Could you use spcr->SerialPort.AccessWidth to find this? It's set to 1 in the copy of the spcr table I have indicating byte access.

I'm fine with breaking from the ABI in the kernel & modules as long as it's just for optimisation and we don't require the breakage to allow the kernel to still work after being built with, for example, gcc & linked with bfd.

Split out the header
Enable clocks, etc

Is there a reason to limit this to arm and arm64?

pmap_qenter each page
Use vm_page_next

Use vm_page_lookup to find the page

Use vm_object_reference rather than manual tracking of the mmap state

Add a missing header

Use vm_page_unwire_noq

In D19097#408644, @jchandra wrote:

Tested with overnight 'make -j 224 buildworld' without failures.

How long does that take?

I've been thinking about reserving a fixed VA on arm64 for this and reworking how we set the range for EARLY_PRINTF. The idea would be we only need to set the physical address, the virtual address base will be fixed, with the offset dependent on the physical address.

Remove a D_TRACKCLOSE that sneaked in

Update the kcov tracking to be based on the number of threads actively
being traced.

You should remove crtsavres.o from gnu/lib/csu/Makefile

Fix the spelling of KCOV_ENTRY_SIZE

Add KCOV_ENTRIY_SIZE
While here update the header copyright

Simplify the locking in kcov_mmap_cleanup
Update the comments to describe the state machine
Add access requirements to struct kcov_info

Rework the cleanup code:

• Use barriers to ensure ordering between info->state and other info data.
• Check if a thread is using the info struct now we don't enter DYING on thread exit

Add a missing atomic_thread_fence_seq_cst call to get_kinfo

Remove the kcov_thread_exit prototype from <sys/kcov.h>

• Fix a comment
• Remove unneeded code from kern_thread.c
• Fix a typo in a test

In D14599#399447, @mhorne063_gmail.com wrote:

Also Andrew, I saw you mention in another thread that you had a man page started but not quite ready. If you'd like help reviewing or finishing it let me know, as I had started a basic one locally myself.

Update based on feedback:

• Remove duplicate code from kern.pre.mk
• Switch to number of entries when setting the buffer size
• Remove the unneeded kcov_thread_exit
• Return to KCOV_STATE_READY on thread exit to allow other threads to use the fd

In D14599#398017, @tuexen wrote:

I was trying to use this patch with syzkaller, but couldn't get it running. To figure out what is going on, I created a small test program,

kcov_test.c511 BDownload

, which results in a reboot of the system. When changing the code such that the ioctl(fd, KIODISABLE, mode) command is executed, to reboot happens. I think the system should not reboot...

Add tests and fix the code based on issues found by these

There is no need to set the cpu affinity, the userspace register view is the same on each core.