Skip to contentSkip to conversation

kvmtool: new package

PATCH Superseded gustavo.zacarias at free-electrons.com 2 participants 3 messages Jun 22, 2015 – Jun 24, 2015

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 kvmtool: new package

From: Gustavo Zacarias <[email protected]>

---
 package/Config.in          |  1 +
 package/kvmtool/Config.in  | 32 ++++++++++++++++++++++++++++
 package/kvmtool/kvmtool.mk | 53 ++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 86 insertions(+)
 create mode 100644 package/kvmtool/Config.in
 create mode 100644 package/kvmtool/kvmtool.mk
package/Config.in +1 -0
index 6dbc32d..2c0d029 100644
@@ -1398,6 +1398,7 @@ endif
 	source "package/irqbalance/Config.in"
 	source "package/keyutils/Config.in"
 	source "package/kmod/Config.in"
+	source "package/kvmtool/Config.in"
 	source "package/lxc/Config.in"
 	source "package/monit/Config.in"
 	source "package/ncdu/Config.in"
package/kvmtool/Config.in new file +32 -0
new file mode 100644
index 0000000..ce24d40
@@ -0,0 +1,32 @@
+config BR2_PACKAGE_KVMTOOL
+	bool "kvmtool"
+	depends on BR2_USE_MMU # guest simple init fork()
+	# No MADV_HUGEPAGE define in uClibc(-ng)
+	depends on (BR2_TOOLCHAIN_USES_GLIBC || BR2_TOOLCHAIN_USES_MUSL)
+	# Only mips big-endian seems to be supported (build breaks)
+	# Should work on powerpc64 but the build breaks with missing types
+	depends on \
+		BR2_aarch64 || BR2_arm       || BR2_armeb || BR2_i386 || \
+		BR2_mips    || BR2_x86_64
+	# libfdt is only required for DT architectures...
+	# ...but it's not able to build for static targets
+	depends on !(BR2_STATIC_LIBS && (BR2_aarch64 || BR2_arm || BR2_armeb))
+	select BR2_PACKAGE_DTC if BR2_aarch64 || BR2_arm || BR2_armeb
+	help
+	  kvmtool is a lightweight tool for hosting KVM guests.
+	  As a pure virtualization tool it only supports guests using
+	  the same architecture, though it supports running 32-bit guests
+	  on those 64-bit architectures that allow this.
+
+	  https://git.kernel.org/cgit/linux/kernel/git/will/kvmtool.git/tree/README
+
+comment "kvmtool needs a (e)glibc or musl toolchain"
+	depends on BR2_USE_MMU
+	depends on !(BR2_TOOLCHAIN_USES_GLIBC || BR2_TOOLCHAIN_USES_MUSL)
+	depends on BR2_i386 || BR2_mips || BR2_x86_64
+
+comment "kvmtool needs a (e)glibc or musl toolchain w/ dynamic library"
+	depends on BR2_USE_MMU
+        depends on !(BR2_TOOLCHAIN_USES_GLIBC || BR2_TOOLCHAIN_USES_MUSL) || \
+		BR2_STATIC_LIBS
+	depends on BR2_aarch64 || BR2_arm || BR2_armeb
package/kvmtool/kvmtool.mk new file +53 -0

53 lines not on this page · raw message

Signature · 1 line
2.1.4
#1 lore raw

PATCH v2 1/1 kvmtool: new package

Quoted 3 lines · gustavo.zacarias at free-electrons.com

From: Gustavo Zacarias <[email protected]>

Some minor whitespace questions below, but basically

Quoted 26 lines
---
 package/Config.in          |  1 +
 package/kvmtool/Config.in  | 32 ++++++++++++++++++++++++++++
 package/kvmtool/kvmtool.mk | 53 ++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 86 insertions(+)
 create mode 100644 package/kvmtool/Config.in
 create mode 100644 package/kvmtool/kvmtool.mk
package/Config.in +1 -0
index 6dbc32d..2c0d029 100644
@@ -1398,6 +1398,7 @@ endif
 	source "package/irqbalance/Config.in"
 	source "package/keyutils/Config.in"
 	source "package/kmod/Config.in"
+	source "package/kvmtool/Config.in"
 	source "package/lxc/Config.in"
 	source "package/monit/Config.in"
 	source "package/ncdu/Config.in"
package/kvmtool/Config.in new file +0 -0
new file mode 100644
index 0000000..ce24d40

package/kvmtool/Config.in line 10

Quoted 10 lines
package/kvmtool/Config.in new file +10 -0
@@ -0,0 +1,32 @@
+config BR2_PACKAGE_KVMTOOL
+	bool "kvmtool"
+	depends on BR2_USE_MMU # guest simple init fork()
+	# No MADV_HUGEPAGE define in uClibc(-ng)
+	depends on (BR2_TOOLCHAIN_USES_GLIBC || BR2_TOOLCHAIN_USES_MUSL)
+	# Only mips big-endian seems to be supported (build breaks)
+	# Should work on powerpc64 but the build breaks with missing types
+	depends on \
+		BR2_aarch64 || BR2_arm       || BR2_armeb || BR2_i386 || \
+		BR2_mips    || BR2_x86_64

Line splitting, whitespace and ordering is a bit weird I think, why not

	depends on BR2_aarch64 || BR2_arm || BR2_armeb || BR2_mips || \
		BR2_i386 || BR2_x86_64
Quoted 6 lines
package/kvmtool/Config.in new file +6 -0
+	# libfdt is only required for DT architectures...
+	# ...but it's not able to build for static targets
+	depends on !(BR2_STATIC_LIBS && (BR2_aarch64 || BR2_arm || BR2_armeb))
+	select BR2_PACKAGE_DTC if BR2_aarch64 || BR2_arm || BR2_armeb
+	help
+	  kvmtool is a lightweight tool for hosting KVM guests.

package/kvmtool/Config.in line 26

Quoted 10 lines
package/kvmtool/Config.in new file +10 -0
+	  As a pure virtualization tool it only supports guests using
+	  the same architecture, though it supports running 32-bit guests
+	  on those 64-bit architectures that allow this.
+
+	  https://git.kernel.org/cgit/linux/kernel/git/will/kvmtool.git/tree/README
+
+comment "kvmtool needs a (e)glibc or musl toolchain"
+	depends on BR2_USE_MMU
+	depends on !(BR2_TOOLCHAIN_USES_GLIBC || BR2_TOOLCHAIN_USES_MUSL)
+	depends on BR2_i386 || BR2_mips || BR2_x86_64

It makes more sense to put the arch-depends together, so the arches just after MMU.

package/kvmtool/Config.in line 30

Quoted 4 lines
package/kvmtool/Config.in new file +4 -0
+
+comment "kvmtool needs a (e)glibc or musl toolchain w/ dynamic library"
+	depends on BR2_USE_MMU
+        depends on !(BR2_TOOLCHAIN_USES_GLIBC || BR2_TOOLCHAIN_USES_MUSL) || \

^^^^^^^^ spaces iso. tab.

Also, glibc can't be static, but probably the way you have it here is clearer.

package/kvmtool/Config.in line 32

Quoted 2 lines
package/kvmtool/Config.in new file +2 -0
+		BR2_STATIC_LIBS
+	depends on BR2_aarch64 || BR2_arm || BR2_armeb

Same here: arches just after MMU.

Quoted 38 lines
package/kvmtool/kvmtool.mk new file +32 -0
new file mode 100644
index 0000000..7f682db
@@ -0,0 +1,53 @@
+################################################################################
+#
+# kvmtool
+#
+################################################################################
+
+KVMTOOL_VERSION = 372f583d359a5bdcbbe7268809c8d1dc179c64d2
+KVMTOOL_SITE = $(BR2_KERNEL_MIRROR)/scm/linux/kernel/git/will/kvmtool.git
+KVMTOOL_SITE_METHOD = git
+KVMTOOL_DEPENDENCIES = \
+	$(if $(BR2_PACKAGE_BINUTILS),binutils) \
+	$(if $(BR2_PACKAGE_DTC),dtc) \
+	$(if $(BR2_PACKAGE_LIBAIO),libaio) \
+	$(if $(BR2_PACKAGE_LIBGTK3),libgtk3) \
+	$(if $(BR2_PACKAGE_LIBVNCSERVER),libvncserver) \
+	$(if $(BR2_PACKAGE_SDL),sdl) \
+	$(if $(BR2_PACKAGE_ZLIB),zlib)
+KVMTOOL_LICENSE = GPLv2
+KVMTOOL_LICENSE_FILES = COPYING
+
+# This is required to convert a static binary (init helper) back into
+# object-file format, and in multilib toolchains like CS 2012.09 for x86
+# the default is i386, hence when building for x86_64 things break since
+# LD doesn't autodetect the input file format.
+# GCC-as-linker can't accomplish this feat easily either since it's mixing
+# static content (guest_init.o) with dynamic one (lkvm) making
+# a relocatable output file.
+# The purpose of this trick is to embed the init helper into the main
+# binary to help users in guest system startup, which would otherwise
+# require more complex guest images.
+# This needs revisiting if/when X32 ABI support is added.
+#

package/kvmtool/kvmtool.mk line 42

Quoted 10 lines
package/kvmtool/kvmtool.mk new file +10 -0
+# If more packages need this (unlikely) an ld wrapper might be a better
+# solution, using gcc -dumpspecs information.
+KVMTOOL_EXTRA_LDFLAGS = \
+	$(if $(BR2_x86_64),-m elf_x86_64)
+
+# Disable -Werror, otherwise musl is not happy
+KVMTOOL_MAKE_OPTS = \
+	CROSS_COMPILE="$(TARGET_CROSS)" \
+	LDFLAGS="$(TARGET_LDFLAGS) $(KVMTOOL_EXTRA_LDFLAGS)" \
+	WERROR=0

Shouldn't we pass TARGET_CFLAGS and friends as well? Perhaps put
TARGET_CONFIGURE_OPTS in the environment?

Regards,
Arnout

Quoted 12 lines
package/kvmtool/kvmtool.mk new file +11 -0
+
+define KVMTOOL_BUILD_CMDS
+	$(TARGET_MAKE_ENV) ARCH=$(KERNEL_ARCH) $(MAKE) -C $(@D) $(KVMTOOL_MAKE_OPTS)
+endef
+
+define KVMTOOL_INSTALL_TARGET_CMDS
+	$(TARGET_MAKE_ENV) ARCH=$(KERNEL_ARCH) $(MAKE) -C $(@D) \
+		$(KVMTOOL_MAKE_OPTS) install DESTDIR=$(TARGET_DIR) prefix=/usr
+endef
+
+$(eval $(generic-package))
Signature · 6 lines
Arnout Vandecappelle                          arnout at mind be
Senior Embedded Software Architect            +32-16-286500
Essensium/Mind                                http://www.mind.be
G.Geenslaan 9, 3001 Leuven, Belgium           BE 872 984 063 RPR Leuven
LinkedIn profile: http://www.linkedin.com/in/arnoutvandecappelle
GPG fingerprint:  7CB5 E4CC 6C2E EFD4 6E3D A754 F963 ECAB 2450 2F1F
#2 lore raw

PATCH v2 1/1 kvmtool: new package

Quoted 8 lines · Arnout Vandecappelle

package/kvmtool/Config.in line 10

Quoted 3 lines
package/kvmtool/Config.in new file +3 -0
+	depends on \
+		BR2_aarch64 || BR2_arm       || BR2_armeb || BR2_i386 || \
+		BR2_mips    || BR2_x86_64

Line splitting, whitespace and ordering is a bit weird I think, why not

	depends on BR2_aarch64 || BR2_arm || BR2_armeb || BR2_mips || \
		BR2_i386 || BR2_x86_64

Hi Arnout.
At first the architecture list was longer but i started chopping off because of build failures for architectures that weren't my target and this particular piece of software doesn't work in general with qemu (since it normally doesn't emulate kvm features of the processors, so no kvm-in-kvm facility to test with).
The weird indentation was from some other $random package, though i don't recall which one it was.

Quoted 6 lines

package/kvmtool/Config.in line 26

Quoted 4 lines
package/kvmtool/Config.in new file +4 -0
+comment "kvmtool needs a (e)glibc or musl toolchain"
+	depends on BR2_USE_MMU
+	depends on !(BR2_TOOLCHAIN_USES_GLIBC || BR2_TOOLCHAIN_USES_MUSL)
+	depends on BR2_i386 || BR2_mips || BR2_x86_64

It makes more sense to put the arch-depends together, so the arches just after MMU.

Ok.

Quoted 6 lines

package/kvmtool/Config.in line 30

Quoted 3 lines
package/kvmtool/Config.in new file +3 -0
+comment "kvmtool needs a (e)glibc or musl toolchain w/ dynamic library"
+	depends on BR2_USE_MMU
+        depends on !(BR2_TOOLCHAIN_USES_GLIBC || BR2_TOOLCHAIN_USES_MUSL) || \
    ^^^^^^^^ spaces iso. tab.

Also, glibc can't be static, but probably the way you have it here is clearer.

Yes, i tried to minimize the amount of comments and conditionals here.

Quoted 4 lines

package/kvmtool/Config.in line 32

Quoted 2 lines
package/kvmtool/Config.in new file +2 -0
+		BR2_STATIC_LIBS
+	depends on BR2_aarch64 || BR2_arm || BR2_armeb

Same here: arches just after MMU.

Ok.

Quoted 8 lines

package/kvmtool/kvmtool.mk line 42

Quoted 5 lines
package/kvmtool/kvmtool.mk new file +5 -0
+# Disable -Werror, otherwise musl is not happy
+KVMTOOL_MAKE_OPTS = \
+	CROSS_COMPILE="$(TARGET_CROSS)" \
+	LDFLAGS="$(TARGET_LDFLAGS) $(KVMTOOL_EXTRA_LDFLAGS)" \
+	WERROR=0

Shouldn't we pass TARGET_CFLAGS and friends as well? Perhaps put
TARGET_CONFIGURE_OPTS in the environment?

Depending on the architecture this package changes target ABI and even builds a BIOS stub for x86/x86_64 (changing for instance cpu target), forcing CFLAGS/LDFLAGS on it is a bad idea in general which causes build failures for certain combinations, at least for x86*.
It's by far the sickest package i've met regarding toolchain tricks, with the static-in-dynamic init helper converted via ld as an example. Regards.