Since TUNE_FEATURES now either contains a CPU or an architecture (but
not both) we can't rely on finding the architecture in TUNE_FEATURES.
Use architecture specific over-rides instead.
(From OE-Core rev: 805dd4807d322dc70cef97edd68fdb3142b60fb1)
Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This allows us to generate a rootFS with a large filesystem for use with
QEMU.
(From OE-Core rev: e06439200e44999c1e2f88d7d6c651da13698ca7)
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Default riscv is little-endian moreover most of other arches define
bigendian as tune and treats absense as litteendian, this make risc-v
fall in line
(From OE-Core rev: cd6f377591a7bd7b3c61ce580f997aaeffab3df3)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This helps in defining LE tunes and at the same time specifies
endianness on compiler cmdline clearly, clang e.g. defaults to
little-endian always, so unless specified with -mbig-endian won't
compile the code right
(From OE-Core rev: e0fd699d398f0e88fb208970dea7b74e6e9431fe)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
There was a discussion about what amount of RAM is appropriate for a
default; the outcome was that for now it is still 256M. Some qemu machine
definitions have however set this to 512M so for the sake of
treating all architectures fairly, they are reset back to 256M.
Also runqemu is adjusted to use 256M if QB_MEM is not set at all.
http://lists.openembedded.org/pipermail/openembedded-core/2019-August/285900.html
(From OE-Core rev: 04c01b6cc5be3e6d45d0e04571640648a5655a8b)
Signed-off-by: Alexander Kanavin <alex.kanavin@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Configuration:
MACHINE = qemumips64
bitbake lib32-core-image-minimal
runqemu slirp nographic qemumips64 ext4
Error:
ERROR - Failed to run qemu: qemu-system-mips: unable to find CPU model 'MIPS64R2-generic'
Fixed by moving QB_SYSTEM_NAME to Respective configuration file
(From OE-Core rev: e724e8836ed614ff8eaa0d0d9c51d22ee62576b3)
Signed-off-by: Changqing Li <changqing.li@windriver.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Now that we have a -bios option for the RISC-V virt machine in QEMU we
can pass OpenSBI in via -bios and the kernel in via -kernel. We no
longer need to pass the kernel in via -device loader so let's remove
that.
(From OE-Core rev: 65e7f371f19e053d0bac7771a80615f6bada74c7)
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Configrations:
MACHINE: qemux86-64
require conf/multilib.conf
MULTILIBS = "multilib:lib32"
DEFAULTTUNE_virtclass-multilib-lib32 = "x86"
Reproduce steps:
bitbake lib32-core-image-minimal
runqemu qemux86-64 nographic lib32-core-image-minimal
Errors:
qemu cannot bootup since:
Booting from ROM...
This kernel requires an x86-64 CPU, but only detected an i686 CPU.
Unable to boot - please use a kernel appropriate for your CPU.
QEMU: Terminated
For lib32 image, override has x86, so the qemubin set to qemu-system-i386,
fix by move QB_SYSTEM_NAME to corresponding conf, don't use the override
(From OE-Core rev: ffaf86f175b2e6caa3a0067f7b3725930b053715)
Signed-off-by: Changqing Li <changqing.li@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The include is split ready to add the 32-bit RISC-V machine as soon as
glibc supports 32-bit RISC-V.
This is based on the work in the meta-riscv layer, thanks to Khem for
starting this.
(From OE-Core rev: 11b6020dff4550fc3a42e04bc1e86baf37942c62)
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The value of PACKAGE_EXTRA_ARCHS_tune-thunderx should be based on
PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-crypto instead of armv8a-crc-crypto.
Otherwise we would get some sanity check error like this:
OE-core's config sanity checker detected a potential misconfiguration.
Either fix the cause of this error or at your own risk disable the checker (see sanity.conf).
Following is the list of potential problems / advisories:
Error, the PACKAGE_ARCHS variable (all any noarch armv8a-crc-crypto thunderx qemuarm64) for DEFAULTTUNE (thunderx) does not contain TUNE_PKGARCH (aarch64)
(From OE-Core rev: 13cc0f7c0bd98ea228e9bdf51043117d38837ce7)
Signed-off-by: Kevin Hao <kexin.hao@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This matches what the qemux86_64 is currently using, and
will allow testing the instructions added in the meantime;
particularly various SSE extensions are now enabled.
(From OE-Core rev: f3b1e577ec94c849d0354f5679257f02ef4e4fe9)
Signed-off-by: Alexander Kanavin <alex.kanavin@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This makes sure, e.g., ${SOC_FAMILY} and ${MACHINE} have higher
priorities than aarch64.
(From OE-Core rev: 4d1339af88543d85930139dbcb87a669f285ea66)
Signed-off-by: Peter Kjellerstedt <peter.kjellerstedt@axis.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The armv8a tune specific PACKAGE_EXTRA_ARCHS contained tune feature
names like "crc" and "crypto" rather than package architecture names
like "armv8a-crc" and "armv8a-crypto".
(From OE-Core rev: 1756f2354745ee709886683422887efed4e10dba)
Signed-off-by: Peter Kjellerstedt <peter.kjellerstedt@axis.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
SIMD instructions are a mandatory part of armv8a
(they were optional in armv7a), and the gcc docs
also say that they are always enabled.
(From OE-Core rev: 02288c94e99e9dd444d8c1af186b6d89085b7b8b)
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
armv5 is not a specific tune feature anymore, there is no need to check
it, since having dsp will define if 'e' should be added or not
(From OE-Core rev: 1d6d5bb30a83f9136b7c33e297d48564ae61b50e)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Cc: Peter Kjellerstedt <peter.kjellerstedt@axis.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
For multilib to work correctly, BASE_LIB overrides must be provided.
(From OE-Core rev: b32ec63e48a3552f2e7f3cc5caf61432af716283)
Signed-off-by: teven Hung (洪于玉) <Steven.Hung@mediatek.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
For multilib to work correctly, BASE_LIB overrides must be provided for
each new tune added in this file.
(From OE-Core rev: e39c5ec90ebbc37064c9cd59eba12603317740cd)
Signed-off-by: Mike Crowe <mac@mcrowe.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Some ARM Cortex devices have the VFPv4-D16, but no NEON.
(From OE-Core rev: 594f8584268d5179c18512beada2bae4a21325de)
Signed-off-by: Phil Edworthy <phil.edworthy@renesas.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
I am not familiar with the big endian, so I don't update it.
I don't have much information about the Cavium ThunderX,
it looks like it supports all the ARM instructions.
(From OE-Core rev: b6c6fa72bdffd5d8995058e8e0e21d5561cc16c6)
Signed-off-by: Randy Li <ayaka@soulik.info>
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
There are some addtional instructions apart from bare armv8,
also there is armv8.1, armv8.2.
Most the processor would support crc, except X-gene 1.
(From OE-Core rev: d1db78b0e284e1a1f370e71183ded0cbdc1475db)
Signed-off-by: Randy Li <ayaka@soulik.info>
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
tune files which inherit the arch definitions already define appropriate
-mcpu option, which is equivalent of right -march and -mtune combination
and is preferred since gcc is getting stricter and stricter with option
check semantics and can now find incompatible -march and -mcpu options
better with every release. It does internal feature consistency check
and if it finds out discrepency between what -mcpu would expand to as
compared to -march it will flag the options to be incompatible, for
naked eye it sounds wrong but gcc would translate -mcpu to a given
-march internally and it might not match to what we set in these arch
files.
The effects are quite subtle, where this can result in configure test
failing to compile due to these incompatible options and a feature
option getting disabled for a recipe for no reason.
e.g. with gcc9 which can now detect that -mcpu=cortex-a5 and
-march=armv7-a are incompatible, many features in libstdc++ ends up
disabled due to configure check failures e.g. size_t size, ptrdiff_t
sizes, which inturn results in compiling libstdc++ with unwanted
disabled features.
(From OE-Core rev: ac83d22eb5031f7fdd09d34a1a46d92fd3e39a3c)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
-march=armv5't'e means that CPU can execute thumb ISA, we do not need to
tie this to exclusively generating thumb ISA, this change means that
when we have thumb in tune features then it can use 't' in -march
options irrespective of ISA being thumb or arm.
This fixes derivative of armv5 tunes and paves way for gcc9 where e.g.
armv5e is dropped and minimum arch supported is armv5te
(From OE-Core rev: 49bc3eab2897307a92b538a045e3d4dce245b0e0)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Historically first PowerPC ABI was big-endian only (elfv1 currently). It
is standard ABI for both 32-bit ppc and 64-bit ppc64 architectures.
With PowerPC little-endian support new ABI was introduced (elfv2) and it
is used primarily with ppc64le target only. While it has support for
big-endian it is not commonly used and elfv1 still preferred.
Musl does support only elfv2 ABI for both LE and BE and does not have
any plans to support elfv1.
Since then to build for powerpc64 with musl new ABI should be used. As
expected it is not compatible with elfv1 but that isn't problem as long
as there is no binary distributed software or assembly code written for
elfv1 ABI.
(From OE-Core rev: 68c9641855199f34aabe1050e863c21830116fe1)
Signed-off-by: Serhey Popovych <serhe.popovych@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The QEMU usermode fails with invalid instruction error when
used with those tunes.
(From OE-Core rev: c1c881ef7f4faf94f385ae742030382122ee5816)
Signed-off-by: Serhey Popovych <serhe.popovych@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
It is ppcpX, not ppcprX, where X is 6 or 7. While there select 32bit
tune for P5 machine to bring it inline with P6 and P7.
(From OE-Core rev: 220eee184a4f510cdf7c55e2aed00330abee1553)
Signed-off-by: Serhey Popovych <serhe.popovych@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Just adding `-device virtio-rng-pci' to the QEMU invocation will
add the device with a default host backend. As of QEMU 1.3+,
the default backend is to use the host's /dev/random as a
source of entropy. [1]
When the entropy pool is empty, reads from /dev/random will
block until additional environmental noise is gathered. [2]
For Yocto, if call runqemu frequently, it will consume lots
of host's /dev/random, and kernel finish crng init in guest get
more and more slowly.
Here are 4 times runqemu boot test:
[ 3.464432] random: crng init done
[ 20.874030] random: crng init done
[ 23.583589] random: crng init done
[ 23.858945] random: crng init done
Modify entropy source to /dev/urandom device on the host which
returns random bytes using a pseudorandom number generator seeded
from the entropy pool. Reads from this device do not block and
kernel finish crng init in guest will not delay.
Of course, the side effect is obviously, we lost the quality of
randomness, but the modification is only on runqemu script
rather than real embedded device, and it benefits oeqa efficiency
in which many cases call runqemu especially multiple oeqa builds
on one host.
After apply the fix:
[ 3.364670] random: crng init done
[ 4.619061] random: crng init done
[ 3.403897] random: crng init done
[ 3.450717] random: crng init done
[1] https://wiki.qemu.org/Features/VirtIORNG
[2] http://man7.org/linux/man-pages/man4/random.4.html
(From OE-Core rev: 853644f82eb3205ef3efc1ea3959c7225dfacf61)
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
*The PowerPC SPE support is split off to a separate powerpcspe port
(From OE-Core rev: c67bc9bdcbcf46ed54dfbafd1bb6e405ed77f86a)
Signed-off-by: Chunrong Guo <chunrong.guo@nxp.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
4.18 is replacing 4.15 as the latest kernel in the upcoming
release, so we update our preferred versions to match.
(From OE-Core rev: e12f9e7ede7eea408d6a97233c7c8df97e8bfb26)
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
SERIAL_CONSOLE was already deprecated in 2013, yet still some
machine configuration files were using it. This patch replaces
it with SERIAL_CONSOLES, which is the successor.
The default value in systemd-serialgetty.bb can also be safely
transitioned from SERIAL_CONSOLE to SERIAL_CONSOLES, as this
recipe already uses SERIAL_CONSOLES within do_install().
The documentation seems to be already up do date.
beaglebone-yocto.conf in the bsp-guide already uses
SERIAL_CONSOLES. The ref-manual redirects from SERIAL_CONSOLE
to SERIAL_CONSOLES.
[YOCTO #12653]
(From OE-Core rev: e5bde3d466869df9fce559e49842ac245153d045)
Signed-off-by: Maciej Pijanowski <maciej.pijanowski@3mdeb.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
qemumips,qemumips64 and x86 based machines include kernel-modules by
default. Remove the kernel modules from RRECOMMENDS to make it consistent
across the machines.
Tested using core-image-sato on qemumips, qemumips64, qemux86 and
qemux86-64
[YOCTO #12383]
(From OE-Core rev: 1c4d9f46d5d94ea15d0c0d68c412ec05b9f51820)
Signed-off-by: Maxin B. John <maxin.john@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Although there may still be specific cases which can benefit from the
ARM instruction set, the Thumb2 instruction set is generally a better
default for armv7a class CPUs. Distros such as Debian and Fedora have
been targeting Thumb2 by default for some time.
Note that setting ARM_INSTRUCTION_SET has no effect unless
TUNE_FEATURES contains "thumb" (which is controlled by the "t" suffix
in DEFAULTTUNE, e.g. armv7vehf-neon -vs- armv7vethf-neon, etc) so out
of tree machine configs may need to update their DEFAULTTUNE to take
advantage of this change.
(From OE-Core rev: c88304a78e528596ca481cabe273749c286c352a)
Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The logic can be improved and the historical NOISO/NOHDD variables
moved into the class and out of common code.
The variables are also then removed in favour of directly controlling
the behaviour from IMAGE_FSTYPES in line with all the other image types.
(From OE-Core rev: a052caed563a09a01f5a3ea1f0477f379c05bee0)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Currently we have local.conf included before qemu.inc, and ${DISTRO}.conf
after qemu.inc. They both possibly specify their expected kernel providers.
To let other config files override it in real use, this commit changes
kernel provider assignment to ??= .
(From OE-Core rev: 07e06abeb6b7ae5047c4b70818cd8873302e4940)
Signed-off-by: He Zhe <zhe.he@windriver.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
