I've picked up Simon's v1 of this series and moved it to an RFC with this v2. I don't intend for this series to go in as-is but rather since I spent a good bit of time iterating over the problems of trying a conversion (in a few places) where we only selectively add back in the header being removed from common.h in the case of a fail to build, I didn't want the work lost.
What I think needs to be done moving forward is even smaller series here where we focus on removing one or two headers, but then only re-add them where required.
Also note that this series shows a few funny issues. The patch to remove <linux/kernel.h> and selectively re-add it shows: bcm958712k : all -4 text -4 u-boot: add: 1/0, grow: 0/-3 bytes: 24/-28 (-4) function old new delta blk_dread - 24 +24 part_test_efi 184 180 -4 is_gpt_valid 736 724 -12 fs_devread 600 588 -12
everywhere that code is used. I don't see why, but there's some underlying problem exposed in the move. I believe it's also that patch which shows, for every big-endian platform something like: T4160RDB : all -592 text -592 u-boot: add: 2/0, grow: 1/-14 bytes: 68/-656 (-588) function old new delta ___arch__swab32 - 48 +48 blk_dread - 12 +12 ext4fs_bg_get_inode_table_id 76 84 +8 ehci_td_buffer 164 156 -8 _ehci_destroy_int_queue 244 236 -8 static.ehci_update_endpt2_dev_n_port 116 104 -12 ext4fs_open 200 188 -12 _ehci_poll_int_queue 236 224 -12 usb_lowlevel_init 916 892 -24 ext4fs_mount 332 304 -28 fs_devread 572 540 -32 ext4fs_find_file1 756 724 -32 ext4fs_iterate_dir 848 812 -36 ext4fs_read_inode 520 452 -68 static._ehci_create_int_queue 1000 908 -92 ehci_submit_async 1632 1520 -112 read_allocated_block 2532 2352 -180
and I lack hardware to see (and it looks like qemu-ppce500 can't be given a disk atm) if the problem is that ext2/4 is broken before and fixed now, or fixed now and broken with this patch, as that's my first concern on seeing ___arch__swab32 show up. But maybe it's a harmless "no, don't inline ..." decision the compiler is now able to make. But very non-obvious and needing a run-time sanity check to be sure.