similar to: [PATCH supermin 0/2] Allow an alternate libc to be used for init.

Allow an alternate libc, such as dietlibc, to be used for the init binary in the supermin appliance. Rich.

Hi, the two patches improve the way memory is handled in supermin, by cleanly exiting on memory allocation failures, and free'ing memory when not needed (to keep working and not run out of memory). Pino Toscano (2): Check for failures in memory allocations Free memory buffers when not used src/ext2fs-c.c | 13 +++++++++++-- src/init.c | 13 +++++++++++++ 2 files changed, 24

Joseph, Please try my modified versions of these patches. These are compile-tested on Fedora and they don't break any existing functionality, but I don't have either urpmi nor a statically-linked xz so I cannot fully test them. I have also fixed detection of zlib (2/3). Rich.

v1 -> v2: - I fixed the end condition of the loop in parse_root_uuid. - Retested. Didn't change the busy wait loop into a function or macro, as per discussion on IRC. Rich.

This patch is the minimum possible to add virtio-scsi support to febootstrap. Problem: SCSI devices are detected asynchronously after the Linux kernel module is loaded. The usual solutions to this involve scsi_wait_scan (deprecated apparently) or udev. However febootstrap doesn't use udev, and adding it is a ton of work with a very uncertain outcome. So the minimum possible change is to

When building the mini initrd, previously we copied the modules into the initrd as-is, so for example if the module was xz-compressed, we copied the foo.ko.xz file to the initrd. This requires that the mini init binary is linked to zlib & lzma, so that it knows how to uncompress these modules when insmoding them at boot time. Also since the init is statically linked, it required _static_

Require the root= parameter is passed to specify which root (apppliance) to mount. Libguestfs has done this since 2012. The other two patches are small code refactorings in the init program. Rich.

After supermin has finished running, the initramfs files sit around occupying swappable memory but serving no further purpose. This saves a little memory, at the cost of about 1ms of extra boot time. --- init/init.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 48 insertions(+), 2 deletions(-) diff --git a/init/init.c b/init/init.c index 733d66e..5ac53e9 100644 ---

Instead of specifying a device name (eg. root=/dev/sdb), this permits specifying an ext4 volume UUID (root=UUID=12345678-...). This allows the appliance to be robust against the non-determinism of SCSI device enumeration. --- init/init.c | 226 +++++++++++++++++++++++++++++++++++++++++++++++++++--------- 1 file changed, 192 insertions(+), 34 deletions(-) diff --git a/init/init.c b/init/init.c

Various patches to make supermin quieter. By outputting fewer messages on the fast path, we use the slow emulated UART less, and this improves boot times. Also remove some kernel modules that we cannot or should not be using, which also improves boot times. Rich.

--- .gitignore | 6 +- .gitmodules | 3 - HACKING | 41 +++ Makefile.am | 6 +- README | 7 +- TODO | 61 +--- autobuild.sh | 65 ---- autogen.sh | 12 - configure.ac | 27

Various changes/fixes to use smaller stack frames. Rich.

This two-part patch for supermin implements device trees (for ARM). The first patch introduces a more rational way to handle command line arguments in 'supermin-helper'. See the commit message for details. The old style is still supported for compatibility. The second patch adds an extra supermin-helper --dtb parameter specifying a wildcard. A device tree file which matches the

This needs loads more testing before we include it. However it's in good shape for testing and review. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones virt-df lists disk usage of guests without needing to install any software inside the virtual machine. Supports Linux and Windows. http://et.redhat.com/~rjones/virt-df/ -------------- next part

--- configure.ac | 8 +++++--- 1 files changed, 5 insertions(+), 3 deletions(-) diff --git a/configure.ac b/configure.ac index da03c9f..7606bca 100644 --- a/configure.ac +++ b/configure.ac @@ -68,7 +68,8 @@ dnl For ArchLinux handler. AC_CHECK_PROG(PACMAN,[pacman],[pacman],[no]) dnl Required programs, libraries. -AC_PATH_PROG([INSMODSTATIC],[insmod.static],[no])

v1 was here: https://www.redhat.com/archives/libguestfs/2016-March/thread.html#00157 Not running the 'hwclock' command reduces boot times considerably. However I'm not sure if it is safe. See the question I posted on qemu-devel: http://thread.gmane.org/gmane.comp.emulators.qemu/402194 At the moment, about 50% of the time is consumed by SeaBIOS. Of this, about ⅓rd is SGABIOS

On Wed, Apr 19, 2017 at 10:57:28AM +0100, Richard W.M. Jones wrote: > + struct timespec t; > +#define NANOSLEEP(ns) do { \ > + t.tv_sec = delay_ns / 1000000000; \ > + t.tv_nsec = delay_ns % 1000000000; \ > + nanosleep (&t, NULL); \ > + } while(0) > + Defining 't' outside its context is a bit

Lots of changes since v2, too much to remember or summarize. Please ignore patch 11/11, it's just for my testing. Rich.

DAX is explained in detail here: https://lwn.net/Articles/610174/ This patch adds support to supermin for using a DAX root filesystem. The corresponding libguestfs patches will be posted shortly once I've tested them a bit more. This requires qemu >= 2.6. Unfortunately it's not really a win for a few reasons: - Requires enabling ACPI, which slows everything down by hundreds of

v4: - Lots more analysis of the /init script and other parts. - Display a list of the longest to shortest activities. - Rebase on top of current head. Rich.