Juergen Gross [Mon, 27 Nov 2023 10:24:53 +0000 (11:24 +0100)]
Mini-OS: explicitly mark symbols to be visible for apps
Add an EXPORT_SYMBOL() macro to explicitly mark a symbol to be visible
for an app or library linked with Mini-OS. This prepares hiding all
other symbols from external components, avoiding any problems with
duplicate symbol names.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Tue, 21 Nov 2023 09:49:53 +0000 (10:49 +0100)]
Mini-OS: fix 9pfs response receiving
When copying a 9pfs response chunk from the ring buffer across the
ring end, the local ring pointer and length field are not updated
correctly. Fix that.
Fixes: 0924fec1de58 ("Mini-OS: add 9pfs transport layer") Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Tue, 21 Nov 2023 09:49:52 +0000 (10:49 +0100)]
Mini-OS: fix 9pfs stat receive format
The format string of the received data for the 9pfs stat command is
missing the initial 2 byte total length specifier. Add it.
Fixes: 2d1dfccd3aa3 ("Mini-OS: add read and write support to 9pfsfront") Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Jason Andryuk <jandryuk@gmail.com>
Juergen Gross [Tue, 21 Nov 2023 09:49:51 +0000 (10:49 +0100)]
Mini-OS: don't crash if no shutdown node is available
It might be perfectly fine not to have a control/shutdown Xenstore
node. If this is the case, don't crash, but just terminate the
shutdown thread after issuing a message that shutdown isn't available.
In fini_shutdown() clearing the watch can result in an error now, in
case the early exit above was taken. Just ignore this error now.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Jason Andryuk <jandryuk@gmail.com>
Juergen Gross [Tue, 21 Nov 2023 09:49:50 +0000 (10:49 +0100)]
Mini-OS: get own domid
Get the own domid via creation of a temporary event channel. There is
no "official" way to read the own domid in PV guests, so use the event
channel interface to get it:
- allocate an unbound event channel specifying DOMID_SELF for the
other end
- read the event channel status which will contain the own domid in
unbound.dom
- close the event channel
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Jan Beulich [Thu, 2 Nov 2023 07:25:59 +0000 (08:25 +0100)]
build: silence GNU ld 2.39 warning about executable stacks
While for C files the compiler is supposed to arrange for emitting
respective information, for assembly sources we're responsible ourselves.
For the new use of cc-option to work we cannot pass -S to the compiler
anymore. We need the compiler to actually invoke the assembler, so switch
to using -c.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Mon, 13 Feb 2023 08:44:10 +0000 (09:44 +0100)]
Mini-OS: add 9pfs transport layer
Add the transport layer of 9pfs. This is basically the infrastructure
to send requests to the backend and to receive the related answers via
the rings.
As a first example add the version and attach requests of the 9pfs
protocol when mounting a new 9pfs device. For the version use the
"9P2000.u" variant, as it is the smallest subset supported by the qemu
based backend.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Mon, 13 Feb 2023 08:44:09 +0000 (09:44 +0100)]
Mini-OS: add 9pfs frontend
Add a frontend for the 9pfs PV device. For now add only the code needed
to connect to the backend and the related disconnect functionality. The
9pfs protocol support will be added later.
Due to its nature (ability to access files) the whole code is guarded
by "#ifdef HAVE_LIBC".
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Mon, 13 Feb 2023 08:44:07 +0000 (09:44 +0100)]
Mini-OS: add concept of mount points
Add the concept of mount points to Mini-OS. A mount point is a path
associated with a device pointer and an open() callback. A mount point
can be either a file (e.g. "/dev/mem") or a directory ("/var/log").
This allows to replace the special casing in the generic open()
handling with a generic mount point handling.
Prepare the open() callbacks to support creating new files by adding a
mode parameter.
Additionally add a close() prototype to include/lib.h, as it is missing
today.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Mon, 13 Feb 2023 08:44:06 +0000 (09:44 +0100)]
Mini-OS: xenbus: add support for reading node from directory
Especially PV device drivers often need to read multiple Xenstore
nodes from a common directory. Add support for reading a string or an
unsigned value by specifying the directory and the node.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Tue, 21 Jun 2022 07:23:13 +0000 (09:23 +0200)]
mini-os: fix number of pages for PVH
When getting the current allocation from Xen, this value includes the
pages allocated in the MMIO area. Fix the highest available RAM page
by subtracting the size of that area.
This requires to read the E820 map before needing this value. Add two
functions returning the current and the maximum number of RAM pages
taking this correction into account.
At the same time add the LAPIC page to the memory map in order to
avoid reusing that PFN for internal purposes.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Tue, 21 Jun 2022 07:23:12 +0000 (09:23 +0200)]
mini-os: prefer memory map via start_info for PVH
Since some time now a guest started in PVH mode will get the memory
map from Xen via the start_info structure.
Modify the PVH initialization to prefer this memory map over the one
obtained via hypercall, as this will allow to add information to the
memory map for a new kernel when supporting kexec.
In case the start_info structure doesn't contain memory map information
fall back to the hypercall.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Sun, 16 Jan 2022 08:33:26 +0000 (09:33 +0100)]
add struct file_ops for file type socket
Even with some special handling needed in select_poll(), add a struct
file_ops for FTYPE_SOCKET. Due to the need of the special handling it
isn't possible to use a dynamically allocated file type.
Most functions calling the file_ops methods can be simplified a lot now
that no file type specific handling is left. Same applies to the file
type name printing in debug/verbose mode.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Sun, 16 Jan 2022 08:33:25 +0000 (09:33 +0100)]
use file_ops and get_file_from_fd() for console
Add struct file_ops for the console related file types (FTYPE_CONSOLE
and savefile). FTYPE_CONSOLE remains statically defined, as it is used
to statically init stdin, stdout and stderr.
Instead of directly accessing the files[] array use get_file_from_fd().
With CONSOLE now handled via file_ops the bogus file descriptor case in
select_poll() now needs to be handled more explicit instead of dropping
into console handling, assuming that this case was basically meant to
cover SAVEFILE.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Sun, 16 Jan 2022 08:33:18 +0000 (09:33 +0100)]
remove gnttab specific member from struct file
The gnttab specific union member in struct file is no longer
needed, so remove it.
The gnttab file type and its associated handling can be removed, too,
as libxengnttab is now supplying a struct file_ops via a call of
alloc_file_type().
This removes all contents of CONFIG_LIBXENGNTTAB guarded sections, so
this config option can be removed.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Sun, 16 Jan 2022 08:33:17 +0000 (09:33 +0100)]
remove event channel specific struct file definitions
The event channel specific union member in struct file is no longer
needed, so remove it together with the associated structure
definitions.
The event channel file type and its associated handling can be removed,
too, as libxenevtchn is now supplying a struct file_ops via a call of
alloc_file_type().
This removes all contents of CONFIG_LIBXENEVTCHN guarded sections, so
this config option can be removed.
Add an extern declaration for event_queue as it is used by
libxenevtchn.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Sun, 16 Jan 2022 06:45:27 +0000 (07:45 +0100)]
add CONFIG_LIBXS item
Mini-OS contains a stripped down version of libxenstore in lib/xs.c.
Today it is being built always if CONFIG_XENBUS is set and libc
support is enabled.
In order to allow a Mini-OS specific build of libxenstore to be used
instead add a new CONFIG_LIBXS item which per default will have the
same setting as CONFIG_XENBUS. A user wanting to replace lib/xs.c
with libxenstore would just need to set CONFIG_XENBUS=y and
CONFIG_LIBXS=n.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Sun, 16 Jan 2022 06:45:26 +0000 (07:45 +0100)]
use function vectors instead of switch for file operations
Introduce file type specific function vectors for all the needed file
operations which are file type specific in order to prepare replacing
the large switch statements in each generic file function.
Add a function to allocate a new file type dynamically in order to
prepare removing direct dependencies to external components in the
future. For this reason switch the file type from an enum to an
unsigned int. Prepare removal of some statically defines file types
by putting them at the end of the defined list.
Change the debug output for the file type from a single character to
a string in order to support a future removal of the file_types[]
array.
Provide some functions useful for file_ops in future patches.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Sun, 16 Jan 2022 06:45:25 +0000 (07:45 +0100)]
remove file type FTYPE_XC
The only reason for the file type FTYPE_XC has been gone long time
ago: it was needed for xc_map_foreign_bulk(), which has been switched
to use libxenforeignmemory and doesn't need this special file any
more.
So remove everything related to FTYPE_XC.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Tue, 21 Dec 2021 08:46:36 +0000 (09:46 +0100)]
mini-os: modify grant mappings to work in PVH mode
For being able to use the grant mapping interface in PVH mode some
changes are required, as the guest needs to specify a physical address
in the hypercall interface.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Tue, 21 Dec 2021 08:46:35 +0000 (09:46 +0100)]
mini-os: prepare grantmap entry interface for use by PVH mode
Instead of passing the pointer of a grantmap entry to the
_gntmap_[un]map_grant_ref() sub-functions use the map pointer and the
entry index instead. This will be needed for PVH mode usage.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Tue, 21 Dec 2021 08:46:33 +0000 (09:46 +0100)]
mini-os: move x86 specific gnttab coding into arch/x86/gnttab.c
Having grant table code in arch/x86/mm.c seems wrong. Move it to the
new file arch/x86/gnttab.c, especially as the amount of code is
expected to grow further.
While doing that replace type casts to pte_t with the more appropriate
__pte() macro.
No functional change.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Tue, 21 Dec 2021 08:46:30 +0000 (09:46 +0100)]
mini-os: respect memory map when ballooning up
Today Mini-OS won't look at the memory map when ballooning up. This can
result in problems for PVH domains with more than 4 GB of RAM, as
ballooning will happily run into the ACPI area.
Fix that by adding only pages being marked as RAM in the memory map and
by distinguishing between the current number of RAM pages and the first
unallocated page.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Mon, 4 Oct 2021 09:40:00 +0000 (11:40 +0200)]
xenbus: support large messages
Today the implementation of the xenbus protocol in Mini-OS will only
allow to transfer the complete message to or from the ring page buffer.
This is limiting the maximum message size to lower values as the xenbus
protocol normally would allow.
Change that by allowing to transfer the xenbus message in chunks as
soon as they are available.
Avoid crashing Mini-OS in case of illegal data read from the ring
buffer.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Mon, 4 Oct 2021 14:19:24 +0000 (16:19 +0200)]
mini-os: add config options for xen libraries
Today close hooks into libxenctrl, libxenevtchn and libxengnttab are
under the CONFIG_XC umbrella. In order to support Mini-OS builds using
stable Xen libraries only, add CONFIG_LIBXENCTRL, CONFIG_LIBXENEVTCHN
and CONFIG_LIBXENGNTTAB config options.
In case CONFIG_XC was specified in the Mini-OS config explicitly, set
the three new variables to the specified value.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Thu, 19 Aug 2021 05:30:56 +0000 (07:30 +0200)]
mini-os: netfront: fix initialization without ip address in xenstore
Commit 4821876fcd2ff ("mini-os: netfront: fix suspend/resume handling")
introduced a NULL pointer dereference in the initialization of netfront
in the case of no IP address being set in Xenstore.
Fix that by testing this condition. At the same time fix a long
standing bug for the same condition if someone used init_netfront()
with a non-NULL ip parameter.
Costin Lupu [Tue, 15 Dec 2020 16:50:12 +0000 (16:50 +0000)]
mini-os: netfront: Handle init failure when no vifs are provided
These changes deal with the case when no vifs are created. This can happen when
no vif is provided in the config file for mini-os images with netfront support.
Signed-off-by: Costin Lupu <costin.lupu@cs.pub.ro> Reviewed-by: Samuel Thibault <samuel.thibault@@ens-lyon.org>
Suspend/resume handling of netfront is completely broken from the
beginning. Commit d225f4012d69a1 ("Save/Restore Support: Add
suspend/restore support for netfront") introduced a new structure
netfront_dev_list referencing the real struct netfront_dev elements.
This list is used to setup the devices when resuming again.
Unfortunately the netfront_dev elements are released during suspend,
so at resume time those references will be stale.
Fix this whole mess by dropping struct netfront_dev_list again and
link the netfront_dev elements directly into a list. When suspending
don't free those elements.
The ip-address, netmask and gateway strings can just be released when
suspending and reread from xenstore at resume time.
Fixes: d225f4012d69a1 ("Save/Restore Support: Add suspend/restore support for netfront") Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
mini-os: netfront: retrieve netmask and gateway via extra function
Commit 1b8ed31f4ce40 ("mini-os: netfront: Read netmask and gateway from
Xenstore") modified init_netfront() to take two additional parameters.
This broke the Xen build as init_netfront() is used in grub stubdom,
too.
So instead of tightly coupling Mini-OS and Xen build via this interface
modification undo this change of init_netfront() and add two other
functions for retrieving the netmask and gateway for a network device.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Costin Lupu [Wed, 19 Aug 2020 16:41:08 +0000 (19:41 +0300)]
mini-os: Add suspend stubs for ARM
This patch introduces the missing suspend-related stubs for ARM. It also
introduces the new arch_suspend() function in order to keep the logic in
kernel_suspend() arch independent.
Signed-off-by: Costin Lupu <costin.lupu@cs.pub.ro> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Costin Lupu [Wed, 19 Aug 2020 15:45:36 +0000 (18:45 +0300)]
mini-os: Use a single start_info_ptr variable
The second definition of the `start_info_ptr` variable was introduced by commit
e05eb0 which was part of a series trying to add suspend/resume support to
mini-os. This patch removes the second definition by reverting some changes of
the mentioned commit and of commit 892b66.
However the logic in shutdown.c is still left in an inconsistent state because
it still doesn't work for ARM. The solution should be part of a future patch.
Signed-off-by: Costin Lupu <costin.lupu@cs.pub.ro> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Costin Lupu [Wed, 19 Aug 2020 12:49:00 +0000 (15:49 +0300)]
mini-os: netfront: Read netmask and gateway from Xenstore
When providing the IP address via the config file, one can also add the
netmask and gateway address, e.g. "ip=192.168.0.2 255.255.255.0
192.168.0.1", which will be further published to Xenstore. This patch adds
support for reading the netmask and gateway values.
Signed-off-by: Costin Lupu <costin.lupu@cs.pub.ro> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Costin Lupu [Tue, 18 Aug 2020 13:44:06 +0000 (16:44 +0300)]
mini-os: console.c: Rename static variable from buf to __print_buf
lwip soure code also has a static variable called 'buf' in ip_frag.c. This can
get confusing when inspecting the binary (e.g. with objdump or something
similar). Therefore this patch renames the 'buf' variable used by print()
function to '__print_buf'.
Signed-off-by: Costin Lupu <costin.lupu@cs.pub.ro> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Juergen Gross [Sat, 15 Aug 2020 11:15:57 +0000 (13:15 +0200)]
mini-os: correct memory access rights for pvh mode
When running as a PVH guest the memory access rights are not set
correctly: _PAGE_USER should not be set and CR0.WP should be set.
Especially CR0.WP is important in order to let the allocate on
demand feature work, as it requires a page fault when writing to a
read-only page.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Mini-OS shouldn't use Xen internal paths for building. Import the
needed paths from Xen and fall back to the current values only if
the import was not possible.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Jason Andryuk [Mon, 27 Apr 2020 03:40:19 +0000 (23:40 -0400)]
mini-os: Avoid segfaults in tc{g,s}etattr
Commit c96c22f1d94 "mini-os: minimal implementations of some termios
functions" introduced implementations of tcgetattr and tcsetattr.
However, they do not check if files[fildes].cons.dev is non-NULL before
dereferencing. This is not a problem for FDs allocated through
alloc_fd, but the files array pre-allocates FDs 0-2 for stdio. Those
entries have a NULL .dev, so tc{g,s}etattr on them segfault.
ioemu-stubdom segfaults when term_init() calls tcgetattr on FD 0.
Restore tcgetattr and tcsetattr behavior when .dev is NULL equivalent to
unsupported_function as it was before c96c22f1d94.
Signed-off-by: Jason Andryuk <jandryuk@gmail.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
mini-os: allow 4096 event channels for 64-bit mini-os
Limiting the number of event channels to 1024 is fine for 32-bit
builds, but not for 64-bit ones. This might be a problem when using
Xenstore-stubdom as the number of domains which can be supported is
then limited to a little bit more than 1000.
So raise the number of event channels to 4096 in 64-bit builds.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
projects / mini-os.git / log