This is largely the work from the projects/uefi branch, with some
additional refinements. This is derived from (and replaces) the
original i386 efi implementation; i386 support will be restored later.
Specific revisions of note from projects/uefi:
r247380:
Adjust our load device when we boot from CD under UEFI.
The process for booting from a CD under UEFI involves adding a FAT
filesystem containing your loader code as an El Torito boot image.
When UEFI detects this, it provides a block IO instance that points at
the FAT filesystem as a child of the device that represents the CD
itself. The problem being that the CD device is flagged as a "raw
device" while the boot image is flagged as a "logical partition". The
existing EFI partition code only looks for logical partitions and so
the CD filesystem was rendered invisible.
To fix this, check the type of each block IO device. If it's found to
be a CD, and thus an El Torito boot image, look up its parent device
and add that instead so that the loader will then load the kernel from
the CD filesystem. This is done by using the handle for the boot
filesystem as an alias.
Something similar to this will be required for booting from other
media as well as the loader will live in the EFI system partition, not
on the partition containing the kernel.
r246231:
Add necessary code to hand off from loader to an amd64 kernel.
r246335:
Grab the EFI memory map and store it as module metadata on the kernel.
This is the same approach used to provide the BIOS SMAP to the kernel.
r246336:
Pass the ACPI table metadata via hints so the kernel ACPI code can
find them.
r246608:
Rework copy routines to ensure we always use memory allocated via EFI.
The previous code assumed it could copy wherever it liked. This is not
the case. The approach taken by this code is pretty ham-fisted in that
it simply allocates a large (32MB) buffer area and stages into that,
then copies the whole area into place when it's time to execute. A more
elegant solution could be used but this works for now.
r247214:
Fix a number of problems preventing proper handover to the kernel.
There were two issues at play here. Firstly, there was nothing
preventing UEFI from placing the loader code above 1GB in RAM. This
meant that when we switched in the page tables the kernel expects to
be running on, we are suddenly unmapped and things no longer work. We
solve this by making our trampoline code not dependent on being at any
given position and simply copying it to a "safe" location before
calling it.
Secondly, UEFI could allocate our stack wherever it wants. As it
happened on my PC, that was right where I was copying the kernel to.
This did not cause happiness. The solution to this was to also switch
to a temporary stack in a safe location before performing the final
copy of the loaded kernel.
r246231:
Add necessary code to hand off from loader to an amd64 kernel.
r246335:
Grab the EFI memory map and store it as module metadata on the kernel.
This is the same approach used to provide the BIOS SMAP to the kernel.
r246336:
Pass the ACPI table metadata via hints so the kernel ACPI code can
find them.
r246608:
Rework copy routines to ensure we always use memory allocated via EFI.
The previous code assumed it could copy wherever it liked. This is not
the case. The approach taken by this code is pretty ham-fisted in that
it simply allocates a large (32MB) buffer area and stages into that,
then copies the whole area into place when it's time to execute. A more
elegant solution could be used but this works for now.
r247214:
Fix a number of problems preventing proper handover to the kernel.
There were two issues at play here. Firstly, there was nothing
preventing UEFI from placing the loader code above 1GB in RAM. This
meant that when we switched in the page tables the kernel expects to
be running on, we are suddenly unmapped and things no longer work. We
solve this by making our trampoline code not dependent on being at any
given position and simply copying it to a "safe" location before
calling it.
Secondly, UEFI could allocate our stack wherever it wants. As it
happened on my PC, that was right where I was copying the kernel to.
This did not cause happiness. The solution to this was to also switch
to a temporary stack in a safe location before performing the final
copy of the loaded kernel.
r247216:
Use the UEFI Graphics Output Protocol to get the parameters of the
framebuffer.
Sponsored by: The FreeBSD Foundation
r247216:
Add the ability for a device to have an "alias" handle.
r247379:
Fix network device registration.
r247380:
Adjust our load device when we boot from CD under UEFI.
The process for booting from a CD under UEFI involves adding a FAT
filesystem containing your loader code as an El Torito boot image.
When UEFI detects this, it provides a block IO instance that points
at the FAT filesystem as a child of the device that represents the CD
itself. The problem being that the CD device is flagged as a "raw
device" while the boot image is flagged as a "logical partition".
The existing EFI partition code only looks for logical partitions and
so the CD filesystem was rendered invisible.
To fix this, check the type of each block IO device. If it's found to
be a CD, and thus an El Torito boot image, look up its parent device
and add that instead so that the loader will then load the kernel from
the CD filesystem. This is done by using the handle for the boot
filesystem as an alias.
Something similar to this will be required for booting from other media
as well as the loader will live in the EFI system partition, not on the
partition containing the kernel.
r247381:
Remove a scatalogical debug printf that crept in.
.. so that consistent compilation algorithms are used for both
architectures as in practice the binaries are expected to be
interchangeable (for time being).
Previously i386 used default setting which were equivalent to
-march=i486 -mtune=generic.
The only difference is using smaller but slower "leave" instructions.
Discussed with: jhb, dim
MFC after: 29 days
code that is used to construct a loader (e.g. libstand, ficl, etc).
There is such a thing as a 64-bit EFI application, but it's not
as standard as 32-bit is. Let's make the 32-bit functional (as in
we can load and actualy boot a kernel) before solving the 64-bit
loader problem.
Add support for Pre-Boot Virtual Memory (PBVM) to the loader.
PBVM allows us to link the kernel at a fixed virtual address without
having to make any assumptions about the physical memory layout. On
the SGI Altix 350 for example, there's no usuable physical memory
below 192GB. Also, the PBVM allows us to control better where we're
going to physically load the kernel and its modules so that we can
make sure we load the kernel in memory that's close to the BSP.
The PBVM is managed by a simple page table. The minimum size of the
page table is 4KB (EFI page size) and the maximum is currently set
to 1MB. A page in the PBVM is 64KB, as that's the maximum alignment
one can specify in a linker script. The bottom line is that PBVM is
between 64KB and 8GB in size.
The loader maps the PBVM page table at a fixed virtual address and
using a single translations. The PBVM itself is also mapped using a
single translation for a maximum of 32MB.
While here, increase the heap in the EFI loader from 512KB to 2MB
and set the stage for supporting relocatable modules.
as this only allows us to access file systems that EFI knows about.
With a loader that can only use EFI-supported file systems, we're
forced to put /boot on the EFI system partition. This is suboptimal
in the following ways:
1. With /boot a symlink to /efi/boot, mergemaster complains about
the mismatch and there's no quick solution.
2. The EFI loader can only boot a single version of FreeBSD. There's
no way to install multiple versions of FreeBSD and select one
at the loader prompt.
3. ZFS maintains /boot/zfs/zpool.cache and with /boot a symlink we
end up with the file on a MSDOS file system. ZFS does not have
proper handling of file systems that are under Giant.
Implement a disk device based on the block I/O protocol instead and
pull in file system code from libstand. The disk devices are really
the partitions that EFI knows about.
This change is backward compatible.
MFC after: 1 week
- It is opt-out for now so as to give it maximum testing, but it may be
turned opt-in for stable branches depending on the consensus. You
can turn it off with WITHOUT_SSP.
- WITHOUT_SSP was previously used to disable the build of GNU libssp.
It is harmless to steal the knob as SSP symbols have been provided
by libc for a long time, GNU libssp should not have been much used.
- SSP is disabled in a few corners such as system bootstrap programs
(sys/boot), process bootstrap code (rtld, csu) and SSP symbols themselves.
- It should be safe to use -fstack-protector-all to build world, however
libc will be automatically downgraded to -fstack-protector because it
breaks rtld otherwise.
- This option is unavailable on ia64.
Enable GCC stack protection (aka Propolice) for kernel:
- It is opt-out for now so as to give it maximum testing.
- Do not compile your kernel with -fstack-protector-all, it won't work.
Submitted by: Jeremie Le Hen <jeremie@le-hen.org>
1. Make libefi portable by removing ia64 specific code and build
it on i386 and amd64 by default to prevent regressions. These
changes include fixes and improvements over previous code to
establish or improve APIs where none existed or when the amount
of kluging was unacceptably high.
2. Increase the amount of sharing between the efi and ski loaders
to improve maintainability of the loaders and simplify making
changes to the loader-kernel handshaking in the future.
The version of the efi and ski loaders are now both changed to 1.2
as user visible improvements and changes have been made.
device (kind) specific unit field to the common field. This change
allows a future version of libefi to work without requiring anything
more than what is defined in struct devdesc and as such makes it
possible to compile said version of libefi for different platforms
without requiring that those platforms have identical derivatives
of struct devdesc.
place.
This moves the dependency on GCC's and other compiler's features into
the central sys/cdefs.h file, while the individual source files can
then refer to #ifdef __COMPILER_FEATURE_FOO where they by now used to
refer to #if __GNUC__ > 3.1415 && __BARC__ <= 42.
By now, GCC and ICC (the Intel compiler) have been actively tested on
IA32 platforms by netchild. Extension to other compilers is supposed
to be possible, of course.
Submitted by: netchild
Reviewed by: various developers on arch@, some time ago
we construct the EFI image. It doesn't seem to actually end up
in the EFI image, AFAICT.
o Replace .quad, .long and .short with data8, data4 and data2 resp.
The former are gnuisms.
o Redefine _start_plabel as a data16 with @iplt(_start) as its
value. This is the preferred way to create user PLT entries.
binutils 2.15. The linker now creates a .rela.dyn section for
dynamic relocations, while our script created a .rela section.
Likewise, we copied the .rela section to the EFI image, but not
the .rela.dyn section. The fix is to rename .rela to .rela.dyn
in the linker script so that all relocations end up in the same
section again. This we copy into the EFI image.
EFI file system. When booting from a CD and there's already an EFI
system partition on the disk, setting the current device to unit 0
will select the harddisk. This invariably breaks installing FreeBSD
when other operating systems have been installed before.
We obviously want to do the same when we're booting over the network.
Maybe later.
Based on a patch (from memory) from: arun
things over floppy size limits, I can exclude it for release builds or
something like that. Most of the changes are to get the load_elf.c file
into a seperate elf32_ or elf64_ namespace so that you can have two
ELF loaders present at once. Note that for 64 bit kernels, it actually
starts up the kernel already in 64 bit mode with paging enabled. This
is really easy because we have a known minimum feature set.
Of note is that for amd64, we have to pass in the bios int 15 0xe821
memory map because once in long mode, you absolutely cannot make VM86
calls. amd64 does not use 'struct bootinfo' at all. It is a pure loader
metadata startup, just like sparc64 and powerpc. Much of the
infrastructure to support this was adapted from sparc64.
introduce a preprocessor define for it. The larger block size
significantly speeds up the loading of the kernel.
Submitted by: Arun Sharma <arun.sharma@intel.com>
NULL is passed. The address of the HCDP table can be found by
iterating over the configuration tables in the EFI system table.
To avoid more duplication, a function can be called with the GUID
of interest. The function will do the scanning. Use the function
in all places where we iterate over the configuration tables in
an attempt to find a specific one.
Bump the loader version number as the result of this.
Approved by: re (blanket)
accept load options (=command line options).
The call graph changes from *entry*->efi_main->efi_init, where
efi_main is the EFI equivalent of main to *entry*->efi_main->main,
where main is what you'd expect. efi_main now is what efi_init was.
The prototype of main follows that of C. The first argument is argc
and the second is argv. There is no third argument.
Allocation of heap pages is now handled by the EFI library and it
now deallocates the pages when main() returns or when exit() is
called. This allows us to safely return to the boot manager (or
EFI shell) without leaks. EFI applications are responsible to free
all memory themselves.
Handling of the load options is a bit tricky. There are either no
load options, load options in ASCII or load options in Unicode.
The EFI library will translate the ASCII options to Unicode options
as to simplify user code. Since the load options are passed as a
single string (if present) and main() accepts argc and argv, the
startup code also has to split the string into words and build the
argv vector. Here the trickiness starts. When the loader is started
from the EFI shell, argv[0] will automaticly load the program name.
In all other cases (ie through the boot manager), this is not the
case. Unfortunately, there's no trivial way to check. Hence, a
set of conditions is checked to determine if we need to fill in
argv[0] ourselves or not. This checking is not perfect. There are
known cases where it fails to do the right thing. The logic works
for most expected cases, though. This includes the case where no
options are given.
Approved by: re (blanket)
a boot option. When the timer expires the machine is rebooted.
Disable the watchdog timer for 2 reasons:
o We're an interactive program. We cannot guarantee that we've
booted the kernel in the time available to us. There have been
situations where netbooting the right kernel took 2 tries and
more time than given. Not to speak of the normal behaviour to
have the loader sitting at the prompt while the user is off
doing other things (such as figuring out what to type next ;-)
o We may not boot a kernel at all. We may exit as the result of
the user typing quit (assuming it took less than 5 minutes to
type it :-). It is documented that loaders should have disabled
the watchdog timer if they return to the boot manager. Not doing
so would cause a reboot while in the boot manager. This appears
to be harmless, besides of course the actual reboot.
Approved by: re (weisse karte)
the signaled state of the apropriate event. As a side-effect of
checking the event, it's signaled state is cleared if it was set.
In efi_cons_getchar we used to wait for the apropriate event to be
signaled before reading a character. This however does not work if
we poll before reading the characteri, such as during autoboot. On
a more compliant EFI implementation this resulted in the behaviour
that hitting a key during autoboot would stop the countdown, but
would then wait for a new character to arrive instead of reading
the already pending key that stopped the countdown.
The correct behaviour for efi_cons_getchar is to try to read a key
and if none is pending, to wait for the apropriate event to signal
the arrival of a new key.
Note that with the previous behaviour, the second key would determine
how the autoboot was interrupted. This would indicate that the first
key got lost. This indicates that EFI does not necessarily maintain
a queue of pending keys. FWIW...
Approved by: re (carte blanche)
French corrected by: various people :-)
Previous kernels unwantingly depended on this mapping, but as
of version 1.123 of src/sys/ia64/ia64/machdep.c this dependency
has been removed. Consequently, one has to update the kernel
before updating the loader. The documented/recommended upgrade
will suffice in this case.
Due to a visible (from the kernels point of view) change in
behaviour, bump the loader version number from 0.3 to 1.0.
Approved by: re (carte blanc)
o Show the contents of the AP wakeup descriptor when dumping SAL
information.
o Increase S/N ratio when listing the itr and dtr. Only show valid
mappings and give the total number of TRs.
Approved by: re (blanket)
pages are 4KB.
o As a second order fix, don't assume we have enough space
after the bootinfo block left in a page to hold the memory
map.
o A third order fix as that we removed the assumption that a
bootinfo block fits in a single 8KB page.
PR: ia64/39415
submitted by: Espen Skoglund <esk@ira.uka.de>
Bug#1: The GetStatus() function returns radically different pointers that
do not match any packets we transmitted. I think it might be pointing to
a copy of the packet or something. Since we do not transmit more than
one packet at a time, just wait for "anything".
Bug#2: The Receive() function takes a pointer and a length. However, it
either ignores the length or otherwise does bad things and writes outside
of ptr[0] through ptr[len-1]. This is bad and causes massive stack
corruption for us since we are receiving packets into small buffers on
the stack. Instead, Receive() into a large enough buffer and bcopy the
data to the requested area.
- Don't include ia64_cpu.h and cpu.h
- Guard definitions by _NO_NAMESPACE_POLLUTION
- Move definition of KERNBASE to vmparam.h
o Move definitions of IA64_RR_{BASE|MASK} to vmparam.h
o Move definitions of IA64_PHYS_TO_RR{6|7} to vmparam.h
o While here, remove some left-over Alpha references.
o We don't expect the PLT relocations to follow the .rela section
anymore. We still assume that PLT relocations are long formed,
o Document register usage,
o Improve ILP,
o Fix the FPTR relocation by creating unique OPDs per function.
Comparing functions is valid now,
o The IPLT relocation naturally handles the addend. Deal with it.
We ignore the addend for FPTR relocations for now. It's not at
all clear what it means anyway.
Fix ABI misinterpretation:
o For Elf_Rela relocations, the addend is explicit and should not
be loaded from the memory address we're relocating. Only do that
for Elf_Rel relocations (ie the short form).
o DIR64LSB is not the same as REL64LSB. DIR64LSB applies to a
symbol (S+A), whereas REL64LSB applies to the base address (BD+A),
the S_IFREG bit for regular files. This caused the path search code to
skip it when it finally did find the kernel (after the common/module.c
buffer overrun bug was fixed)
detects and uses the gas section merge support. As a result, a whole bunch
of new sections arrive, including .rodata.str1.8, which was not included
in our custom ldscript.ia64. The result was a loader binary that EFI
rejected.
While here, collect the loader shell commands linker set and include it
in the data area rather than having its own section.
/boot/loader.efi was the last holdout for having a 100% self built ia64
system.
register r8. We continue to write the bootinfo block at the same
hardwired address, because the kernel still expects it there.
It is expected that future kernels use register r8 to get to the
bootinfo block and don't depend on the hardwired address anymore.
Bump the loader version once again due to the interface change.
only care if it's network or not at this time. If we're loaded from
the network, we set currdev (=loaddev) so that the kernel is loaded
from the network as well. In all other cases we initialize to disk.
This makes netbooting more convenient and can easily be enhanced to
do more elaborate checking.
Most significantly (from an interfacing point of view) is the
support for the FPSWA pointer passing. Even though that was added
4 months ago, it's probably not a bad idea to bump the version
number to reflect this.
o Query the state field of the protocol mode to determine whether
we need to start and/or initialize the protocol. When we're
loaded across the network, the protocol has already been started
and is already initialized. When no networking has happened yet,
we have to start and initialize the protocol ourselves.
o After initialization, we have to set the receive filters. Not
doing this results in a deaf interface. We set the unicast and
broadcast filters. Multicast may not be supported. This specific
change fixes the problem we had that we could not netboot if
the loader was started from the EFI shell.
o To help future debugging, add a function that dumps the current
mode of the interface. It's conditional on EFINET_DEBUG.
o To help in runtime problems, emit a diagnostic message when we
could not initialize the protocol properly.
an efi_devdesc structure. When we're netbooting, f->f_devdata holds
the address of the network socket variable. Dereferencing this caused
some very unpredictable behaviour, including proper functioning.
So, as a sanity check, we first make sure f->f_dev points to our
own devsw. If not, the open will fail before we use f->f_devdata.
This solves the netboot hangs I invariably got whenever I used the
latest toolchain to compile the EFI loader.
layer to signal transmission of the packet. This resolves the
problem I'm seeing that an immediate call to net->Receive
after calling net->Transmit returns EFI_DEVICE_ERROR. This
condition seems to be sufficiently persistent that BOOTP and
RARP fail.
o While here, unify all functions to have 'nif' defined. Some
have it as arguments. The others now have them as locals. We
now always get the protocol interface by using the 'nif' var.
The current status of netbooting is that even though we now reliably
have BOOTP working (again), opening a file (ie loading a kernel)
across the network causes the loader to hang. I'm working on that now.
o Make <stdint.h> a symbolic link to <sys/stdint.h>.
o Move most of <sys/inttypes.h> into <sys/stdint.h>, as per C99.
o Remove <sys/inttypes.h>.
o Adjust includes in sys/types.h and boot/efi/include/ia64/efibind.h
to reflect new location of integer types in <sys/stdint.h>.
o Remove previously symbolicly linked <inttypes.h>, instead create a
new file.
o Add MD headers <machine/_inttypes.h> from NetBSD.
o Include <sys/stdint.h> in <inttypes.h>, as required by C99; and
include <machine/_inttypes.h> in <inttypes.h>, to fill in the
remaining requirements for <inttypes.h>.
o Add additional integer types in <machine/ansi.h> and
<machine/limits.h> which are included via <sys/stdint.h>.
Partially obtain from: NetBSD
Tested on: alpha, i386
Discussed on: freebsd-standards@bostonradio.org
Reviewed by: bde, fenner, obrien, wollman
(I'll be we know which compiler and platform they developed this on...)
Minimally change them to C89 comments to make GCC happy. (this is kinda funny
as the file has piece derived from FreeBDS 3.2)
Also fix FreeBSD id style.
page of the image to load section headers and if we let the text section
start at zero, it corrupts the section table when its loaded. With this
change, the loader gets as far as the 'ok' prompt.
