markj (Mark Johnston)
User

Projects

User Details

User Since
Mar 12 2014, 1:00 AM (201 w, 1 d)

Recent Activity

Yesterday

markj accepted D13956: PTI: Trap if we returned to userspace with kernel (full) page table still active..
Wed, Jan 17, 10:40 PM
markj added a comment to D13926: Require the SHF_ALLOC flag for program sections from kernel object modules..
In D13926#292633, @jhb wrote:
In D13926#292526, @kib wrote:

I think that this change does not affect correctness of the loader and kernel linkers code. I mean that, in my opinion, it is fine to either load or not load the non-allocatable sections. More, it is fine to introduce arbitrary inter-section gaps between the sections, because compiler or assembler cannot know the final linking layout of the resulting executable or shared object, so they must provide enough relocations to make the linking result correct. From this PoV, presence of the non-allocatable sections between progbits is just a gap.

It does potentially waste RAM (not so much for .comment, but possibly for CTF for which we load a duplicate copy). I think markj@ also ran into this with dtrace and had to employ some kind of workaround actually.

Wed, Jan 17, 10:31 PM
markj committed rS328096: Annotate a couple of changes from r328083..
Annotate a couple of changes from r328083.
Wed, Jan 17, 9:52 PM
markj added a comment to D13956: PTI: Trap if we returned to userspace with kernel (full) page table still active..
In D13956#292712, @kib wrote:
In D13956#292703, @alc wrote:

Maybe I'm confused. Is that wording actually correct? Shouldn't it be "Ensure that the return to user mode immediately traps if the kernel-mode page table is mistakenly left active."

Well, it is not return to usermode which traps, but usermode itself after we already returned. At least, I hope there is no such bug in the CPUs.

Wed, Jan 17, 8:16 PM
markj accepted D13956: PTI: Trap if we returned to userspace with kernel (full) page table still active..
Wed, Jan 17, 7:23 PM
markj committed rS328086: MFC r327768:.
MFC r327768:
Wed, Jan 17, 3:13 PM

Tue, Jan 16

markj added inline comments to D13797: PTI for amd64..
Tue, Jan 16, 10:55 PM
markj added inline comments to D13797: PTI for amd64..
Tue, Jan 16, 7:05 PM
markj added inline comments to D13797: PTI for amd64..
Tue, Jan 16, 5:15 PM
markj accepted D13735: Assign map->header values to avoid boundary checks.
Tue, Jan 16, 3:01 PM

Mon, Jan 15

markj committed rS328003: Remove changes left over from debugging..
Remove changes left over from debugging.
Mon, Jan 15, 2:57 PM
markj committed rS327988: MFC r327700:.
MFC r327700:
Mon, Jan 15, 1:09 AM
markj committed rS327987: Add netdump support to iflib..
Add netdump support to iflib.
Mon, Jan 15, 1:07 AM
markj committed rS327980: Create a branch for integrating changes needed for netdump..
Create a branch for integrating changes needed for netdump.
Mon, Jan 15, 12:59 AM
markj committed rS327985: Add netdump support to re(4)..
Add netdump support to re(4).
Mon, Jan 15, 12:58 AM
markj committed rS327986: Add netdump support to vtnet(4)..
Add netdump support to vtnet(4).
Mon, Jan 15, 12:58 AM
markj committed rS327984: Add netdump support to alc(4)..
Add netdump support to alc(4).
Mon, Jan 15, 12:58 AM
markj committed rS327983: Add dumpon(8) support for netdump client configuration..
Add dumpon(8) support for netdump client configuration.
Mon, Jan 15, 12:54 AM
markj committed rS327982: Add netdump client code and mbuf allocator hooks..
Add netdump client code and mbuf allocator hooks.
Mon, Jan 15, 12:50 AM
markj committed rS327981: Add a flag to disable per-CPU item caching in a given zone..
Add a flag to disable per-CPU item caching in a given zone.
Mon, Jan 15, 12:47 AM

Sun, Jan 14

markj committed rS327973: Remove tst.zonename.d from the list of expected failures..
Remove tst.zonename.d from the list of expected failures.
Sun, Jan 14, 5:56 PM
markj committed rS327972: Use the thread's ucred struct when fetching jid or jailname..
Use the thread's ucred struct when fetching jid or jailname.
Sun, Jan 14, 5:56 PM
markj accepted D13707: Defer and aggregate swap_pager_meta_build frees.
Sun, Jan 14, 5:22 PM

Fri, Jan 12

markj committed rS327888: Add "jid" and "jailname" variables to DTrace..
Add "jid" and "jailname" variables to DTrace.
Fri, Jan 12, 8:00 PM
markj closed D13877: DTrace: Add jailname/jid builtins.
Fri, Jan 12, 8:00 PM · DTrace
markj accepted D13877: DTrace: Add jailname/jid builtins.

Thanks. I'll commit this later tonight.

Fri, Jan 12, 5:37 PM · DTrace
markj accepted D13877: DTrace: Add jailname/jid builtins.

Looks ok to me, thanks. Did you end up writing test cases for safety/ as we discussed on IRC?

Fri, Jan 12, 4:17 PM · DTrace

Thu, Jan 11

markj committed rS327833: Add mallocarray.9 to malloc.9's MLINKS..
Add mallocarray.9 to malloc.9's MLINKS.
Thu, Jan 11, 4:41 PM
markj committed rS327830: MFC r324125 (andreast):.
MFC r324125 (andreast):
Thu, Jan 11, 3:32 PM
markj committed rS327804: MFC r327698:.
MFC r327698:
Thu, Jan 11, 12:55 AM
markj accepted D13845: Don't hang on crashes with dumps to ada devices..
Thu, Jan 11, 12:52 AM

Wed, Jan 10

markj committed rS327795: Add a regression test for r327794..
Add a regression test for r327794.
Wed, Jan 10, 9:40 PM
markj committed rS327794: Fix an off-by-one in dt_opt_setenv()..
Fix an off-by-one in dt_opt_setenv().
Wed, Jan 10, 9:37 PM
markj committed rS327789: MFC r327525:.
MFC r327525:
Wed, Jan 10, 9:06 PM
markj committed rS327786: MFC r327168, r327213:.
MFC r327168, r327213:
Wed, Jan 10, 8:48 PM
markj committed rS327785: MFC r325530 (jeff), r325566 (kib), r325588 (kib):.
MFC r325530 (jeff), r325566 (kib), r325588 (kib):
Wed, Jan 10, 8:39 PM
markj committed rS327780: Add regression tests for r327779..
Add regression tests for r327779.
Wed, Jan 10, 7:41 PM
markj committed rS327779: Fix handling of read errors during mirror synchronization..
Fix handling of read errors during mirror synchronization.
Wed, Jan 10, 7:37 PM
markj committed rS327769: Remove the executable bit from some recently added test scripts..
Remove the executable bit from some recently added test scripts.
Wed, Jan 10, 3:55 PM
markj committed rS327768: Clarify the use of the gmirror flag mask constants..
Clarify the use of the gmirror flag mask constants.
Wed, Jan 10, 3:21 PM
markj added inline comments to D13633: Add support for zstd to subr_compressor.c..
Wed, Jan 10, 3:06 PM
markj committed rS327760: Avoid referencing a possibly freed consumer after r327496..
Avoid referencing a possibly freed consumer after r327496.
Wed, Jan 10, 5:06 AM

Mon, Jan 8

markj committed rS327707: Generalize the gzio API..
Generalize the gzio API.
Mon, Jan 8, 9:28 PM
markj closed D13632: Rework the gzio API..
Mon, Jan 8, 9:27 PM
markj added a comment to D13633: Add support for zstd to subr_compressor.c..
In D13633#289172, @cem wrote:

If by "other pieces" you mean D13632, I am going to commit them momentarily. :)

No, I meant the corresponding userspace and documentation changes. Though yes, this patch doesn't make sense without D13632 either :-).

Mon, Jan 8, 9:09 PM
markj added a comment to D13633: Add support for zstd to subr_compressor.c..
In D13633#289164, @cem wrote:

Yes, this is only the kernel code changes from your original patch. Sorry for not specifying that.

Ok. I like the approach, but don't think it makes sense to commit without the other pieces in place as well.

Mon, Jan 8, 8:56 PM
markj added a comment to D13633: Add support for zstd to subr_compressor.c..
In D13633#289147, @cem wrote:

We're missing savecore support, manual page additions, as well as inclusion in GENERIC, right?

Mon, Jan 8, 8:42 PM
markj updated the diff for D13632: Rework the gzio API..
  • Add a SPDX identifier to subr_compressor.c
Mon, Jan 8, 5:59 PM
markj updated the diff for D13633: Add support for zstd to subr_compressor.c..
  • Need to define COMPRESSOR_ZSTD
Mon, Jan 8, 5:02 PM
markj updated the diff for D13632: Rework the gzio API..
  • Sprinkle const
  • Remove unintended reference to zstd
Mon, Jan 8, 5:00 PM
markj added inline comments to D13632: Rework the gzio API..
Mon, Jan 8, 4:56 PM
markj committed rS327701: MFC r322547:.
MFC r322547:
Mon, Jan 8, 4:36 PM
markj committed rS327700: Sort and remove unneeded includes..
Sort and remove unneeded includes.
Mon, Jan 8, 3:56 PM
markj committed rS327698: Release the queue lock before restarting the worker loop..
Release the queue lock before restarting the worker loop.
Mon, Jan 8, 3:42 PM

Thu, Jan 4

markj committed rS327553: MFC r326861-r326863:.
MFC r326861-r326863:
Thu, Jan 4, 4:20 PM
markj committed rS327552: MFC r326912:.
MFC r326912:
Thu, Jan 4, 4:19 PM
markj committed rS327551: MFC r326774, r326811:.
MFC r326774, r326811:
Thu, Jan 4, 4:17 PM
markj added inline comments to D13719: Fix CVE-2010-4670, CVE-2010-4671, CVE-2010-4669, CVE-2011-2393.
Thu, Jan 4, 3:56 PM
markj added a comment to D13719: Fix CVE-2010-4670, CVE-2010-4671, CVE-2010-4669, CVE-2011-2393.
In D13719#287678, @ae wrote:
Thu, Jan 4, 3:19 PM
markj added a comment to D13693: Restructure swapout tests after vm map locking was removed..
In D13693#288066, @kib wrote:

Alan. Mark. Should I go with the commit of the current patch ? There are two supposed comments changing which can be done later.

Thu, Jan 4, 3:12 PM

Wed, Jan 3

markj added a comment to D13671: Collection of fixes for OOM handling of some corner cases..
In D13671#287790, @kib wrote:

Remove double *hz.

Wed, Jan 3, 7:16 PM
markj committed rS327525: Add missing newlines to a couple of error messages..
Add missing newlines to a couple of error messages.
Wed, Jan 3, 6:20 PM

Tue, Jan 2

markj committed rS327496: Fix some I/O ordering issues in gmirror..
Fix some I/O ordering issues in gmirror.
Tue, Jan 2, 6:12 PM
markj closed D13559: Fix gmirror BIO_FLUSH and sync collision queuing..
Tue, Jan 2, 6:12 PM
markj added a comment to D13559: Fix gmirror BIO_FLUSH and sync collision queuing..
In D13559#287445, @imp wrote:

I'm a little uneasy still about gmirror, but to erase that discomfort is too high a bar for me to set for this review :)

Tue, Jan 2, 5:03 PM
markj committed rS327493: MFC r326983:.
MFC r326983:
Tue, Jan 2, 4:20 PM
markj committed rS327492: MFC r326935:.
MFC r326935:
Tue, Jan 2, 4:15 PM
markj committed rS327491: MFC r326919:.
MFC r326919:
Tue, Jan 2, 4:14 PM
markj added a comment to D13559: Fix gmirror BIO_FLUSH and sync collision queuing..

I plan to commit this in the next day or two if there are no objections.

Tue, Jan 2, 3:49 PM
markj added a comment to D13693: Restructure swapout tests after vm map locking was removed..
In D13693#287221, @kib wrote:
In D13693#287219, @alc wrote:

Exactly. I see nothing that makes this loop body of FOREACH_PROC particularly more expensive than some of the others. In essence, I'm looking for a reason why we should be concerned about lock contention here, but not in the other cases.

I think with bigger and bigger systems and gradually bigger average workloads, were current hundreds of processes become thousands, we want to avoid holding the global lock for the whole FOREACH_PROC (esp. nesting FOREACH_THREAD_IN_PROC() three times). We should not regress the code which already does not hold the lock for whole loop.

Tue, Jan 2, 3:41 PM
markj accepted D13732: rpcbind: Fix race in signal termination.
Tue, Jan 2, 5:24 AM
markj accepted D13732: rpcbind: Fix race in signal termination.
Tue, Jan 2, 5:14 AM
markj accepted D13728: PR 224503: rpcbind: Do not use signal-unsafe functions in SIGTERM handler.
Tue, Jan 2, 12:30 AM

Mon, Jan 1

markj added a comment to D13728: PR 224503: rpcbind: Do not use signal-unsafe functions in SIGTERM handler.

The _exit() bit looks right to me.

Mon, Jan 1, 9:53 PM
markj accepted D13693: Restructure swapout tests after vm map locking was removed..
In D13693#287136, @kib wrote:
In D13693#287127, @alc wrote:

To be clear, I am saying that the original reason for releasing the lock doesn't apply to this code anymore. If we are going to release it, we should be able to articulate a new reason for doing so.

As I said above, not holding the lock for whole scan time makes it possible for other numerous consumers of the lock to make the progress. We are reducing the contention on allproc_lock by dropping it, either on each iteration or on should_yeild().

Mon, Jan 1, 9:24 PM

Sat, Dec 30

markj added inline comments to D13693: Restructure swapout tests after vm map locking was removed..
Sat, Dec 30, 4:23 PM
markj added inline comments to D13693: Restructure swapout tests after vm map locking was removed..
Sat, Dec 30, 2:31 PM
markj added a comment to D13644: Eliminate "pass" from vm_pageout_scan() and vm_pageout_worker().
In D13644#285630, @alc wrote:

I left this patch running overnight on a machine that was configured to perform a moderate amount of paging.

Is there any reason not to commit the part that makes vm_pageout_wanted static?

Sat, Dec 30, 1:23 AM

Thu, Dec 28

markj added a comment to D13644: Eliminate "pass" from vm_pageout_scan() and vm_pageout_worker().
In D13644#285475, @alc wrote:

I don't see a better option for dealing with the remaining race than your proposed change to pagedaemon_wakeup().

Combined, won't these changes eliminate the need for the "|| !vm_pages_needed)"-triggered wakeup in pagedaemon_wait()?

Thu, Dec 28, 5:22 AM
markj added a comment to D13559: Fix gmirror BIO_FLUSH and sync collision queuing..
In D13559#285371, @imp wrote:

A question: Does it make sense to 'resilver' those sectors that return EIO rather than simply kicking the disk to the curb and only kicking it to the curb if the resilver fails to readback correctly?

Thu, Dec 28, 2:13 AM
markj added a comment to D13644: Eliminate "pass" from vm_pageout_scan() and vm_pageout_worker().
In D13644#285437, @alc wrote:

The following version would address the issue that I raised yesterday without eliminating "pass". That said, I am no longer incrementing it. Its value is either 0 or 1.

Thu, Dec 28, 2:09 AM

Tue, Dec 26

markj added a comment to D13644: Eliminate "pass" from vm_pageout_scan() and vm_pageout_worker().

With the elimination of "pass", I want to point out that the page daemon will attempt to bring the number of free pages up to the target no matter how it was awakened. I do not think that that is a bad idea.

Tue, Dec 26, 8:31 PM
markj added inline comments to D13559: Fix gmirror BIO_FLUSH and sync collision queuing..
Tue, Dec 26, 8:23 PM
markj closed D13640: Ensure that pass > 0 when starting a scan with vm_pages_needed = 1..
Tue, Dec 26, 4:30 PM
markj committed rS327213: Ensure that pass > 0 when starting a scan with vm_pages_needed == 1..
Ensure that pass > 0 when starting a scan with vm_pages_needed == 1.
Tue, Dec 26, 4:30 PM
markj added reviewers for D13640: Ensure that pass > 0 when starting a scan with vm_pages_needed = 1.: alc, kib.
Tue, Dec 26, 3:40 PM
markj created D13640: Ensure that pass > 0 when starting a scan with vm_pages_needed = 1..
Tue, Dec 26, 3:38 PM
markj added inline comments to D13101: Add support for Zstd-compressed kernel and userspace dumps.
Tue, Dec 26, 3:16 AM
markj updated the diff for D13633: Add support for zstd to subr_compressor.c..
  • Remove dead code.
Tue, Dec 26, 3:16 AM
markj updated the test plan for D13633: Add support for zstd to subr_compressor.c..
Tue, Dec 26, 3:14 AM
markj updated the test plan for D13632: Rework the gzio API..
Tue, Dec 26, 3:14 AM
markj created D13633: Add support for zstd to subr_compressor.c..
Tue, Dec 26, 3:11 AM
markj created D13632: Rework the gzio API..
Tue, Dec 26, 3:09 AM
markj abandoned D13631: Configure ZSTDIO..
Tue, Dec 26, 3:09 AM
markj created D13631: Configure ZSTDIO..
Tue, Dec 26, 3:08 AM

Mon, Dec 25

markj updated subscribers of D13611: Optimize the vm object bypass and collapse counters.
In D13611#284781, @alc wrote:

As an aside, it seems to me that sys/counter.h and/or the man page ought to provide guidance on when counter_u64_alloc() becomes possible. It appears to me that it is earlier than SI_SUB_CPU, specifically, we need the following to have completed:

SYSINIT(pcpu_zones, SI_SUB_KMEM, SI_ORDER_ANY, pcpu_zones_startup, NULL);
Mon, Dec 25, 7:54 PM
markj accepted D13611: Optimize the vm object bypass and collapse counters.
Mon, Dec 25, 5:05 PM

Sun, Dec 24

markj committed rS327168: Fix two problems with the page daemon control loop..
Fix two problems with the page daemon control loop.
Sun, Dec 24, 7:45 PM
markj closed D13424: Fix two problems with the page daemon control loop..
Sun, Dec 24, 7:45 PM
markj added inline comments to D13424: Fix two problems with the page daemon control loop..
Sun, Dec 24, 7:25 PM