core_link_address { } { return "0x03000000" }
 # Populate directory with binaries on OKL4
 #
 proc run_boot_dir {binaries} {
-	global weaver_xml_template
 
-	build_core_image $binaries
-	exec mv [run_dir]/image.elf [run_dir].image
+	global weaver_xml_template
 
 	#
 	# Build kernel if needed
@@ -127,7 +128,14 @@ proc run_boot_dir {binaries} {
 	# the kernel does not stand in the way of the everyday's work flow of
 	# executing run scripts as quick as possible.
 	#
-	if {![okl4_external] && ![file exists [okl4]]} { build { kernel } }
+	if {![okl4_external] && ![file exists [okl4]]} {
+		build { lib/ld/okl4 kernel }
+	} else {
+		build { lib/ld/okl4 }
+	}
+
+	build_core_image $binaries
+	exec mv [run_dir]/image.elf [run_dir].image
 
 	exec cp [okl4] [run_dir]/kernel
 

tool/run/boot_dir/pistachio

@@ -6,6 +6,8 @@
 # This file is meant to be used as '--include' argument for 'tool/run'.
 #
 
+proc binary_name_ld_lib_so { } { return "ld-pistachio.lib.so" }
+
 
 ##
 # Install files needed to boot via PXE
@@ -74,12 +76,16 @@ proc core_link_address { } { return "0x02000000" }
 #
 proc run_boot_dir {binaries} {
 
+	if {![kernel_external] && ![file exists [pistachio_kernel]]} {
+		build { lib/ld/pistachio kernel }
+	} else {
+		build { lib/ld/pistachio }
+	}
+
 	build_core_image $binaries
 
 	exec mkdir -p [run_dir]/pistachio
 
-	if {![kernel_external] && ![file exists [pistachio_kernel]]} { build { kernel } }
-
 	exec cp [pistachio_kernel] [run_dir]/pistachio/kernel
 	exec cp [pistachio_user_dir]/serv/sigma0/sigma0 [run_dir]/pistachio
 	exec cp [pistachio_user_dir]/util/kickstart/kickstart [run_dir]/pistachio

tool/run/boot_dir/sel4

@@ -1,6 +1,5 @@
-#
-# Based on boot_dir/hw
-#
+proc binary_name_ld_lib_so { } { return "ld-sel4.lib.so" }
+
 
 proc run_boot_string { } { return "\n\rStarting node #0" }
 
@@ -14,7 +13,7 @@ proc core_link_address { } { return "0x02000000" }
 proc run_boot_dir {binaries} {
 
 	# build sel4 kernel
-	build { kernel }
+	build { lib/ld/sel4 kernel }
 	exec cp bin/sel4 [run_dir]/sel4
 
 	build_core_image $binaries

tool/run/run

@@ -438,6 +438,33 @@ proc kernel_location_from_config_file { config_file default_location } {
 }
 
 
+##
+# Return name of kernel-specific binary for a given generic name
+#
+# The boot_dir plugin may provide functions named 'binary_name_<binary>'
+# where '<binary>' stands for a generic name like "timer" or "nic". The
+# function returns the name of the binary to integrate into the boot
+# directory under the name '<binary>'.
+#
+# If no such function exists, it returns the argument as is. This is the
+# case for regular binaries that appear in the boot directory under their
+# original name.
+#
+proc kernel_specific_binary { binary } {
+
+	regsub -all {\.} $binary "_" function_suffix
+	set function_name "binary_name_$function_suffix"
+
+	if {[info procs $function_name] == $function_name} {
+		set binary_name [$function_name]
+		puts "using '$binary_name' as '$binary'"
+		return [$function_name]
+	}
+
+	return $binary
+}
+
+
 ##
 # Copy the specified binaries from the 'bin/' directory to the run
 # directory and try to strip executables.
@@ -445,7 +472,7 @@ proc kernel_location_from_config_file { config_file default_location } {
 proc copy_and_strip_genode_binaries_to_run_dir { binaries } {
 
 	foreach binary $binaries {
-		exec cp bin/$binary [run_dir]/genode
+		exec cp bin/[kernel_specific_binary $binary] [run_dir]/genode/$binary
 		catch {
 			exec [cross_dev_prefix]strip [run_dir]/genode/$binary || true }
 	}