Defalut font color on PC console seems to be light-gray; this is
what user also gets in rescue prompt and what is defined as
GRUB_TERM_DEFAULT_NORMAL_COLOR. But normal.mod defaults to white.
This makes unpleasant visual effect as colors are changed after kernel
is booted.
Use the same color eveywhere for consistency and default to light-gray
as this is also what at least Linux kernel is using by default.
This lets us cope with block device drivers that don't implement
HDIO_GETGEO. Fixes Ubuntu bug #1237519.
* grub-core/osdep/linux/hostdisk.c (sysfs_partition_path): New
function.
(sysfs_partition_start): Likewise.
(grub_util_find_partition_start_os): Try sysfs_partition_start
before HDIO_GETGEO.
This was the only instance of "library" in core config. A bug was
reported that -fno-stack-protector wasn't passed to it. Instead of
figuring out why it failed just remove this construction used
needlessly.
Add a new timeout_style environment variable and a corresponding
GRUB_TIMEOUT_STYLE configuration key for grub-mkconfig. This
controls hidden-timeout handling more simply than the previous
arrangements, and pressing any hotkeys associated with menu entries
during the hidden timeout will now boot the corresponding menu entry
immediately.
GRUB_HIDDEN_TIMEOUT=<non-empty> + GRUB_TIMEOUT=<non-zero> now
generates a warning, and if it shows the menu it will do so as if
the second timeout were not present. Other combinations are
translated into reasonable equivalents.
A bit tricky because this function has to continue to work without
heap for short strings. Fixing prealloc to 32 arguments is reasonable
but make all stack references use 32-bit offset rather than 8-bit one.
So split va_args preparsing to separate function and put the prealloc
into the caller.
references to mdadm from otherwise generic code.
(grub_util_exec_pipe): Likewise.
(grub_util_exec_pipe_stderr): Likewise.
* grub-core/osdep/unix/getroot.c (grub_util_pull_lvm_by_command):
This function calls vgs, not mdadm; adjust variable names
accordingly.
Add grub_util_disable_fd_syncs call to turn grub_util_fd_sync calls into
no-ops, and use it in programs that copy files but do not need to take
special care to sync writes (grub-mknetdir, grub-rescue,
grub-mkstandalone).
On my laptop, this reduces partmap_test's runtime from 1236 seconds to
204 seconds.
Similar to check for target linking format, also check for efiemu64
instead of hardcoding -melf_x86_64. This fixes compilation on *BSD
variants. We cannot easily reuse main target check because platforms
are different (main target is 32 bit and efiemu64 - 64 bit).
This commit adds EFIEMU64_LINK_FORMAT that contains detected
link option and is used in efiemu64.o linking instead of hardcoded
value.
Reported-By: Beeblebrox <zaphod@berentweb.com>
Previously we misaligned stack by 8 in startup.S and compensated
for it in callwrap.S. According to ABI docs (EFI and sysv amd64)
right behaviour is to align stack in startup.S and keep it aligned
in callwrap.S. startup.S part was committed few commits before. This
takes care of callwrap.S.
Reported by: Gary Lin.
16-byte boundary, as required by the x86-64 ABI, before calling
grub_main. In some cases, GCC emits code that assumes this
alignment, which crashes if not aligned. The EFI firmware is also
entitled to assume that stack alignment without checking.
pointer to 32K. This is the size of cache element which is the most
common allocation >1K. This way the pointer is always around blocks
of 32K and so we keep performance while decreasing fragmentation.
the function of these files exceeds what can be sanely handled in shell
in posix-comaptible way. Also writing it in C extends the functionality
to non-UNIX-like OS and minimal environments.
We have only 92K of stack and using over 4K per frame is wasteful
* grub-core/script/yylex.l (yyalloc), (yyfree), (yyrealloc): Declare
as macros so that compiler would remove useless structure on stack.
Better solution would be to fix flex not to put this structure on
the stack but flex is external program.
We have only 92K of stack and using over 4K per frame is wasteful
* grub-core/commands/verify.c (grub_load_public_key): Allocate on heap
rather than stack.
(grub_verify_signature_real): Likewise.
We have only 92K of stack and using over 4K per frame is wasteful
* grub-core/disk/mdraid_linux.c (grub_mdraid_detect): Allocate on heap
rather than stack.
We have only 92K of stack and using over 4K per frame is wasteful
* grub-core/fs/btrfs.c (grub_btrfs_lzo_decompress): Allocate on heap
rather than stack.
We have only 92K of stack and using over 4K per frame is wasteful
* grub-core/fs/jfs.c (getblk): Allocate on heap rather than on
stack. Note: this function is recursive.
(grub_jfs_read_inode): Read only part we care about.
Commit c9cd02c broke the u-boot syscall API for va_args that spill over
to the stack, causing the disk support to stop working. This patch
resolves the problem, while keeping the new, cleaner transition_space
handling.
On upcoming arm64 port libgcc ctz* are not usable in standalone
environment. Since we need ctz* for this case and implementation is
in C we may as well use it on all concerned platforms.