aurel32 [Sat, 7 Mar 2009 22:10:40 +0000 (22:10 +0000)]
do not pretend to support low voltage operation
Eliminate "mmc0: SD card claims to support the incompletely defined 'low voltage
range'. This will be ignored." warning. Qemu says the card is a SD card, and SD
spec doesn't define low-voltage cards, so do now pretend to be one.
aurel32 [Sat, 7 Mar 2009 22:00:56 +0000 (22:00 +0000)]
Work around QEMU GDB stub suboptimality
The current XML files claim, on floating point-supporting Power chips,
that $f0 is register 70. This would be fine, except that register 70
for non-XML-aware GDB is FPSCR. More importantly, 70 is less than
NUM_CORE_REGS (71) for Power, so a request for register 70 goes to the
"core" register reading routines, rather than the floating-point
register read routine we registered with gdb_register_coprocessor.
Therefore, when we are talking to an XML-aware GDB, we claim that
register has zero width, which causes the rest of QEMU's GDB stub to
send an error back to GDB, which causes GDB to be unable to read the
floating-point registers. (The problem is also present for SPE
registers and occurs in a slightly different way for Altivec registers.)
The best way to fix this is to have the "core register" XML files for
PPC32 and PPC64 claim that there is a 4-byte register 70, which causes
$f0 to be register 71, and everything works just fine from that point
forward.
aurel32 [Sat, 7 Mar 2009 22:00:49 +0000 (22:00 +0000)]
Fix off-by-one errors for Altivec and SPE registers
Altivec and SPE both have 34 registers in their register sets, not 35
with a missing register 32.
GDB would ask for register 32 of the Altivec (resp. SPE) registers and
the code would claim it had zero width. The QEMU GDB stub code would
then return an E14 to GDB, which would complain about not being sure
whether p packets were supported or not.
aurel32 [Sat, 7 Mar 2009 21:47:53 +0000 (21:47 +0000)]
arm: Fix gic_irq_state.level bitfield type
Found while cleaning up compiler warnings: GIC_*_LEVEL macros strongly
suggest that gic_irq_state.level is intended to be per-CPU and not just
a single, global bit. I'm unable to test the effect, but it seems to be
the most reasonable fix for the apparent brokenness.
blueswir1 [Sat, 7 Mar 2009 20:57:42 +0000 (20:57 +0000)]
Keep SLB in-CPU
Real 970 CPUs have the SLB not memory backed, but inside the CPU.
This breaks bridge mode for 970 for now, but at least keeps us from
overwriting physical addresses 0x0 - 0x300, rendering our interrupt
handlers useless.
I put in a stub for bridge mode operation that could be enabled
easily, but for now it's safer to leave that off I guess (970fx doesn't
have bridge mode AFAIK).
blueswir1 [Sat, 7 Mar 2009 20:53:18 +0000 (20:53 +0000)]
Activate uninorth AGP bridge
Linux tries to poke the AGP bridge port and is pretty sad when it can't,
so let's activate the old code again and throw out the bit modifications,
as we don't really do anything with the values anyways.
blueswir1 [Sat, 7 Mar 2009 20:52:22 +0000 (20:52 +0000)]
Implment tlbiel
Linux uses tlbiel to flush TLB entries in PPC64 mode. This special TLB
flush opcode only flushes an entry for the CPU it runs on, not across
all CPUs in the system.
blueswir1 [Sat, 7 Mar 2009 20:50:01 +0000 (20:50 +0000)]
Implement slbmte
In order to modify SLB entries on recent PPC64 machines, the slbmte
instruction is used.
This patch implements the slbmte instruction and makes the "bridge"
mode code use the slb set functions, so we can move the SLB into
the CPU struct later.
pbrook [Sat, 7 Mar 2009 15:24:59 +0000 (15:24 +0000)]
The _exit syscall is used for both thread termination in NPTL applications,
and process termination in legacy applications. Try to guess which we want
based on the presence of multiple threads.
Also implement locking when modifying the CPU list.
aurel32 [Fri, 6 Mar 2009 21:48:00 +0000 (21:48 +0000)]
Fix race condition on access to env->interrupt_request
env->interrupt_request is accessed as the bit level from both main code
and signal handler, making a race condition possible even on CISC CPU.
This causes freeze of QEMU under high load when running the dyntick
clock.
The patch below move the bit corresponding to CPU_INTERRUPT_EXIT in a
separate variable, declared as volatile sig_atomic_t, so it should be
work even on RISC CPU.
We may want to move the cpu_interrupt(env, CPU_INTERRUPT_EXIT) case in
its own function and get rid of CPU_INTERRUPT_EXIT. That can be done
later, I wanted to keep the patch short for easier review.
aliguori [Fri, 6 Mar 2009 20:27:37 +0000 (20:27 +0000)]
Support ACLs for controlling VNC access ("Daniel P. Berrange")
This patch introduces a generic internal API for access control lists
to be used by network servers in QEMU. It adds support for checking
these ACL in the VNC server, in two places. The first ACL is for the
SASL authentication mechanism, checking the SASL username. This ACL
is called 'vnc.username'. The second is for the TLS authentication
mechanism, when x509 client certificates are turned on, checking against
the Distinguished Name of the client. This ACL is called 'vnc.x509dname'
The internal API provides for an ACL with the following characteristics
- A unique name, eg vnc.username, and vnc.x509dname.
- A default policy, allow or deny
- An ordered series of match rules, with allow or deny policy
If none of the match rules apply, then the default policy is
used.
There is a monitor API to manipulate the ACLs, which I'll describe via
examples
(qemu) acl show vnc.username
policy: allow
(qemu) acl policy vnc.username denya
acl: policy set to 'deny'
(qemu) acl allow vnc.username fred
acl: added rule at position 1
(qemu) acl allow vnc.username bob
acl: added rule at position 2
(qemu) acl allow vnc.username joe 1
acl: added rule at position 1
(qemu) acl show vnc.username
policy: deny
0: allow fred
1: allow joe
2: allow bob
(qemu) acl show vnc.x509dname
policy: allow
(qemu) acl policy vnc.x509dname deny
acl: policy set to 'deny'
(qemu) acl allow vnc.x509dname C=GB,O=ACME,L=London,CN=*
acl: added rule at position 1
(qemu) acl allow vnc.x509dname C=GB,O=ACME,L=Boston,CN=bob
acl: added rule at position 2
(qemu) acl show vnc.x509dname
policy: deny
0: allow C=GB,O=ACME,L=London,CN=*
1: allow C=GB,O=ACME,L=Boston,CN=bob
By default the VNC server will not use any ACLs, allowing access to
the server if the user successfully authenticates. To enable use of
ACLs to restrict user access, the ',acl' flag should be given when
starting QEMU. The initial ACL activated will be a 'deny all' policy
and should be customized using monitor commands.
eg enable SASL auth and ACLs
qemu .... -vnc localhost:1,sasl,acl
The next patch will provide a way to load a pre-defined ACL when
starting up
aliguori [Fri, 6 Mar 2009 20:27:28 +0000 (20:27 +0000)]
Add SASL authentication support ("Daniel P. Berrange")
This patch adds the new SASL authentication protocol to the VNC server.
It is enabled by setting the 'sasl' flag when launching VNC. SASL can
optionally provide encryption via its SSF layer, if a suitable mechanism
is configured (eg, GSSAPI/Kerberos, or Digest-MD5). If an SSF layer is
not available, then it should be combined with the x509 VNC authentication
protocol which provides encryption.
eg, if using GSSAPI
qemu -vnc localhost:1,sasl
eg if using TLS/x509 for encryption
qemu -vnc localhost:1,sasl,tls,x509
By default the Cyrus SASL library will look for its configuration in
the file /etc/sasl2/qemu.conf. For non-root users, this can be overridden
by setting the SASL_CONF_PATH environment variable, eg to make it look in
$HOME/.sasl2. NB unprivileged users may not have access to the full range
of SASL mechanisms, since some of them require some administrative privileges
to configure. The patch includes an example SASL configuration file which
illustrates config for GSSAPI and Digest-MD5, though it should be noted that
the latter is not really considered secure any more.
Most of the SASL authentication code is located in a separate source file,
vnc-auth-sasl.c. The main vnc.c file only contains minimal integration
glue, specifically parsing of command line flags / setup, and calls to
start the SASL auth process, to do encoding/decoding for data.
There are several possible stacks for reading & writing of data, depending
on the combo of VNC authentication methods in use
- Clear. read/write straight to socket
- TLS. read/write via GNUTLS helpers
- SASL. encode/decode via SASL SSF layer, then read/write to socket
- SASL+TLS. encode/decode via SASL SSF layer, then read/write via GNUTLS
Hence, the vnc_client_read & vnc_client_write methods have been refactored
a little.
vnc_client_read: main entry point for reading, calls either
- vnc_client_read_plain reading, with no intermediate decoding
- vnc_client_read_sasl reading, with SASL SSF decoding
These two methods, then call vnc_client_read_buf(). This decides
whether to write to the socket directly or write via GNUTLS.
The situation is the same for writing data. More extensive comments
have been added in the code / patch. The vnc_client_read_sasl and
vnc_client_write_sasl method implementations live in the separate
vnc-auth-sasl.c file.
The state required for the SASL auth mechanism is kept in a separate
VncStateSASL struct, defined in vnc-auth-sasl.h and included in the
main VncState.
The configure script probes for SASL and automatically enables it
if found, unless --disable-vnc-sasl was given to override it.
aliguori [Fri, 6 Mar 2009 20:27:23 +0000 (20:27 +0000)]
Move TLS auth into separate file ("Daniel P. Berrange")
This patch refactors the existing TLS code to make the main VNC code
more managable. The code moves to two new files
- vnc-tls.c: generic helpers for TLS handshake & credential setup
- vnc-auth-vencrypt.c: the actual VNC TLS authentication mechanism.
The reason for this split is that there are other TLS based auth
mechanisms which we may like to use in the future. These can all
share the same vnc-tls.c routines. In addition this will facilitate
anyone who may want to port the vnc-tls.c file to allow for choice
of GNUTLS & NSS for impl.
The TLS state is moved out of the VncState struct, and into a separate
VncStateTLS struct, defined in vnc-tls.h. This is then referenced from
the main VncState. End size of the struct is the same, but it keeps
things a little more managable.
The vnc.h file gains a bunch more function prototypes, for functions
in vnc.c that were previously static, but now need to be accessed
from the separate auth code files.
The only TLS related code still in the main vl.c is the command line
argument handling / setup, and the low level I/O routines calling
gnutls_send/recv.
aliguori [Fri, 6 Mar 2009 20:27:13 +0000 (20:27 +0000)]
Move VNC structs into header file ("Daniel P. Berrange")
This patch moves the definitions of VncState and VncDisplay structs
out into a vnc.h header file. This is to allow the code for TLS
and SASL auth mechanisms to be moved out of the main vnc.c file.
aliguori [Fri, 6 Mar 2009 20:27:10 +0000 (20:27 +0000)]
Refactor keymap code to avoid duplication ("Daniel P. Berrange")
Each of the graphical frontends #include a .c file, for keymap code
resulting in duplicated definitions & duplicated compiled code. A
couple of small changes allowed this to be sanitized, so instead of
doing a #include "keymaps.c", duplicating all code, we can have a
shared keymaps.h file, and only compile code once. This allows the
next patch to move the VncState struct out into a header file without
causing clashing definitions.
aliguori [Fri, 6 Mar 2009 20:27:05 +0000 (20:27 +0000)]
Enhance 'info vnc' monitor output ("Daniel P. Berrange")
The current 'info vnc' monitor output just displays the VNC server address
as provided by the -vnc command line flag. This isn't particularly useful
since it doesn't tell you what VNC is actually listening on. eg, if you
use '-vnc :1' it is useful to know whether this translated to '0.0.0.0:5901'
or chose IPv6 ':::5901'. It is also useful to know the address of the
client that is currently connected. It is also useful to know the active
authentication (if any).
This patch tweaks the monitor output to look like:
(qemu) info vnc
Server:
address: 0.0.0.0:5902
auth: vencrypt+x509
Client: none
More data will be added to this later in the patch series...
The 'addr_to_string' helper method in this patch is overly generic
for the needs of this patch alone. This is because it will be re-used
by the later SASL patches in this series, where the flexibility is
important.
In the case where the TLS handshake does *not* block on I/O, QEMU
sends the next 'start sub-auth' message twice. This seriously confuses
the VNC client :-) Fortunately the chances of the handshake not blocking
are close to zero for a TCP socket, which is why it has not been noticed
thus far. Even with both client & server on localhost, I can only hit the
bug 1 time in 20.
NB, the diff context here is not too informative. If you look at the
full code you'll see that a few lines early we called vnc_start_tls()
which called vnc_continue_handshake() which called the method
start_auth_vencrypt_subauth(). Hence, fixing the bug, just involves
removing the 2nd bogus call to start_auth_vencrypt_subauth() as per
this patch.
aliguori [Thu, 5 Mar 2009 23:01:55 +0000 (23:01 +0000)]
monitor: Pass-through for gdbstub (Jan Kiszka)
Create a monitor terminal and pass it through the gdbstub. This allows
to use gdb's monitor command to access the QEMU monitor. Works for all
commands except for non-detached migration and password retrieval (user
will receive error messages instead).
aliguori [Thu, 5 Mar 2009 23:01:51 +0000 (23:01 +0000)]
monitor: Introduce MONITOR_USE_READLINE flag (Jan Kiszka)
This allows to create monitor terminals that do not make use of the
interactive readline back-end but rather send complete commands. The
pass-through monitor interface of the gdbstub will be an example.
Up to now, you never really knew if you already switched the console
after pressing CTRL-A C or if you mistyped it again. This patch
clarifies the situation by providing a prompt in a new line and
injecting a linebreak when switching away again. For this purpose, the
two events CHR_EVENT_MUX_IN and CHR_EVENT_MUX_OUT are introduced and
distributed on focus switches.
aliguori [Thu, 5 Mar 2009 23:01:42 +0000 (23:01 +0000)]
monitor: Decouple terminals (Jan Kiszka)
Currently all registered (and activate) monitor terminals work in
broadcast mode: Everyone sees what someone else types on some other
terminal and what the monitor reports back. This model is broken when
you have a management monitor terminal that is automatically operated
and some other terminal used for independent guest inspection. Such
additional terminals can be multiplexed device channels or a gdb
frontend connected to QEMU's stub.
Therefore, this patch decouples the buffers and states of all monitor
terminals, allowing the user to operate them independently. It finally
starts to use the 'mon' parameter that was introduced earlier with the
API rework. It also defines the default monitor: the first instantance
that has the MONITOR_IS_DEFAULT flag set, and that is the monitor
created via the "-monitor" command line switch (or "vc" if none is
given).
As the patch requires to rework the monitor suspension interface, it
also takes the freedom to make it "truely" suspending (so far suspending
meant suppressing the prompt, but inputs were still processed).
aliguori [Thu, 5 Mar 2009 23:01:37 +0000 (23:01 +0000)]
monitor: Introduce ReadLineState (Jan Kiszka)
As another step towards decoupled monitor terminals encapsulate the
state of the readline processor in a separate data structure called
ReadLineState and adapt all interfaces appropriately. For now the
monitor continues to instantiate just a single readline state.
aliguori [Thu, 5 Mar 2009 23:01:33 +0000 (23:01 +0000)]
monitor: Drop banner hiding (Jan Kiszka)
There is no use for the hide/show banner option, and it is applied
inconsistently anyway (or what makes the difference between
-serial mon:stdio and -nographic for the monitor?). So drop this mode.
aliguori [Thu, 5 Mar 2009 23:01:23 +0000 (23:01 +0000)]
monitor: Rework API (Jan Kiszka)
Refactor the monitor API and prepare it for decoupled terminals:
term_print functions are renamed to monitor_* and all monitor services
gain a new parameter (mon) that will once refer to the monitor instance
the output is supposed to appear on. However, the argument remains
unused for now. All monitor command callbacks are also extended by a mon
parameter so that command handlers are able to pass an appropriate
reference to monitor output services.
For the case that monitor outputs so far happen without clearly
identifiable context, the global variable cur_mon is introduced that
shall once provide a pointer either to the current active monitor (while
processing commands) or to the default one. On the mid or long term,
those use case will be obsoleted so that this variable can be removed
again.
Due to the broad usage of the monitor interface, this patch mostly deals
with converting users of the monitor API. A few of them are already
extended to pass 'mon' from the command handler further down to internal
functions that invoke monitor_printf.
At this chance, monitor-related prototypes are moved from console.h to
a new monitor.h. The same is done for the readline API.
aliguori [Thu, 5 Mar 2009 23:01:15 +0000 (23:01 +0000)]
monitor: Rework modal password input (Jan Kiszka)
Currently, waiting for the user to type in some password blocks the
whole VM because monitor_readline starts its own I/O loop. And this loop
also screws up reading passwords from virtual console.
Patch below fixes the shortcomings by using normal I/O processing also
for waiting on a password. To keep to modal property for the monitor
terminal, the command handler is temporarily replaced by a password
handler and a callback infrastructure is established to process the
result before switching back to command mode.
aliguori [Thu, 5 Mar 2009 23:01:01 +0000 (23:01 +0000)]
monitor: Rework early disk password inquiry (Jan Kiszka)
Reading the passwords for encrypted hard disks during early startup is
broken (I guess for quiet a while now):
- No monitor terminal is ready for input at this point
- Forcing all mux'ed terminals into monitor mode can confuse other
users of that channels
To overcome these issues and to lay the ground for a clean decoupling of
monitor terminals, this patch changes the initial password inquiry as
follows:
- Prevent autostart if there is some encrypted disk
- Once the user tries to resume the VM, prompt for all missing
passwords
- Only resume if all passwords were accepted
aliguori [Thu, 5 Mar 2009 23:00:57 +0000 (23:00 +0000)]
monitor: Report encrypted disks in snapshot mode (Jan Kiszka)
If the backing file is encrypted, 'info block' currently does not report
the disk as encrypted. Fix this by using the standard API to check disk
encryption mode. Moreover, switch to a canonical output format.
aliguori [Thu, 5 Mar 2009 23:00:53 +0000 (23:00 +0000)]
monitor: Use reasonable default virtual console size (Jan Kiszka)
If a target uses a tiny display (like the MusicPal), the default monitor
is currently set to the same size. Fix this by applying the same
defaults like already used serial and virtio consoles.
aliguori [Thu, 5 Mar 2009 23:00:02 +0000 (23:00 +0000)]
char-mux: Use separate input buffers (Jan Kiszka)
Currently, the intermediate input buffer of mux'ed character devices
records data across all sub-devices. This has the side effect that we
easily leak data recorded over one sub-devices to another once we switch
the focus. Avoid data loss and confusion by defining exclusive buffers.
Note: In contrast to the original author's claim, the buffering concept
still breaks down when the fifo of the currently active sub-device is
full. As we cannot accept futher data from this point on without risking
to loose it, we will also miss escape sequences, just like without all
that buffering. In short: There is no reliable escape sequence handling
without infinite buffers or the risk of loosing some data.
aliguori [Thu, 5 Mar 2009 22:59:58 +0000 (22:59 +0000)]
char: Fix initial reset (Jan Kiszka)
Recent changes to the graphical console initialization broke the initial
CHR_EVENT_RESET distribution. The reset BHs generated on char device
initialization are now already consumed during machine init (ide init
... -> qemu_aio_wait -> qemu_bh_poll). Therefore, this patch moves the
initial qemu_chr_reset calls into a separate funtion which is called
after machine init.
Signed-off-by: Ryan Harper <[email protected]>
---
Subject: [PATCH] lsi: add ISTAT1 register read
From: Ryan Harper <[email protected]> Cc: [email protected]
SLES10 SP2 installer complains when probing a scsi disk and exits qemu when
failing to read one of the registers.
aliguori [Wed, 4 Mar 2009 19:25:22 +0000 (19:25 +0000)]
Update cocoa.m to match new DisplayState code (Samuel Benson)
Version 2 does as follows:
[1]: Corrects endianness on issues by using native BGR to RGB conversion
[2]: Uses DisplayState accessors for obtaining graphics context information,
which
[3]: Removes now unused variables, and
[4]: Allows reading of varying color modes (32bit/24/16), and converting to
native colorspace
[5]: Attempts to keep itself centered on screen (as opposed to bottom right,
which immediately goes off screen after bios load) on context changes
(window resizes)
Testing working on i386 (gentoo, Windows 2000) and PPC (debian) guests on PPC
and x86 Macs.
In regards to [4], Windows 2000 displays fine on quick tests, but on the lowest
setting I could test, 16bit color depth at 4bpp, colors are slightly off. I
used gentoo install-x86-minimal-2008.0 in framebuffer mode to test above
setting; the usual grey text is now blue, and Tux appears to be BGR shifted. I
do not know if previous code worked at such a low color setting.
pbrook [Wed, 4 Mar 2009 12:18:15 +0000 (12:18 +0000)]
The different VFP variants will never occur at the same time, and giving them
different names confuses GDB, so use org.gnu.gdb.arm.vfp for all of them.
aliguori [Tue, 3 Mar 2009 17:37:21 +0000 (17:37 +0000)]
Fix SDL on evdev hosts (Anthony Liguori)
This patch corrects SDL support on X11 hosts using evdev. It's losely based
on the previous patch by Dustin Kirkland and the evdev support code in gtk-vnc
written by Daniel Berrange.
Introduce a growable flag that's set by bdrv_file_open(). Block devices should
never be growable, only files that are being used by block devices.
I went through Fabrice's early comments about the patch that was first applied.
While I disagree with that patch, I also disagree with Fabrice's suggestion.
There's no good reason to do the checks in the block drivers themselves. It
just increases the possibility that this bug could show up again. Since we're
calling bdrv_getlength() to determine the length, we're giving the block drivers
a chance to chime in and let us know what range is valid.
Basically, this patch makes the BlockDriver API guarantee that all requests are
within 0..bdrv_getlength() which to me seems like a Good Thing.
projects / qemu.git / log