Page MenuHomeFreeBSD

markj (Mark Johnston)
User

Projects

User Details

User Since
Mar 12 2014, 1:00 AM (323 w, 5 d)

Recent Activity

Today

markj updated the test plan for D25001: Fix boot on systems where NUMA domain 0 is unpopulated..
Mon, May 25, 3:25 PM
markj requested review of D25001: Fix boot on systems where NUMA domain 0 is unpopulated..
Mon, May 25, 3:24 PM
markj committed rS361455: Fix a mismerge in r361454..
Fix a mismerge in r361454.
Mon, May 25, 1:14 PM
markj committed rS361454: MFC r360906:.
MFC r360906:
Mon, May 25, 12:50 PM
markj committed rS361453: MFC r360903:.
MFC r360903:
Mon, May 25, 12:48 PM
markj committed rS361451: MFC r360903:.
MFC r360903:
Mon, May 25, 12:45 PM

Sat, May 23

markj committed rS361408: Revert mlock_helper.c test code changes from r361350..
Revert mlock_helper.c test code changes from r361350.
Sat, May 23, 3:32 AM

Fri, May 22

markj updated the diff for D24755: Permit deferred creation of SMR structures for the VM radix zone..

Create the radix SMR structures earlier, during SI_SUB_SMR, right after
SI_SUB_VM. Use a global list of SMR structures to perform deferred
initialization of per-CPU structures. This allows us to choose a layout
for per-CPU data at SI_SUB_CPU.

Fri, May 22, 10:23 PM
markj committed rS361386: MFC r361352:.
MFC r361352:
Fri, May 22, 3:43 PM
markj committed rS361352: Fix the build after r361033 when ACPI is disabled..
Fix the build after r361033 when ACPI is disabled.
Fri, May 22, 1:19 AM

Thu, May 21

markj committed rS361350: MFC r347532, r347541.
MFC r347532, r347541
Thu, May 21, 10:48 PM
markj committed rS361338: Fix ACCEPT_FILTER_DEFINE to pass the version to MODULE_VERSION..
Fix ACCEPT_FILTER_DEFINE to pass the version to MODULE_VERSION.
Thu, May 21, 6:39 PM
markj accepted D24910: Add macros simplifying the fake preload setup.

Seems ok to me aside from the comment.

Thu, May 21, 6:12 PM
markj accepted D24946: Revert to the initial version of D24918.
Thu, May 21, 5:52 PM
markj accepted D24946: Revert to the initial version of D24918.

This fixes the problem for me.

Thu, May 21, 5:45 PM
markj added inline comments to D24946: Revert to the initial version of D24918.
Thu, May 21, 5:31 PM
markj accepted D24941: libprocstat: fix ZFS support.
Thu, May 21, 4:55 PM
markj added inline comments to D24941: libprocstat: fix ZFS support.
Thu, May 21, 3:59 PM
markj added inline comments to D24910: Add macros simplifying the fake preload setup.
Thu, May 21, 3:53 PM
markj committed rS361335: MFC r361039:.
MFC r361039:
Thu, May 21, 3:29 PM
markj committed rS361334: MFC r361033:.
MFC r361033:
Thu, May 21, 3:28 PM
markj committed rS361333: MFC r361031:.
MFC r361031:
Thu, May 21, 3:19 PM

Wed, May 20

markj added a comment to D24755: Permit deferred creation of SMR structures for the VM radix zone..

This makes me a little uneasy. Allocation before attach is okay, but free is not. Do we have some kind of guarantee that no free will occur?

Wed, May 20, 9:33 PM
markj committed rS361287: Don't block on the range lock in zfs_getpages()..
Don't block on the range lock in zfs_getpages().
Wed, May 20, 6:30 PM
markj closed D24839: Don't block on the range lock in zfs_getpages()..
Wed, May 20, 6:30 PM
markj added a comment to D24839: Don't block on the range lock in zfs_getpages()..
In D24839#548684, @avg wrote:

Mark, thank you very much for this!
I think that this change is good.
I still wonder if we can do some trick to avoid busying valid pages.
Maybe there is some way to be more sloppy when checking the validity.
E.g., we could possibly rely on a fact that with ZFS the range lock makes sure the validity cannot change because both page-in and page-out would need to lock the range.

Wed, May 20, 5:59 PM
markj committed rS361286: Avoid hard-coding pipe buffer sizes in the pipe and fifo kqueue tests..
Avoid hard-coding pipe buffer sizes in the pipe and fifo kqueue tests.
Wed, May 20, 5:48 PM
markj committed rS361285: MFC r360690:.
MFC r360690:
Wed, May 20, 5:33 PM
markj updated the diff for D24758: Introduce a dynamic pcpu layout on amd64..
  • Simplify layout calculation.
  • Add a block comment above pcpu_layout().
Wed, May 20, 6:21 AM
markj added inline comments to D24758: Introduce a dynamic pcpu layout on amd64..
Wed, May 20, 6:21 AM
markj added inline comments to D24758: Introduce a dynamic pcpu layout on amd64..
Wed, May 20, 3:51 AM
markj updated the diff for D24758: Introduce a dynamic pcpu layout on amd64..
  • Fix cpuhead initialization.
  • Weaken the amd64 assertion about sizeof(struct pcpu).
Wed, May 20, 3:51 AM

Tue, May 19

markj updated the diff for D24839: Don't block on the range lock in zfs_getpages()..

Fix a syntax error reported by pho@.

Tue, May 19, 6:36 PM
markj committed rS361263: Define a module version for accept filter modules..
Define a module version for accept filter modules.
Tue, May 19, 6:35 PM
markj committed rS361262: Use the symbolic name for "modmetadata_set"..
Use the symbolic name for "modmetadata_set".
Tue, May 19, 6:34 PM
markj accepted D24853: Stop performing a full icache sync when the DIC and IDC flags are set.
Tue, May 19, 3:47 PM
markj added a comment to D24839: Don't block on the range lock in zfs_getpages()..

@avg what do you think of this patch?

Tue, May 19, 2:57 PM
markj added a reviewer for D24839: Don't block on the range lock in zfs_getpages().: avg.
Tue, May 19, 2:57 PM
markj added a comment to D24621: Avoid switching CPUs in epoch_drain_callbacks()..

I need some more time to review this. Any updates?

Tue, May 19, 1:56 PM
markj committed rS361253: Belatedly add a RELNOTES entry for r347532..
Belatedly add a RELNOTES entry for r347532.
Tue, May 19, 1:54 PM
markj accepted D24845: Reorder manipulations in RB_REMOVE.
Tue, May 19, 1:43 PM
markj committed rS361240: MFC r360966:.
MFC r360966:
Tue, May 19, 3:25 AM

Mon, May 18

markj added inline comments to D24853: Stop performing a full icache sync when the DIC and IDC flags are set.
Mon, May 18, 4:16 PM
markj accepted D24845: Reorder manipulations in RB_REMOVE.
Mon, May 18, 2:49 PM

Sat, May 16

markj committed rS361096: MFC r360848:.
MFC r360848:
Sat, May 16, 12:29 AM
markj committed rS361095: pf: Add a new zone for per-table entry counters..
pf: Add a new zone for per-table entry counters.
Sat, May 16, 12:28 AM
markj closed D24843: pf: Add a new zone for per-table entry counters..
Sat, May 16, 12:28 AM

Fri, May 15

markj added a comment to D24790: libalias: retire cuseeme support.

We should also check the other modules -- for example, ftp and irc are in common use; I'm not sure off hand what skinny and smedia are.

Fri, May 15, 2:47 PM

Thu, May 14

markj updated the diff for D24843: pf: Add a new zone for per-table entry counters..

Don't forget to destroy the zone.

Thu, May 14, 9:51 PM
markj accepted D24790: libalias: retire cuseeme support.

yes please

Thu, May 14, 8:48 PM
markj updated the test plan for D24843: pf: Add a new zone for per-table entry counters..
Thu, May 14, 7:52 PM
markj requested review of D24843: pf: Add a new zone for per-table entry counters..
Thu, May 14, 7:51 PM
markj accepted D24842: Fix r361037..
Thu, May 14, 7:42 PM
markj accepted D22053: Add domainset policy allocation for fpu_kern_ctx.
Thu, May 14, 5:59 PM
markj committed rS361039: Fix the i386 build after r361033..
Fix the i386 build after r361033.
Thu, May 14, 5:57 PM
markj added inline comments to D22053: Add domainset policy allocation for fpu_kern_ctx.
Thu, May 14, 4:41 PM
markj closed D24757: Call acpi_pxm_set_proximity_info() slightly earlier..
Thu, May 14, 4:07 PM
markj committed rS361033: Call acpi_pxm_set_proximity_info() slightly earlier on x86..
Call acpi_pxm_set_proximity_info() slightly earlier on x86.
Thu, May 14, 4:07 PM
markj closed D24756: Allocate UMA counters earlier..
Thu, May 14, 4:07 PM
markj committed rS361032: Allocate UMA per-CPU counters earlier..
Allocate UMA per-CPU counters earlier.
Thu, May 14, 4:07 PM
markj committed rS361031: Assert that page table traversal functions don't operate on superpages..
Assert that page table traversal functions don't operate on superpages.
Thu, May 14, 3:49 PM
markj accepted D22053: Add domainset policy allocation for fpu_kern_ctx.
Thu, May 14, 3:44 PM
markj requested review of D24839: Don't block on the range lock in zfs_getpages()..
Thu, May 14, 3:25 PM

Wed, May 13

markj added reviewers for D24828: Assert that we are not descending through a leaf pointer.: alc, kib.
Wed, May 13, 5:43 PM
markj accepted D24827: riscv: Fix pmap_protect for superpages.
Wed, May 13, 5:11 PM
markj accepted D24819: Fix spurious ENOTCONN from closed unix domain socket other' side..
Wed, May 13, 2:50 PM
markj added a comment to D23848: Reduce object locking in zfs..

I think deadlock at https://lists.freebsd.org/pipermail/freebsd-current/2020-May/076100.html may be related to this but I don't understand a lot of it.

Wed, May 13, 1:40 PM
markj added a comment to D24819: Fix spurious ENOTCONN from closed unix domain socket other' side..

I see the recursion: socantrcvmore_locked() calls sorwakeup_locked(), which may call sowakeup(), which may call soisconnected() if an accept filter decided to accept the connection based on data resident in the receive buffer. In this case, the so_state update is rolled back...

Wed, May 13, 1:03 PM
markj updated the diff for D24758: Introduce a dynamic pcpu layout on amd64..
  • Restore DPCPU indirection in kernel modules for now. The value of DPCPU_START is kld-specific, so DPCPU_BASE_OFFSET() gives the wrong value for dpcpu fields allocated from the kld region (the modspace field).
  • Fix a bug in uma_pcpu_init2(): when marking the bootstrap region as allocated, we have to allocate the region at the same granularity as it gets freed, i.e., a page at a time.
Wed, May 13, 12:43 PM
markj committed rS361008: MFC r360691:.
MFC r360691:
Wed, May 13, 12:34 PM

Tue, May 12

markj added inline comments to D24652: Non-transparent superpages support..
Tue, May 12, 10:25 PM
markj added a comment to D24652: Non-transparent superpages support..

Looking at DPDK, it wants to determine the number of huge pages available in each domain before performing any allocation. I think the only way to get that information today is to read vm.phys_free, which is meant to be human-readable. We could add a set of sysctls vm.domain.<domidx>.largepage.<psind>.free to provide that information.

Tue, May 12, 10:24 PM
markj committed rS360980: Re-enable proc_test:symbol_lookup after r360979..
Re-enable proc_test:symbol_lookup after r360979.
Tue, May 12, 5:06 PM
markj committed rS360979: librtld_db: Fix shlib mapping offsets..
librtld_db: Fix shlib mapping offsets.
Tue, May 12, 5:01 PM
markj committed rS360966: rtwn: Add a USB ID for the TP-Link TL-WN727N..
rtwn: Add a USB ID for the TP-Link TL-WN727N.
Tue, May 12, 4:10 PM
markj added inline comments to D24819: Fix spurious ENOTCONN from closed unix domain socket other' side..
Tue, May 12, 2:55 PM
markj added a comment to D24819: Fix spurious ENOTCONN from closed unix domain socket other' side..

I thought the socket and sockbuf lock already had a well-defined lock order, the socket lock is acquired first. There are many examples in e.g., the poll and kevent code for sockets.

Tue, May 12, 2:21 PM
markj committed rS360962: MFC r360438:.
MFC r360438:
Tue, May 12, 2:01 PM

Mon, May 11

markj committed rS360906: Add a basic table entry counter regression test..
Add a basic table entry counter regression test.
Mon, May 11, 7:08 PM
markj closed D24809: Add a basic table entry counter regression test..
Mon, May 11, 7:08 PM
markj committed rS360905: Remove the svn:executable property from some pf test files..
Remove the svn:executable property from some pf test files.
Mon, May 11, 7:07 PM
markj closed D24803: pf: Don't allocate per-table entry counters unless requested..
Mon, May 11, 6:47 PM
markj committed rS360903: pf: Don't allocate per-table entry counters unless required..
pf: Don't allocate per-table entry counters unless required.
Mon, May 11, 6:47 PM
markj updated the diff for D24809: Add a basic table entry counter regression test..

Remove executable bits.

Mon, May 11, 5:12 PM
markj added a reviewer for D24809: Add a basic table entry counter regression test.: kp.
Mon, May 11, 5:11 PM
markj requested review of D24809: Add a basic table entry counter regression test..
Mon, May 11, 5:11 PM
markj accepted D24780: geom(4) mirror: Do not panic on gmirror(8) insert, resize.

I verified that head panics when running the test suite with kern.geom.mirror.launch_mirror_before_timeout=0, and that it passes with this patch applied.

Mon, May 11, 4:10 PM
markj committed rS360898: MFC r360381:.
MFC r360381:
Mon, May 11, 3:21 PM
markj committed rS360897: MFC r360380:.
MFC r360380:
Mon, May 11, 3:21 PM
markj committed rS360896: MFC r360379:.
MFC r360379:
Mon, May 11, 3:20 PM
markj committed rS360895: MFC r360378:.
MFC r360378:
Mon, May 11, 3:18 PM
markj added a comment to D24803: pf: Don't allocate per-table entry counters unless requested..
In D24803#545731, @kp wrote:

It'd be nice (but it's not required for this to proceed) to have a test to exercise this, i.e. create a table with counters, send some traffic through, check that the counters incremented.

Mon, May 11, 3:07 PM

Sun, May 10

markj updated the summary of D24803: pf: Don't allocate per-table entry counters unless requested..
Sun, May 10, 10:15 PM
markj added reviewers for D24803: pf: Don't allocate per-table entry counters unless requested.: kp, glebius.
Sun, May 10, 10:14 PM
markj requested review of D24803: pf: Don't allocate per-table entry counters unless requested..
Sun, May 10, 10:13 PM

Sat, May 9

markj committed rS360848: rtwn: Add a new USB ID..
rtwn: Add a new USB ID.
Sat, May 9, 2:50 PM

Fri, May 8

markj updated the summary of D24758: Introduce a dynamic pcpu layout on amd64..
Fri, May 8, 3:36 PM
markj updated the diff for D24758: Introduce a dynamic pcpu layout on amd64..
  • Add a vm_phys_seg[] entry for the bootstrap pcpu region.
  • When pcpu_layout() backs the rest of the bootstrap region with 2MB pages, it must subtract 2MB from the allocation size for domain 0, since it is already backed by the initial pcpu bootstrap allocation.
  • Rebase on D24755 to shrink the uma_core.c diff slightly.
Fri, May 8, 3:11 PM
markj updated the diff for D24755: Permit deferred creation of SMR structures for the VM radix zone..

Remove handling of pcpu kegs from zone_kva_available(). Now that such
kegs never use startup_alloc(), we don't need code to switch them from
startup_alloc() to the real pcpu slab allocator.

Fri, May 8, 3:09 PM
markj added inline comments to D24757: Call acpi_pxm_set_proximity_info() slightly earlier..
Fri, May 8, 3:03 PM