Kernel API changes IO acct, global_page_state, etc

[dinatale2]

Description

generic_start_io_acct/generic_end_io_acct in the master
branch of the linux kernel requires that the request_queue
be provided.

Move the logic from freemem in the spl to arc_free_memory
in arc.c. Do this so we can take advantage of global_page_state
interface checks in zfs.

Upstream kernel replaced struct block_device with
struct gendisk in struct bio. Determine if the
function bio_set_dev exists during configure
and have zfs use that if it exists.

Motivation and Context

Fix the Kernel builder in the ZFS buildbot.

How Has This Been Tested?

Builds in a VM locally. Going to see if buildbot builds work.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Performance enhancement (non-breaking change which improves efficiency)
• Code cleanup (non-breaking change which makes code smaller or more readable)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation (a change to man pages or other documentation)

Checklist:

• My code follows the ZFS on Linux code style requirements.
• I have updated the documentation accordingly.
• I have read the CONTRIBUTING document.
• I have added tests to cover my changes.
• All new and existing tests passed.
• All commit messages are properly formatted and contain Signed-off-by.
• Change has been approved by a ZFS on Linux member.

[chrisrd]

It's early days, but the global_page_state => global_node_page_state part of this fix, along with #6528, looks to have greatly stabilised my Linux 4.9 / 192GB system. Contrast the ARC, with the same workload, from Monday's boot prior to fix with previously-inexplicable wild gyrations in arcstats:c (Target Size in the graph) etc., to the current post-fix boot from Wed afternoon:

The summary of those fixes is that with the Linux API change introduced by torvalds/linux@599d0c9, our free memory calcs were based on completely the wrong thing. E.g. where we're looking at NR_FILE_PAGES, after the Linux change, and before these fixes, we were actually looking at one of NR_ZSPAGES, NUMA_HIT or NR_FREE_CMA_PAGES, depending on how your kernel is configured. (That was a particularly egregious API change in Linux - instead of breaking the compilation it silently did the wrong thing 👎 👎 👎)

This also goes a long way towards explaining various arc_prune issues (including #6630), with the arcstats:c collapses causing extreme efforts to reduce arcstats:size - particulary on larger memory boxes where a collapse of arcstats:c can cause the box to struggle to release many, many GB of memory.

[behlendorf]

Functionally this looks good. Just a little suggested refactoring. I completely agree with @chrisrd , it's really unfortunate the kernel changed this behavior so significantly in a way which didn't break the build. I'm just glad it's been caught. @chrisrd your graph looks much more like what I'd expect, thank you for the quick feedback!

[behlendorf]

This would be more readable if we defined our own version of bio_set_dev() in linux/blkdev_compat.h when !HAVE_BIO_SET_DEV. A similar thing was done for bio_set_op_attrs().

#ifndef HAVE_BIO_SET_DEV
static inline void
bio_set_dev(struct bio *bio, struct block_device *bdev)
{
        bio->bi_bdev = bdev;
}
#endif /* !HAVE_BIO_SET_DEV */

[behlendorf]

Use bio_set_dev() which will now always be available.

[behlendorf]

@dinatale2 I added a commit to this PR to add the bio_set_dev() helper when needed.