fix(90kernel-modules): fix MMC and NVMe on kernels 6.0+

[mikhailnov]

Commit 6f8191fdf41 ("block: simplify disk shutdown") (torvalds/linux@6f8191f) in Linux kernels 6.0 and above renamed blk_cleanup_disk to blk_mq_destroy_queue. Adapt code to handle both cases.

Because of this the kernel module mmc_block.ko was not included into initrd on kernel 6.1:
dracut-install: No symbol or path match for '/lib/modules/6.1.0-generic-0.rc7.1rosa2021.1-x86_64//kernel/drivers/mmc/core/mmc_block.ko.zst
while on kernel 5.15 it was included:
dracut-install: Handling /lib/modules/5.15.79-generic-1rosa2021.1-x86_64//kernel/drivers/mmc/core/mmc_block.ko.zst
dracut-install: Module mmc_block: symbol blk_cleanup_queue matched inclusion filter

It made booting OS from root on /dev/mmcblk* impossible.

Just adding blk_mq_destroy_queue was not enough to make mmc_block.ko be included into initrd (maybe because call of blk_cleanup_disk() was removed and call of blk_mq_destroy_queue() was not added).
So also adding blk_mq_alloc_request which is used from mmc_block.

Fixes: #2102

[LaszloGombos]

@mikhailnov

Error: commit_message:1:24: error: commit message's description contains blacklisted word or repeats type tag
fix(90kernel-modules): fix MMC and NVMe on kernels 6.0+
^^^
commit_message:1:24: note: prefer using the imperative for verbs

maybe "fix(90kernel-modules): fixed MMC and NVMe....' ?

[mikhailnov]

maybe "fix(90kernel-modules): fixed MMC and NVMe....' ?

Done

[LaszloGombos]

@mikhailnov

Error: commit_message:1:24: error: commit message's description contains blacklisted word or repeats type tag
fix(90kernel-modules): fixed MMC and NVMe on kernels 6.0+
^^^^^
commit_message:1:24: note: prefer using the imperative for verbs

Please try to find a solution (it is me not doing this) - https://github.com/dracutdevs/dracut/actions/runs/3610891530/jobs/6084978300

Perhaps "correct"

[johannbg]

You need to rephrase your commit message excluding the word fixed other than LGTM

[LaszloGombos]

LGTM

[mikhailnov]

@LaszloGombos thanks for rephrasing the commit message

[aafeijoo-suse]

Related: redhat-plumbers/dracut-rhel9#57
@pvalena is there any reason why RH didn't send this upstream first, so all distros using dracut with the latest kernel version could benefit from this patch?

your commit message rephrased in the latest PR

[aafeijoo-suse]

Related: redhat-plumbers/dracut-rhel9#57 @pvalena is there any reason why RH didn't send this upstream first, so all distros using dracut with the latest kernel version could benefit from this patch?

@johannbg what's your opinion here? I'm a bit upset, things like this should never happen with RH and the worst thing is we are only getting silence and no justification.

[johannbg]

@aafeijoo-suse I can only assume the reason but how can you be upset? Any corporation or individual is free to submit patches upstream or not RH and or it's employees are not obliged to submit patches here upstream or owe upstream somekind of justification if they dont.

[aafeijoo-suse]

Yeah, better forget about this.

In another vein, maybe we should add that patch, because I think this PR does not cover the case of the nvdimm drivers.

# zstd -d /lib/modules/5.14.21-150400.24.18-default/kernel/drivers/nvdimm/nd_pmem.ko.zst -o nd_pmem.ko
/lib/modules/5.14.21-150400.24.18-default/kernel/drivers/nvdimm/nd_pmem.ko.zst: 38904 bytes 
# nm nd_pmem.ko | grep blk
                 U __blk_alloc_disk
-------------->  U blk_cleanup_disk
                 U blk_freeze_queue_start
                 U blk_queue_flag_set
                 U blk_queue_logical_block_size
                 U blk_queue_max_hw_sectors
                 U blk_queue_physical_block_size
                 U blk_queue_write_cache
                 U blk_status_to_errno
                 U errno_to_blk_status

# zstd -d /lib/modules/6.1.0-3.g7f1864f-default/kernel/drivers/nvdimm/nd_pmem.ko.zst -o nd_pmem.ko
/lib/modules/6.1.0-3.g7f1864f-default/kernel/drivers/nvdimm/nd_pmem.ko.zst: 46191 bytes 
# nm nd_pmem.ko | grep blk
                 U __blk_alloc_disk
                 U blk_queue_flag_set
                 U blk_queue_logical_block_size
                 U blk_queue_max_hw_sectors
                 U blk_queue_physical_block_size
                 U blk_queue_write_cache
                 U blk_status_to_errno
                 U errno_to_blk_status

[johannbg]

@aafeijoo-suse yeah I dont see why not but I probably should mention he did not add his patch to fedora either ( which might raise some eyebrows in Raleigh, rarely is a fix limited to RHEL ) so there might be a catch 22 for doing that

[mikhailnov]

Also, it would be very nice to somehow clean up hackery like explicitly adding drivers/mmc on aarch64 only

dracut/modules.d/90kernel-modules/module-setup.sh

Line 77 in e0d57a8

"=drivers/mmc/core" \

, while relying on detecting used kernel symbols on other architectures