Add a platform cleanup hook before starting the kernel. The openfirmware
and PXE loaders clean up their network stack there, while the other
loaders currently do nothing.
This closes ticket #6166
Change-Id: I34765892dfd9b2310c6af97c9ff7d414afae49e5
Reviewed-on: https://review.haiku-os.org/c/50
Reviewed-by: Adrien Destugues <[email protected]>
Reviewed-by: waddlesplash <[email protected]>
The NetStack.h header is currently not usable from C code. So while
net_stack_init() is called from platform code, we cannot call
net_stack_cleanup() from OpenFirmware's platform_start_kernel().
Thus call it directly from main()'s cleanup TODO, having assured
that the function is a no-op when no network stack was initialized.
Signed-off-by: Augustin Cavalier <[email protected]>
The UDP service does not own the UDP sockets. When shutting down,
inform the bound sockets that the service is no longer available.
This allows subsequent method calls to error out cleanly.
Signed-off-by: Augustin Cavalier <[email protected]>
Add a cleanup function net_stack_cleanup() that calls a new NetStack::ShutDown() method.
Make sure this method works even if the network stack was never initialized.
Signed-off-by: Augustin Cavalier <[email protected]>
It is currently done in both ~EthernetService() and ~NetStack().
Since NetStack is where it's added and where an explicit accessor function is provided,
choose that location.
Signed-off-by: Augustin Cavalier <[email protected]>
If no Haiku remote disk server is found, try an iSCSI discovery session on
the TFTP server, using the IP address from the boot command or OF options.
All available IPv4 targets are considered, Target Portal Groups are ignored.
RFC 4173 suggests a mechanism that avoids a discovery session by using DHCP;
that requires a compatibly configured DHCP server though and we wouldn't have
access to such data currently anyway. iSCSI is currently used as fallback,
and when it doesn't succeed it falls back to the menu as before.
Resolves ticket #5319.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38537 a95241bf-73f2-0310-859d-f6bbb57e9c96
Add support for both discovery and regular iSCSI sessions. Command and status
sequence numbers do differentiate between session and connection but only
one connection per session is currently supported.
Code is Big Endian for now, so compile it for ppc only.
Based on RFC 3720 ff. Tested against OpenSolaris 2009.06.
Resolves most of ticket #5319.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38536 a95241bf-73f2-0310-859d-f6bbb57e9c96
Don't enqueue a packet whose sequence number matches the first one in queue
or is between the next and the acknowledged on. This would lead to packet
duplication in the queue and leftovers after dequeuing.
Optimize calculation of acknowledgement number for out-of-order reception.
If a single missing packet arrived, this could have resulted in a timeout
depending on the window size.
In case the connection is closed, do dequeue remaining packets. This would
have caused a read error after a unidirectional FIN.
Prepare a dynamic window size calculation.
Do checksum calculation before attempting to trace packet contents. I have
seen a few and couldn't spot an error in the checksum calculation.
More fine-grained control over trace output.
Trace the Maximum Segment Size option if present; break after End of Options.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38535 a95241bf-73f2-0310-859d-f6bbb57e9c96
Pass the desired window size from the socket to the service.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38530 a95241bf-73f2-0310-859d-f6bbb57e9c96
Define structs for iSCSI messages, to be used by boot loader and kernel add-on.
For now it will refuse to compile for Little Endian systems (e.g., x86).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38528 a95241bf-73f2-0310-859d-f6bbb57e9c96
In some error conditions the packet would have been leaked.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38503 a95241bf-73f2-0310-859d-f6bbb57e9c96
Initialize the PRNG seed for the sequence number with real_time_clock_usecs()
rather than system_time(). The remaining uses of system_time() for timeout
calculation match UDP.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38482 a95241bf-73f2-0310-859d-f6bbb57e9c96
Partially revert r38465 and move the code from system_time() into a new
real_time_clock_usecs() function. The system_time() implementation was correct
in relying on of_milliseconds(), as pointed out by Axel. Add the correct
function for the desired functionality.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38481 a95241bf-73f2-0310-859d-f6bbb57e9c96
Fix coding style issues and bump copyright (although there's not much anyway).
No functional changes.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38480 a95241bf-73f2-0310-859d-f6bbb57e9c96
Modelled after UDP, add limited TCP support to the boot net stack. The socket
works by queuing received packets as well as sent packets that have not yet
been ACK'ed. Some known issues are documented, especially there's only limited
congestion control. I.e., we send immediately and in unlimited quantity, thus
its use should be restricted to local networks, and due to a fixed window size
there is potential for our socket being overrun with data packets before they
are read. Some corner cases like wrapping sequence numbers may cause a timeout.
The TCP implementation is based on Andrew S. Tanenbaum's "Computer Networks",
4th ed., as well as lecture notes from Prof. W. Effelsberg, the relevant RFCs
and Wikipedia. The pseudo-random number Galois LFSR used for the sequence
number was suggested by Endre Varga.
Since the code is unlikely to get much smaller, better merge it now so that
subsequent changes get easier to review. No platform actively uses TCP sockets
yet, and the receiving code has been reviewed for endianness issues and should
terminate okay after verifying the checksum if no sockets are open.
Based on a version tested with custom code (#5240) as well as with iSCSI.
Compile-tested boot_loader_openfirmware, pxehaiku-loader with gcc4 and
haiku_loader with gcc2. Closes ticket #5240.
Changes from #5240 proposed patch:
* Various bug fixes related to queuing, some memory leaks fixed.
* Never bump the sequence number when dequeuing a packet. It's done afterwards.
* Don't bump the sequence number again when resending the queue or ACK'ing.
* Aggressively ACK while waiting for packets.
* Don't queue sent ACK-only packets.
* More trace output, esp. for queue inspection.
* Adapted use of TCP header flags to r38434.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38472 a95241bf-73f2-0310-859d-f6bbb57e9c96
system_time() was based on of_milliseconds(), which returns the number of
milliseconds since power-on. This would produce very similar or identical
results for subsequent boots due to limited clock resolution; therefore it
was unsuited as PRNG seed, e.g., for TCP ports.
Try to inquire the RTC device node with get-time to return an improved time
value, if possible.
Closes ticket #6061.
Changes from proposed patch:
* Obtain the RTC handle once and reuse it for each inquiry.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38465 a95241bf-73f2-0310-859d-f6bbb57e9c96
Fix style issue, pointed out by Axel.
No functional changes.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38456 a95241bf-73f2-0310-859d-f6bbb57e9c96
Add protocol number and struct for TCP header.
First minuscule part of ticket #5240.
Checked that pxehaiku_loader still compiles, too.
Changes from proposed patch:
* Simplify struct by merging flags into one 8-bit field.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38434 a95241bf-73f2-0310-859d-f6bbb57e9c96
Based on a SCSI Command Reference Manual by Seagate and Wikipedia.
Tested against OpenSolaris iSCSI target on ppc. It choked on the READ (12) opcode.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38425 a95241bf-73f2-0310-859d-f6bbb57e9c96
The future cookie was leaked in three of four error scenarios.
Backported from grackle.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38345 a95241bf-73f2-0310-859d-f6bbb57e9c96
Based on Motorola MPC106 User Manual and NetBSD's
src/sys/arch/macppc/pci/grackle.c 1.11. The MPC106 apparently deviates by
requiring the E (enable) bit set to 1 for config read/write to work.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38344 a95241bf-73f2-0310-859d-f6bbb57e9c96
A grackle-compatible Motorola MPC106 PCI controller can be found in the
PowerMac G3 (ticket #6247). For now just let the user know.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38323 a95241bf-73f2-0310-859d-f6bbb57e9c96
Read the EDID from OF and pass it on to the kernel.
Last part of ticket #6105.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38307 a95241bf-73f2-0310-859d-f6bbb57e9c96
Use the OF "screen" device alias to query frame buffer properties. Postpone
the opening of the device as far as possible as it erases the screen output.
Initialize the kernel arguments and hook into generic code to display the
splash screen.
Like on x86, the frame buffer is enabled by default. To disable it, either
press ESC during early boot or, for debugging, make sure in start.cpp that
platform_boot_options() has the BOOT_OPTION_DEBUG_OUTPUT flag set.
Resolves ticket #6105.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38306 a95241bf-73f2-0310-859d-f6bbb57e9c96
Avoid writing to stdout if the frame buffer is enabled.
Part of ticket #6105.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38305 a95241bf-73f2-0310-859d-f6bbb57e9c96
Introduce a non-blocking function for checking keyboard input, and refactor
existing code to share the escape key translations. The comment that key-up and
key-down result in a zero char is confirmed to apply.
If the space bar is pressed, enter the boot menu. If the escape key is pressed,
disable the frame buffer and keep showing the usual debug output. Apparently
the key press must come after console/keyboard initialization but before the
"Welcome to the Haiku bootloader!" line.
Closes ticket #6140.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38304 a95241bf-73f2-0310-859d-f6bbb57e9c96
The frame buffer will need more generic sources than just the text menu.
Adapted from bios_ia32 platform.
Part of ticket #6105.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38301 a95241bf-73f2-0310-859d-f6bbb57e9c96
Debug output calls OpenFirmware methods writing text into the frame buffer.
Depending on the screen mode this leads to garbled display or parts of the
screen overwritten, so mute the output during frame buffer use for now.
A better solution would be to redirect the output elsewhere (e.g., #6168),
so that ConsoleHandle::WriteAt() is not called for the problematic handle
in the first place. That does not work inside the kernel yet.
Part of ticket #6105.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38300 a95241bf-73f2-0310-859d-f6bbb57e9c96
Initialize the frame buffer as not enabled. Add checks for this condition
and for the debug boot option to the video functions.
Code is adapted from bios_ia32 platform. Part of ticket #6105.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38299 a95241bf-73f2-0310-859d-f6bbb57e9c96
This is a leftover from the r25913 build fix. It used to be an out parameter.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38294 a95241bf-73f2-0310-859d-f6bbb57e9c96
This remedies "Current thread pointer [...] is an address we can't read from."
on entering KDL. Closes ticket #6163.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38293 a95241bf-73f2-0310-859d-f6bbb57e9c96
Let setting the memory type to B_MTR_UC succeed.
Add initial support for those memory types in the translation map,
pointed out by Ingo in ticket #5193.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38292 a95241bf-73f2-0310-859d-f6bbb57e9c96
Revert r36886 and fix compilation of insert_virtual_range_to_keep().
The use of void* vs. addr_t matches the surrounding boot loader code
but should probably be revised in favour of addr_t.
create_area() in the kernel wrongly assumed a RAM-backed address range,
which was destined to fail since ranges below the kernel address space
were ignored anyway. Use vm_map_physical_memory() instead.
This fixes a hang once the frame buffer and other resources used by OF
get unmapped. Closes ticket #5193 again.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38291 a95241bf-73f2-0310-859d-f6bbb57e9c96
Fix some more #pragma export warnings treated as errors on ppc.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38258 a95241bf-73f2-0310-859d-f6bbb57e9c96
PowerPC was special-cased to not include posix/regex.h but a local copy.
This caused a warning for undefining __STDC__ and a pointer signedness error.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38257 a95241bf-73f2-0310-859d-f6bbb57e9c96
#pragma export on
and
#pragma export off
result in warnings on ppc, which are now treated as errors.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38256 a95241bf-73f2-0310-859d-f6bbb57e9c96
As an interim solution for fixing the build,
download an ICU package from Haiku Ports.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38254 a95241bf-73f2-0310-859d-f6bbb57e9c96
Cast uint32* arguments for atomic_or() to int32*.
Resolves warnings treated as errors for Big Endian targets.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38128 a95241bf-73f2-0310-859d-f6bbb57e9c96
Cast further pointer arguments for B_HOST_TO_LENDIAN_INT32 to uint32.
Resolves warnings treated as errors for Big Endian targets.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38127 a95241bf-73f2-0310-859d-f6bbb57e9c96
Cast pointer arguments for B_HOST_TO_LENDIAN_INT32 to uint32.
Resolves warnings treated as errors for Big Endian targets.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38114 a95241bf-73f2-0310-859d-f6bbb57e9c96
Cast return value of __swap_int32 to uint32_t, as suggested by Ingo.
Remedies uint32 %08x warnings treated as errors in the Big Endian code path.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38106 a95241bf-73f2-0310-859d-f6bbb57e9c96
Add Jam rule SuppressOptionalHaikuImagePackages as a mechanism to keep
packages from being installed.
Extend the UserBuildConfig.ReadMe document to cover the new command.
Closes ticket #6260.
Changes from v1:
* Simplified IsOptionalHaikuImagePackageAdded as suggested by Ingo.
* Added example as documentation.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38100 a95241bf-73f2-0310-859d-f6bbb57e9c96
The updated glibc regex sources use source-local includes, such as
"regex_internal.h", so add the regex directory to the search path.
Define __USE_GNU as in r38031 to define GNU extensions used and
HAVE_STDBOOL_H to avoid build failure due to type bool.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38097 a95241bf-73f2-0310-859d-f6bbb57e9c96
Cygwin and Solaris rely on Haiku supplying a glibc-compatible regex.h header.
Before r38031, Solaris was able to reuse cygwin/regex.h successfully. Import
the latest stable 2.11 posix/regex.h to avoid local trickery and to silence
some RE_* errors.
It is very likely necessary to update cygwin/regex.h as well, but I don't have
a test setup to verify.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38096 a95241bf-73f2-0310-859d-f6bbb57e9c96
Suggested by Ingo in ticket #6139. This shortens the stack trace and makes it
independent of whether called implicitly on KDL entry or manually from the
kernel debugger prompt.
For a kernel panic, "panic" is now the top-most frame printed.
Closes ticket #6160.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@37282 a95241bf-73f2-0310-859d-f6bbb57e9c96
Suggested by Ingo in ticket #6139. Code is adapted from x86.
Note that on ppc64 GPR1 needs to be 64-bit, thus the choice of addr_t.
Resolves part of ticket #6160.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@37281 a95241bf-73f2-0310-859d-f6bbb57e9c96
On QEMU's emulation of a PowerPC Old World Mac (g3beige), determine_machine()
returns without setting the machine type to unknown, since the root device
does not provide a device_type property.
Move the variable assignment up a few lines to remedy this.
Closes ticket #6123.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@37280 a95241bf-73f2-0310-859d-f6bbb57e9c96
The only user of of_milliseconds(), the boot loader's system_time(),
checks for an error that would never be signalled since the zero-initialized
number of milliseconds would always be returned.
Return OF_FAILED instead to allow explicit error handling in that case.
While this is a behavioral change, it is not yet a functional change,
for system_time()'s implementation translates OF_FAILED back to zero currently.
Resolves part of ticket #6061.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@37279 a95241bf-73f2-0310-859d-f6bbb57e9c96
In r33670 the svn:eol-style property was dropped, which took care of
locally converting the line endings to the user's native style.
While most files use Unix-style LF line endings, some files have
Windows-style CR LF line endings.
Assure that the following r37262 directories use Unix-style line endings:
src/system/boot/
src/system/boot/arch/
src/system/boot/arch/ppc/
src/system/boot/loader/
src/system/boot/loader/net/
src/system/boot/platform/
src/system/boot/platform/openfirmware/
src/system/boot/platform/openfirmware/arch/
src/system/boot/platform/openfirmware/arch/ppc/
src/system/kernel/
src/system/kernel/arch/
src/system/kernel/arch/ppc/
src/system/kernel/platform/
src/system/kernel/platform/openfirmware/
headers/private/kernel/
headers/private/kernel/arch/
headers/private/kernel/arch/ppc/
headers/private/kernel/platform/
headers/private/kernel/platform/openfirmware/
headers/private/kernel/boot/
headers/private/kernel/boot/net/
headers/private/kernel/boot/platform/
headers/private/kernel/boot/platform/openfirmware/
This avoids patches containing irrelevant lines unintentionally converted.
No functional changes.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@37265 a95241bf-73f2-0310-859d-f6bbb57e9c96
Andreas Faerber