Based on top of 38 release and patches added on top of it up to September
2022, all of which are fixes that should have no functional changes, but are
useful for porting, although "LLD: fix detection and remove not needed
workarounds" was reverted to make build work with lld 13.0.
Debug logging is off as it is on Android systems and for the same reason:
the implementation relies on fopencookie() GNU extension.
There is an extra mode_t parameter in efi_set_variable() as well as
efi_chmod_variable(), which are specific to pseudo file-systems of Linux.
The parameter is ignored and the function succeed while doing nothing.
This provides means for manipulating data structures that describe device paths. Should work fine as it's not system-specific.
Should work properly, it's about time converting and formatting.
Should work as expected. It's another data formatting API for things like
Boot0001 variables.
efi_generate_ipv4_device_path() doesn't work. Needs platform-specific
implementation for querying things like MAC address.
efi_generate_file_device_path() should work only if abbreviated device path is
requested and implementation can handle disk/partition devices on the system
(find parent device from partition device, see more on this below). Drives
without a partition table might have some issues, needs testing.
efi_generate_file_device_path_from_esp() is also restricted to abbreviated
device paths and might not work if implementation can't derive partition device
name from a combination of disk device name and partition number.
When non-abbreviated device path is requested, the invocation fails with
EINVAL error.
Should work. It's about manipulating data in files with no special requirements on the environment.
These use device naming scheme that's easy to parse and thus shouldn't have much issues.
Assumption is that wedges are enabled (dk driver) and are used.
Conventional partition names (i.e., "slices" like sd3f) seem to lack an API
for querying partition number and their support is thus limited. Disk label
obtained from DIOCGDINFO ioctl() has empty partition table.
Might need to add a new ioctl() for partitions to make this work reliably.
Need to know partition number in some way. Getting just partition's offset on a
disk might take less work in the kernel, will then need to look it up in
MBR/GPT, which is done by the code anyway to fetch other information.
No wedges here, only slices, but DIOCGPDINFO ioctl() (note that "P" which
NetBSD doesn't have, not a typo) provides offsets at least for some drives
(weirdly, root drive seems missing), allowing the translation between partition
number and device name.
A nice thing to do would be to group more of platform specific things to get rid
of #ifdefs. Also dev_probes array (currently in linux.c) could be changed
to be constructed by the linker to remove declaration dependencies between files
and allow managing available probes by the build system by including only
certain source files.
Supporting unabbreviated device paths would take querying logical topology of drives (ACPI, PCI, USB, etc.).
Partition number is base 1, partition number 0 is the drive itself.
FreeBSD has only p in device names and DragonflyBSD has only s in them or
both p and s can appear in both kinds of systems?