Paths

Table of Contentst

ktrace: generate events on extended errors
ClosedPublic
Actions

Authored by kib on Sat, May 31, 11:07 PM.

Details

Reviewers

brooks
jrtc27

Commits

rG2a35b00732d9: kern/kern_ktrace.c: normalize includes
rGe26f3836d4e1: kdump(1): Use static array for header types string literals
rGde6012c48e21: kdump: staticise functions in kdump.c
rG9a0360355792: kdump(1): pretty-print KTR_EXTERR
rG96f4be881e8e: ktrace(1): teach about KTR_EXTERR
rG6d43260d49e6: ktrace: generate events on extended errors
rG6abe7e2a929c: exterr: add category for ktrace
rGac1a57f6c325: ktrace: define user interface for tracing extended errors
rGe5586b32d2dc: exterr: move the definition of user-visible struct uexterror into…
rGb3a93154db74: Extract exterr_to_ue()

Summary

The ktrexterr() function is provided unconditionally, to mitigate the
issue of missed include of "opt_ktrace.h" before using exterrvar.h
otherwise.

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Skipped

Unit

Tests Skipped

Event Timeline

kib created this revision.Sat, May 31, 11:07 PM

Herald added subscribers: ziaee, olce, imp. · View Herald TranscriptSat, May 31, 11:07 PM

kib requested review of this revision.Sat, May 31, 11:07 PM

Looks good. I can replace our CheriBSD SYSERRCAUSE with this pretty easily.

The verbose formatted output does somewhat make me want a valid fields bitmask (particularly as I'm planning to steal the reserved space for two 128-bit capabilities so we can pretty print them.)

sys/kern/kern_ktrace.c
107	Hard to be sure in phab, but it looks like there might be a whitespace difference to the existing lines after `struct`
sys/sys/exterr_cat.h
16	I don't understand this comment.

This revision is now accepted and ready to land.Mon, Jun 2, 8:46 PM

kib marked 2 inline comments as done.Mon, Jun 2, 9:26 PM

kib added inline comments.

sys/sys/exterr_cat.h
16	I mean that this category is only to have something to satisfy requirement of exterrvar.h so it can be included into kern_ktrace.c. ktrace facility is not supposed to generate extended errors.

Change to tab in the structure def.
Expand comment in exterr_cat.h.

This revision now requires review to proceed.Mon, Jun 2, 9:27 PM

Herald added a subscriber: jonathan. · View Herald TranscriptMon, Jun 2, 9:27 PM

In D50633#1156331, @brooks wrote:

The verbose formatted output does somewhat make me want a valid fields bitmask (particularly as I'm planning to steal the reserved space for two 128-bit capabilities so we can pretty print them.)

Do you intend to code this yourself?

In D50633#1156349, @kib wrote:

In D50633#1156331, @brooks wrote:

The verbose formatted output does somewhat make me want a valid fields bitmask (particularly as I'm planning to steal the reserved space for two 128-bit capabilities so we can pretty print them.)

Do you intend to code this yourself?

I'm happy to code to the capability bits. I'm less sure where to find bits for a mask.

brooks added inline comments.Mon, Jun 2, 10:29 PM

usr.bin/kdump/kdump.c
526	Does this want a `/* FALLTHROUGH */`?

In D50633#1156363, @brooks wrote:

I'm happy to code to the capability bits. I'm less sure where to find bits for a mask.

I am not sure what do you mean. Are you saying that you do not want to use one of the rsrv1[] array member for the flags?

We can break ABI in main, I believe.

usr.bin/kdump/kdump.c
526	No, it would be too ugly. I fixed ktrexterr() by adding \n.

Add break after kexterror() call.

In D50633#1156370, @kib wrote:

In D50633#1156363, @brooks wrote:

I'm happy to code to the capability bits. I'm less sure where to find bits for a mask.

I am not sure what do you mean. Are you saying that you do not want to use one of the rsrv1[] array member for the flags?

We can break ABI in main, I believe.

And we can bump version, to not break things too much. I will not add compat code though.

In D50633#1156370, @kib wrote:

In D50633#1156363, @brooks wrote:

I'm happy to code to the capability bits. I'm less sure where to find bits for a mask.

I am not sure what do you mean. Are you saying that you do not want to use one of the rsrv1[] array member for the flags?

I'd need all of the space in rsrv1 for two capabilities (128-bits each, strongly aligned (although just for convenience in this case)). I'll try a couple things and see what makes sense once this is in.

In D50633#1156401, @brooks wrote:

In D50633#1156370, @kib wrote:

In D50633#1156363, @brooks wrote:

I'm happy to code to the capability bits. I'm less sure where to find bits for a mask.

I am not sure what do you mean. Are you saying that you do not want to use one of the rsrv1[] array member for the flags?

I'd need all of the space in rsrv1 for two capabilities (128-bits each, strongly aligned (although just for convenience in this case)). I'll try a couple things and see what makes sense once this is in.

For cheri you can break ABI as you want, anyway.
But I will add uint32_t flags field and bump the version after this change is committed, as promised.

This revision was not accepted when it landed; it landed in state Needs Review.Tue, Jun 3, 4:07 AM

Closed by commit rGb3a93154db74: Extract exterr_to_ue() (authored by kib). · Explain Why

This revision was automatically updated to reflect the committed changes.

kib added a commit: rGb3a93154db74: Extract exterr_to_ue().

kib added a commit: rGe5586b32d2dc: exterr: move the definition of user-visible struct uexterror into….

kib added a commit: rGac1a57f6c325: ktrace: define user interface for tracing extended errors.

kib added a commit: rG6abe7e2a929c: exterr: add category for ktrace.

kib added a commit: rG6d43260d49e6: ktrace: generate events on extended errors.

kib added a commit: rG96f4be881e8e: ktrace(1): teach about KTR_EXTERR.

kib added a commit: rG9a0360355792: kdump(1): pretty-print KTR_EXTERR.

kib added a commit: rGde6012c48e21: kdump: staticise functions in kdump.c.

kib added a commit: rGe26f3836d4e1: kdump(1): Use static array for header types string literals.