Currently custom rootfs init is built along with the main KVM tools executable
and is copied into custom rootfs directories when they are created with
'kvm setup'. The problem there is that if the init code changes, they have
to be manually copied to custom rootfs directories.
Instead, this patch splits init process into two parts. One part that simply
handles mounts, and passes it to stage 2 of the init.
Stage 2 really sits along in the code tree, and does all the heavy lifting.
This allows us to make init changes in the code tree and have it automatically
be updated in custom rootfs guests without having to copy files over manua
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
[ penberg@kernel.org: fix 'make check' breakage in Makefile ]
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Free up the BAR to make space for the new virtio BARs. It isn't required
to have the PBA and the table in the separate BARs, and uniting them will
just give us extra BARs to play with.
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
We don't really support that, so no point in using 64bit BARs.
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Sasha reported, that a non RT guest reports "too much work for irq 4"
with the previous serial overhaul.
The reason is, that the new code allows unlimited tx transfers, which
triggers the sanity check in the 8250.c interrupt handler.
Limit the consecutive TX chars to 16 and let the guest kernel escape
from the 8250 interrupt handler. Set the TEMT/THRE bits in the
periodic serial console update.
Reported-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
The interrupt injection of the serial emulation is completely
broken. It's just doing random toggling of the interrupt line, which
can lead to complete console hangs.
The real hardware asserts the interrupt line when a condition
(RX/TX/Status) is met and the corresponding interrupt is enabled in
the IER. It's deasserted when the condition is cleared or the
corresponding interrupt is disabled in the IER.
So the correct emulation just needs to check after each state change
in the LSR or the IER which bits in the IIR need to be set and update
the interrupt line accordingly. To avoid setting the same state over
and over keep an internal state of the last set interrupt line state
and only update via the kvm ioctl when the new state differs.
Rename serial8250__inject_interrupts() to serial8250__update_consoles()
which reflects what the function really is about.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
There is no point to have the same switch case construct for all the
registers, just to take care of the oddball case of DLL/DLM.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
It's nice to align struct initializers, but random tab insertion into
the code flow is a horrible idea.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
pci_get_io_space_block() used to grab addresses from
KVM_32BIT_GAP_START + 0x1000000, which is x86-specific. Create a new define,
KVM_PCI_MMIO_AREA, to specify a bus address these allocations can come from.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
This allows config space access in a more natural manner than clunky x86 IO ports,
and is useful for other architectures. Internally, the x86 IO port access uses
these new config space interfaces.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
The BAR addresses are set up fine, but missed the bar_size[] array which is now
updated correspondingly.
Use PCI_IO_SIZE instead of '0x100'.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
vesa, pci-shmem and virtio-pci devices need to set up config space with
little-endian conversions (as config space is LE). The pci_config_address
bitfield also needs to be reversed when building on BE systems.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Other pieces of kvmtool may be interested in PCI_MAX_DEVICES.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
IRQ init may involve per-CPU setup/allocation of resources, so make sure
kvm->nrcpus is initialised before calling irq__init().
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Currently some devices (in this case kbd, fb, vesa) are initialised after
CPU/firmware setup. On some platforms (e.g. PPC) kvm__arch_setup_firmware() may
be making a device tree. Any devices added after this point will be missed!
Tiny refactor of builtin-run.c, moving timer start, firmware setup, cpu init
to occur last.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Re-arrange pci__init() in builtin-run such that it comes before devices are
initialised.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
This patch passes the initrd fd and commandline to load_flat_binary(), which may
be used to load both the kernel & an initrd (stashing or inserting the
commandline as appropriate) in the same way that load_bzimage() does. This is
especially useful when load_bzimage() is unused for a particular
architecture. :-)
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
cpios are valid as initrds too, so allow them through the check.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
term_getc()'s int c has one byte written into it (at its lowest address) by
read_in_full(). This is expected to be the least significant byte, but that
isn't the case on BE! Use correct type, unsigned char. A similar issue exists
in term_getc_iov(), which needs to write a char to the iov rather than an int.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
This patch paves the way for adding a hypervisor console, useful on systems that
support one out of the box yet don't have either serial port or virtio console
support (e.g. kernels expecting POWER SPAPR).
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Different systems will want different base kernel commandlines, e.g. non-x86
systems probably don't need noapic, i8042.* etc., so set the commandline up in
arch-specific code. Then, if the resulting commandline is empty, don't strcat a
space onto the front.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Currently, the SIGALRM handler calls device poll functions (for serial, virtio
console) directly. Which devices are present and which require polling is a
system-specific decision, so create a new function called from common code &
move the x86-specific poll calls into it.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
kvm_cpu__run() currently die()s if KVM_RUN returns non-zero. Some architectures
may return positive values in non-error cases, whereas real errors are always
negative return values. Check for those instead.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
We die() if we can't read KVM_CAP_NR_VCPUS, but the API docs suggest to assume
the value 4 in this case. This is pertinent to PPC KVM, which currently
does not support this CAP.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
This patch creates a new function in x86/kvm-cpu.c, kvm_cpu__handle_exit(), in
which arch-specific exit reasons can be handled outside of the common runloop.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
The checks for optional libraries build code to perform the tests, so should
respect certain CFLAGS -- in particular, -m64 so we check for 64bit libraries if
they're required.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
This patch adds an 'I' parameter to override the default kernel include path of
'../../include'.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Not every architecture has an i8042 kbd controller, so only use this when
building for x86.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
The process of probing the 8250 serial port is as follows:
1. Start detecting IRQs
2. Enable the IER register [At this point, the port is supposed to light
the INTR].
3. Stop detecting IRQs [At this point, the driver detects which IRQ
belongs to that port].
4. Disable IER register.
Since we weren't enabling and disabling the IRQ based on IER writes, we
would often fail the probing since the driver couldn't detect which IRQ
is used by the port, and would just default that to 0.
This would cause slowness and may have caused hangs. For me there is a
significant increase in speed of the terminal after this patch.
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
[ penberg@kernel.org: whitspace cleanup ]
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Fix a typo where kvm->nrcpus was set, then immediately overwritten with
older value.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
This allows users to pass a pre-configured fd to use for the network
interface.
For example:
kvm run -n mode=tap,fd=3 3<>/dev/net/tap3
Acked-by: Daniel P. Berrange <berrange@redhat.com>
Cc: Osier Yang <jyang@redhat.com>
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Ingo Molnar writes:
On Tue, 6 Dec 2011, Ingo Molnar wrote:
> > FYI, today's version fails to build:
> >
> > In file included from x86/include/kvm/barrier.h:13:0,
> > from virtio/core.c:5:
> > ../../arch/x86/include/asm/system.h:404:1: error: unknown type
> > name ‘bool’
> > make: *** [virtio/core.o] Error 1
> > make: *** Waiting for unfinished jobs....
> >
> > latest Fedora Rawhide.
>
> There's no 'bool' in system.h for 3.2-rc4. Is this something that's
> changed in -tip?
It got introduced by a post-rc4 fix:
e5fd47bfab2d: xen/pm_idle: Make pm_idle be default_idle under Xen.
Fix it by include <stdbool.h> in kvm/barrier.h
Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
This enables the custom rootfs to run executables which have dylibs located
in non standard paths by taking these settings from the host.
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Initialize irq_routing, and specifically irq_routing->nr before using it.
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Zero out struct epoll_event before calling epoll_ctl(). This isn't really
required, but is done to avoid warnings.
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
The field size is currently wrong, read into a 32bit word instead of 16. This
casues trouble when BE.
Acked-by: Asias He <asias.hejun@gmail.com>
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Create a new arch-specific subdirectory to contain architecture-specific code
and includes.
The Makefile now adds various arch-specific objects based on detected
architecture. That aside, this patch should only contain code moves. These
include:
- x86-specific kvm_cpu setup, kernel loading, memory setup etc. now in
x86/kvm{-cpu}.c
- BIOS now lives in x86/bios/
- ioport setup
- KVM extensions are asserted in arch-specific kvm.c now, so each architecture
can manage its own dependencies.
- Various architecture-specific #defines are moved into $(ARCH)/include/kvm{-cpu}.h
such as struct kvm_cpu, KVM_NR_CPUS, KVM_32BIT_GAP_SIZE.
Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Rusty has just removed it out of the spec. Since we probably the only ones
who implemented support for it, we should remove it out of our code as well.
There is no issue with breaking anything since nothing else worked with it,
so it's fully backwards compatible.
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
It is supposed to have no write ops in ro_ops_nowrite disk operation.
However, there is one. Let's remove it.
Signed-off-by: Asias He <asias.hejun@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
ro_ops is never used after the assignment, so no need to do the
assignment.
Signed-off-by: Asias He <asias.hejun@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
This patch reduces the number of calls to io_getevents() by getting
multiple io events at a time instead of one in disk image thread.
Signed-off-by: Asias He <asias.hejun@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
When meeting request to write the cluster without copied flag,
allocate a new cluster and write original data with modification
to the new cluster. This also adds support for the writing operation
of the qcow2 compressed image. After testing, image file can pass
through "qemu-img check". The performance is needed to be improved.
Signed-off-by: Lan Tianyu <tianyu.lan@intel.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Indirect buffers are ring descriptors which point to more (even more)
descriptors.
This can be used to increase the effective ring capacity, which helps the
guest to batch large requests - very useful for blk devices.
This patch also enables indirect buffers for virtio-net and virtio-blk.
The patch is based on the lguest's code which does the same.
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
There are at most bdev->reqs[VIRTIO_BLK_QUEUE_SIZE] outstanding requests
at any time. We can simply use the head of each request to fetch the
right 'struct blk_dev_req' in bdev->reqs[].
So, we can eliminate the list and lock operations which introduced by
virtio_blk_req_{pop, push}.
Signed-off-by: Asias He <asias.hejun@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
This adds a ->cmos_data array to 'struct rtc_device' that is used to emulate
nonvolatile BIOS memory and the RTC registers.
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Add missing RTC register names to hw/rtc.c and rename the current ones to
follow <linux/mc146818rtc.h> naming. It would be nice to use the header
directly but unfortunately it includes <linux/spinlock.h>.
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Sasha Levin