by Michele Giacomone
Detailed description:
-Observed issues
Due to some limitations only dates between 1980 and 2038 can be
used in a reliable way.
Also, bochs incorrectly assumes a linear correspondence between
the data returned by the <time.h> functions localtime() and
mktime(), and isn't setting the latter properly.
Bochs keeps its internal time value dependent to these functions
after setup, assuming that their internal settings won't change
on the go - which is not the case.
In my OS, and in my timezone, this leads to incorrect startup values
for 5 months each year and unreliable values if the simulation is
kept going for a long time. (a feedback between localtime() and
mktime() is created which keeps shifting back the time)
Also, the RTC simulation is not realistic since the clock fixes
itself across DST changes, without updating any DST related flag,
a behavior that no guest OS expects.
-Proposed fix
This is implemented in such way that no bochs' previous behavior
is changed, a part from the broken ones, with legacy in mind
== the user can keep using bochs exactly as before knowing nothing
of this patch
+Make the internal s.timeval variable a Bit64s, so it can fit all
values that the cmos can correctly represent, reported below:
MIN setting -62167219200 => 0000/01/01 SAT 0:00:00
MAX BCD setting 253402300799 => 9999/12/31 FRI 23:59:59
MAX BIN setting 745690751999 => 25599/12/31 FRI 23:59:59
And then fix each reference to these so it can handle such values
And make bochs correctly wrap around for under/overflows, so that
only the most significant bits of the century are lost.
+Do the same thing to the bochs time0 parameter, so all the above
values can be chosen at startup (despite being now legal values,
1 and 2 will still be treated as "local" and "utc"). Note that
normally only BCD settings are valid since bochs' CMOS defaults
to such operating mode - the only way to use the binary range
is by loading a cmos memory map.
+Make the internal s.timeval variable independent from external
factors. This means providing a small set of time handling
functions, contained in "iodev/utctime.h", which must work in
any environment in which bochs compiles, accessing no external
resource. This also means that after startup, s.timeval will only
be changed internally, and no call to the OS time functions will
be made.
+Make the internal s.timeval variable timezone independent, to
have a linear correlation between its values and valid CMOS
settings. To make it easier, s.timeval is gonna be treated as
if the current timezone was UTC: so,
- if the user selects UTC as time0, s.timeval will become current
time(NULL)
- if the user selects localtime, s.timeval will be computed as
the value which will display the same broken down time as
localtime(&now)
- if the user inputs a time formatted string the proper s.timeval
to displayed will be easily calculated,
- if the user inputs a starting time value, s.timeval will be
computed as the value which will display the same broken down
time as localtime(&user_input) to ensure the same operation as
before.
A "tz=utc" is displayed when bochs prints out the current time
value, to warn users about the difference in meaning between the
internally kept time value and the value they can set through
the "time0=" parameter. This might be changed to communicate
instead the time value they can input to get the same setting,
but performing such calculation (except for the startup time)
suffers from all the mktime()/localtime() problems listed above
so I did not do it.
The range of "time0" is automatically adjusted so all users in
all time zones can set any legal value despite "time0=" having a
local meaning.
A thorough explanation of what I did and why can be found in the
"iodev/utctime.h" library header.
---------
Co-authored-by: Stanislav Shwartsman <sshwarts@users.sourceforge.net>
Co-authored-by: Volker Ruppert <Volker.Ruppert@t-online.de>
- Added new PCI chipset choice for the i440BX AGPset. Some basic work is done,
but AGP support is not present yet.
- Added new class for the "virtual" PCI-to-PCI bridge that should manage the
secondary bus (AGP). Since this device must appear with device number #1 at
the primary bus, it was required to change the PCI device numbers for the
i440BX case. Moved the PIIX4 module to device number #7. The presence of the
PCI base address regions now depends on the header type as expected.
- Since the Bochs BIOS cannot handle the modified PCI device layout, all tests
continued with an external BIOS designed for this chipset (GA-6BA_F1.bin).
This BIOS requires additional changes in some devices.
- ACPI: Return value 0 for some status registers and the GPI registers.
- CMOS: Since the PIIX4 supports a 256 byte CMOS RAM, prepared support for it
and enable it in case a 256 byte CMOS image is used.
- PCI: The device numbers for 4 slots starting at #8. The 5th slot could be
used for AGP when available.
Make save/restore default feature, the configure option for save/restore removed from configure script and save/restore made available forever. All code now assume it is exists. Bochs save/restore tree previosly called "save_restore" renamed to "bochs" tree and it will be havily used everywhere, starting from save/restore and ending by various bochs debugger functions. I am going to rework debugger code to get rid of debug CPU access functions and use this "bochs" param tree instead
done by the object destructor (changes in cmos, hard drive and keyboard)
- bx_unload_plugins() now deletes the devmodel object in non-plugin mode
- CMOS device now prints the time on exit in human-readable format
(TODO: find a way to call plugin_fini() for non-core devices if plugins are
disabled)
- set the device pointers back to stubs after unloading plugins
- added debug message "Exit" in all device plugin destructors
- cmos image options renamed
- new cmos option controls the usage of the RTC values from image
- report time0 in use after handling the cmos image stuff
- MSVC warning fixed
- unsupported shutdown status values no longer cause a panic
- definition of BX_NUM_CMOS_REGS moved from config.h to cmos.h (TODO: get rid
of this and implement 128 registers)
- indent mode fixed in modified section
- implementation of the UIP bit using a new timer handler. The one_second_timer()
function only sets the UIP bit and starts the UIP timer. The uip_timer()
function handles the date / time update, the alarm check and finally clears
the UIP bit.
- writing to control register A doesn't change the UIP bit
register values to a new timeval after date/time change. There is only a
BX_ERROR for now.
- added separate cases for the alarm time registers and a BX_DEBUG message that
reports the new alarm time.
Some devices already had one. Some I had to add an empty one.
I did a little cleaning of init() methods to make them more uniform
but generally I left them alone.
- I also put these exact diffs into a patch "patch.iodev-add-reset"
in case I want to revert these changes for some reason, for example
if they break an old patch. It should be deleted after a while.
To see the commit logs for this use either cvsweb or
cvs update -r BRANCH-io-cleanup and then 'cvs log' the various files.
In general this provides a generic interface for logging.
logfunctions:: is a class that is inherited by some classes, and also
. allocated as a standalone global called 'genlog'. All logging uses
. one of the ::info(), ::error(), ::ldebug(), ::panic() methods of this
. class through 'BX_INFO(), BX_ERROR(), BX_DEBUG(), BX_PANIC()' macros
. respectively.
.
. An example usage:
. BX_INFO(("Hello, World!\n"));
iofunctions:: is a class that is allocated once by default, and assigned
as the iofunction of each logfunctions instance. It is this class that
maintains the file descriptor and other output related code, at this
point using vfprintf(). At some future point, someone may choose to
write a gui 'console' for bochs to which messages would be redirected
simply by assigning a different iofunction class to the various logfunctions
objects.
More cleanup is coming, but this works for now. If you want to see alot
of debugging output, in main.cc, change onoff[LOGLEV_DEBUG]=0 to =1.
Comments, bugs, flames, to me: todd@fries.net
