Page MenuHomeFreeBSD

kevans (Kyle Evans)
User

Projects

User Details

User Since
Oct 2 2015, 1:17 PM (211 w, 1 d)

Recent Activity

Yesterday

kevans added inline comments to D21315: bhyve: tap: add support for offloads.
Sat, Oct 19, 2:50 PM

Fri, Oct 18

kevans added a comment to D21263: tap: add support for virtio-net offloads.

Il giorno ven 18 ott 2019 alle ore 23:47 kevans (Kyle Evans) <
phabric-noreply@freebsd.org> ha scritto:

kevans added a comment.

In D21263#482518 <https://reviews.freebsd.org/D21263#482518>,

@vmaffione wrote:

> Thanks. Yes, I don't see either any reason why we want to reset device

configuration (vnet hdr, mtu, etc.) on close.

I'm more curious why we don't want to reset it all on close -- once it's

closed, it can be reused by any other software that would then have to
assume the tun/tap device can be in some indeterminate state with respect
to parameters it may not care about at all (e.g. vnethdr, TUNSIFHEAD) that
can affect its usage.

That's right, and if we want to be conservative we should indeed reset the
interface configuration to a "standard state" at tunclose(). The only
potential problem I see is that then it would not be possible to have a
process to configure the tap interface and a separate ones (e.g. maybe
children processes) read/write frames from/to the device. But maybe it's
not something useful, as processes can just pass a tap file descriptor
across fork() and unix sockets.

Fri, Oct 18, 10:22 PM
kevans added a comment to D21263: tap: add support for virtio-net offloads.

Thanks. Yes, I don't see either any reason why we want to reset device configuration (vnet hdr, mtu, etc.) on close.

Fri, Oct 18, 9:47 PM

Thu, Oct 17

kevans updated the diff for D21535: patch(1): give /dev/null patches special treatment.

Drop in a lot more comments that should help explain the funky flow and make the decisions made (incorrect or not =-)) a little more clear. Some of the weird structure is so we can still be verbose about what we're doing without rewriting more than we need to (though a good chunk of it needs a rewrite anyways, because it feels a little unwieldy in spots).

Thu, Oct 17, 11:27 PM
kevans added a comment to D22062: Allow loader.efi to identify non-standard boot setup.

Downstream OS/vendors that don't have /boot/defaults/loader.conf or /boot/kernel/kernel because they store all of this elsewhere as part of a versioned boot scheme (or something to this effect) -- they'd then define PATH_BOOTABLE_TOKEN as another path to check to determine if this is a valid rootfs.

So how does the loader find the loader.conf and kernel to load? Should users set a different PATH_KERNEL?

Thu, Oct 17, 8:29 PM
kevans accepted D22062: Allow loader.efi to identify non-standard boot setup.

Just curious what is this a fix for? I assume some specific type or class of hardware.

Thu, Oct 17, 5:31 PM
kevans updated the diff for D21775: (WIP) Rewrite makesyscalls.sh in Lua.

Prefer multiline string format over multiple subsequent calls to write_line of the same file in most places.

Thu, Oct 17, 1:52 AM

Wed, Oct 16

kevans added a comment to D22057: always add standard kernel configuration include path.

I'd be inclined to explicitly spell it as -I ${KRNLCONFDIR} instead of -I., and then we later rename KRNLCONFDIR to make it more clear. We can't quickly change the cd, at least, because:

Wed, Oct 16, 10:57 PM
kevans committed rS353663: libbe(3): Fix destroy of imported BE w/ AUTOORIGIN.
libbe(3): Fix destroy of imported BE w/ AUTOORIGIN
Wed, Oct 16, 6:33 PM
kevans committed rS353646: bectl(8): destroy: use BE_DESTROY_AUTOORIGIN if -o is not specified.
bectl(8): destroy: use BE_DESTROY_AUTOORIGIN if -o is not specified
Wed, Oct 16, 2:56 PM
kevans committed rS353644: libbe(3): add needed bits for be_destroy to auto-destroy some origins.
libbe(3): add needed bits for be_destroy to auto-destroy some origins
Wed, Oct 16, 2:43 PM
kevans accepted D21263: tap: add support for virtio-net offloads.

I think I'm happy with this. I did wonder if we should reset capabilities on close now that we support SIOCSIFCAP, but we don't reset mtu or other state manipulated by ioctls (TUNSLMODE/TUNSIFHEAD), so I guess that's a "no."

Wed, Oct 16, 1:33 AM

Wed, Oct 9

kevans added a comment to D20293: ucom(4): synchronously execute param changes.

Yes, that's the simplest.
I think it is important we still keep on using a USB task/process queue.
Multiple threads may be exercising the API simultaneously, likely causing USB low level commands out of order!
There may be more than one atomic USB request involved!
Use a SX lock perhaps in TTY layers?

Wed, Oct 9, 2:20 PM

Tue, Oct 8

kevans added inline comments to D21263: tap: add support for virtio-net offloads.
Tue, Oct 8, 6:08 PM
kevans added a comment to D20293: ucom(4): synchronously execute param changes.

Yeah, so, looking at this again- I've walked through a couple of different approaches, and none of them really work because we have to drop the tty/usb lock.

Tue, Oct 8, 5:46 PM
kevans updated the diff for D21903: Add per-process flag to disable logsigexit.

General wordsmithing/relabelling for consistency; no functional change.

Tue, Oct 8, 3:24 AM

Mon, Oct 7

kevans added a comment to D21921: devfs: don't invoke dev_clone for !ISOPEN inquiries.

dev_clone(9) does certainly document it as being invoked for all name lookups; I'm thinking passing the flags along would be more appropriate for this.

Mon, Oct 7, 4:43 PM
kevans created D21921: devfs: don't invoke dev_clone for !ISOPEN inquiries.
Mon, Oct 7, 4:24 PM
kevans committed rS353161: Revert r352557: powerpc/loader: Install ubldr without stripping.
Revert r352557: powerpc/loader: Install ubldr without stripping
Mon, Oct 7, 3:28 AM
kevans committed rS353159: MFS r353157: tuntap(4): loosen up tunclose restrictions.
MFS r353157: tuntap(4): loosen up tunclose restrictions
Mon, Oct 7, 2:57 AM
kevans updated the diff for D21697: Fix MIPS CSU compilation w/ LLVM.

There seems to be some magic at play here, and I don't understand it. I've added the stack space in a reasonable spot, this still builds and passes the csu kyua tests just happily enough for me.

Mon, Oct 7, 2:30 AM
kevans committed rS353157: MFC r353103: tuntap(4): loosen up tunclose restrictions.
MFC r353103: tuntap(4): loosen up tunclose restrictions
Mon, Oct 7, 1:03 AM

Sun, Oct 6

kevans added a comment to D21903: Add per-process flag to disable logsigexit.
In D21903#478626, @kib wrote:

It is impossible to turn logsigexit back to default with proccontrol(1), but this is arguably a wart in the current structure of the arguments parser.

Sun, Oct 6, 3:33 PM
kevans closed D21696: Re-add ALLOW_MIPS_SHARED_TEXTREL, sprinkle it around.
Sun, Oct 6, 4:20 AM
kevans committed rS353140: Re-add ALLOW_MIPS_SHARED_TEXTREL, sprinkle it around.
Re-add ALLOW_MIPS_SHARED_TEXTREL, sprinkle it around
Sun, Oct 6, 4:20 AM
kevans committed rS353139: MFC r348503, r351769: bsdgrep nits.
MFC r348503, r351769: bsdgrep nits
Sun, Oct 6, 4:12 AM
kevans committed rS353138: MFC r352691: bsdgrep(1): fixes of empty pattern/exit code/-c behavior.
MFC r352691: bsdgrep(1): fixes of empty pattern/exit code/-c behavior
Sun, Oct 6, 4:10 AM
kevans committed rS353137: MFC r352559: loader: Respect loader_color=YES for serial consoles.
MFC r352559: loader: Respect loader_color=YES for serial consoles
Sun, Oct 6, 4:02 AM
kevans committed rS353136: MFC r352314: lualoader: Add reload-conf loader command.
MFC r352314: lualoader: Add reload-conf loader command
Sun, Oct 6, 4:01 AM
kevans committed rS353135: MFC r353009: libusb: LIBUSB_DEBUG env var override of libusb_set_debug.
MFC r353009: libusb: LIBUSB_DEBUG env var override of libusb_set_debug
Sun, Oct 6, 3:59 AM
kevans committed rS353134: MFC r352668: cron: log suppression and mail suppression for successful runs.
MFC r352668: cron: log suppression and mail suppression for successful runs
Sun, Oct 6, 3:56 AM
kevans closed D21888: riscv: use the common sub-word {,f}cmpset implementation.
Sun, Oct 6, 1:35 AM
kevans committed rS353131: riscv: use the common sub-word {,f}cmpset implementation.
riscv: use the common sub-word {,f}cmpset implementation
Sun, Oct 6, 1:35 AM

Sat, Oct 5

kevans updated the diff for D21903: Add per-process flag to disable logsigexit.

Some renames to be consistent with the sysctl:

  • sigexitlog -> logsigexit
  • SIGEXIT_LOG -> LOGSIGEXIT
Sat, Oct 5, 10:44 PM
kevans committed rS353129: Remove the remnants of SI_CHEAPCLONE.
Remove the remnants of SI_CHEAPCLONE
Sat, Oct 5, 9:52 PM
kevans committed rS353128: kern_conf: fully initialize cloned devices with make_dev_args, too.
kern_conf: fully initialize cloned devices with make_dev_args, too
Sat, Oct 5, 9:44 PM
kevans closed D21904: kern_conf: fully initialize cloned devices with make_dev_args, too.
Sat, Oct 5, 9:44 PM
kevans added inline comments to D21904: kern_conf: fully initialize cloned devices with make_dev_args, too.
Sat, Oct 5, 7:31 PM
kevans added inline comments to D21904: kern_conf: fully initialize cloned devices with make_dev_args, too.
Sat, Oct 5, 7:23 PM
kevans edited reviewers for D21903: Add per-process flag to disable logsigexit, added: jilles; removed: manpages.

(roping Jilles in, too)

Sat, Oct 5, 5:45 PM
kevans added a comment to D21903: Add per-process flag to disable logsigexit.
In D21903#478452, @kib wrote:

I find it somewhat strange that you require global control to enable local setting. Less cryptic, I would expect to see per-process setting of 3 cases: 1. default/use sysctl, 2. enabled, 3. disabled.

I wasn't entirely sure how important the global control is. This doesn't gate a highly important feature in the first place, so I suppose the answer is "not really, let proccontrol override it all the time."

An application can already override the global control (in most cases) by adding signal handlers. So making it hard for proccontrol to override the global control seems inappropriate.

Sat, Oct 5, 5:44 PM
kevans added a comment to D21263: tap: add support for virtio-net offloads.

I've created some more minor merge conflicts for you in the interim (sorry about that), but they should be trivial to resolve. I have some larger tun/tap changes that may produce more annoying conflicts that I'll shelve until we get this in.

Sat, Oct 5, 5:15 PM
kevans added a comment to D21903: Add per-process flag to disable logsigexit.
In D21903#478452, @kib wrote:

You need to handle compat32.
I find it somewhat strange that you require global control to enable local setting. Less cryptic, I would expect to see per-process setting of 3 cases: 1. default/use sysctl, 2. enabled, 3. disabled.

Sat, Oct 5, 4:26 PM
kevans created D21904: kern_conf: fully initialize cloned devices with make_dev_args, too.
Sat, Oct 5, 12:55 PM

Fri, Oct 4

kevans added inline comments to D21903: Add per-process flag to disable logsigexit.
Fri, Oct 4, 11:09 PM
kevans created D21903: Add per-process flag to disable logsigexit.
Fri, Oct 4, 11:06 PM
kevans committed rS353103: tuntap(4): loosen up tunclose restrictions.
tuntap(4): loosen up tunclose restrictions
Fri, Oct 4, 1:43 PM
kevans committed rP513744: MFH: r513668.
MFH: r513668
Fri, Oct 4, 2:42 AM
kevans committed rS353095: caroot: commit initial bundle.
caroot: commit initial bundle
Fri, Oct 4, 2:34 AM
kevans created D21894: Convert in-tree sysent targets to use new makesyscalls.lua.
Fri, Oct 4, 2:11 AM
kevans updated the diff for D21775: (WIP) Rewrite makesyscalls.sh in Lua.

Minimize diff to just makesyscalls.lua; the flua bits are getting broken out into a parent review, and converting sysent targets into a child.

Fri, Oct 4, 2:05 AM
kevans created D21893: Add flua to the base system, install to /usr/libexec.
Fri, Oct 4, 2:04 AM

Thu, Oct 3

kevans committed rS353070: certctl(8): let one blacklist based on hashed filenames.
certctl(8): let one blacklist based on hashed filenames
Thu, Oct 3, 8:46 PM
kevans committed rS353066: certctl(8): realpath the file before creating the symlink.
certctl(8): realpath the file before creating the symlink
Thu, Oct 3, 8:06 PM
kevans created D21888: riscv: use the common sub-word {,f}cmpset implementation.
Thu, Oct 3, 7:25 PM
kevans added a comment to D21682: powerpc: Implement atomic_(f)cmpset_ for short and char.

I have applied this patch on top of yours: https://people.freebsd.org/~kevans/ppc-nonisa206.diff -- to just have powerpc/ use the common sys/_atomic_subword.h version entirely if ISA_206_ATOMICS is not defined. That diff also includes atomic_fcmpset_{,acq_,rel_}{8,16} #defines, as those were missed here.

Thu, Oct 3, 6:06 PM
kevans committed rS353057: if_tuntap: create /dev aliases when a tuntap device gets renamed.
if_tuntap: create /dev aliases when a tuntap device gets renamed
Thu, Oct 3, 5:54 PM
kevans committed rS353056: if_tuntap: add a busy/unbusy mechanism, replace destroy OPEN check.
if_tuntap: add a busy/unbusy mechanism, replace destroy OPEN check
Thu, Oct 3, 5:46 PM
kevans committed rP513668: net/ocserv: fix tun handoff between parent and worker process.
net/ocserv: fix tun handoff between parent and worker process
Thu, Oct 3, 2:46 PM
kevans committed rS353046: MFS r353041: fdt_slicer: bump to SI_ORDER_THIRD following r347183.
MFS r353041: fdt_slicer: bump to SI_ORDER_THIRD following r347183
Thu, Oct 3, 2:27 PM
kevans committed rS353041: MFC r352863: fdt_slicer: bump to SI_ORDER_THIRD following r347183.
MFC r352863: fdt_slicer: bump to SI_ORDER_THIRD following r347183
Thu, Oct 3, 12:47 PM

Wed, Oct 2

kevans added a comment to D21626: vfs: scale foffset_lock.

Sure, that seems reasonable. I mostly don't care about that as long as the ILP32 unlocking part is fixed.=)

Wed, Oct 2, 10:24 PM
kevans added a reviewer for D21626: vfs: scale foffset_lock: kevans.
Wed, Oct 2, 9:37 PM
kevans added a comment to D21626: vfs: scale foffset_lock.

This patch is not quite right.

Wed, Oct 2, 9:31 PM
kevans updated the diff for D21696: Re-add ALLOW_MIPS_SHARED_TEXTREL, sprinkle it around.
  • One last instance of "yes" where the value doesn't matter has been fixed.
  • Add enough explanation to make it clear that this won't spontaneously go away due to work in the FreeBSD tree, but rather from clang/lld revisions.
Wed, Oct 2, 5:54 PM
kevans updated the diff for D21805: freebsd-update: rehash certs as needed.

Take #2; still know very little about freebsd-update. =-)

Wed, Oct 2, 5:32 PM
kevans committed rS353019: Override the TLS model when building mips64 binaries and static libraries.
Override the TLS model when building mips64 binaries and static libraries
Wed, Oct 2, 5:18 PM
kevans closed D21699: Hack around incorrect TLS defaults for mips64.
Wed, Oct 2, 5:18 PM
kevans closed D21698: usr.bin/clang: use -mxgot for MIPS.
Wed, Oct 2, 5:15 PM
kevans committed rS353018: clang: use -mxgot for 32-bit mips.
clang: use -mxgot for 32-bit mips
Wed, Oct 2, 5:15 PM
kevans committed rS353017: sparc64: use generic sub-word atomic *cmpset.
sparc64: use generic sub-word atomic *cmpset
Wed, Oct 2, 5:08 PM
kevans committed rS353016: mips: use generic sub-word atomic *cmpset.
mips: use generic sub-word atomic *cmpset
Wed, Oct 2, 5:08 PM
kevans closed D21822: Provide generic sub-word atomic *cmpset.
Wed, Oct 2, 5:06 PM
kevans committed rS353015: Provide generic sub-word atomic *cmpset.
Provide generic sub-word atomic *cmpset
Wed, Oct 2, 5:06 PM
kevans committed rS353009: libusb: LIBUSB_DEBUG environment variable override of libusb_set_debug.
libusb: LIBUSB_DEBUG environment variable override of libusb_set_debug
Wed, Oct 2, 3:19 PM
kevans closed D21877: libusb: let LIBUSB_DEBUG environment variable override libusb_set_debug.
Wed, Oct 2, 3:19 PM
kevans committed rS353008: mips: fcmpset: do not spin on sc failure.
mips: fcmpset: do not spin on sc failure
Wed, Oct 2, 3:14 PM
kevans closed D21836: mips: fcmpset: do not spin on sc failure.
Wed, Oct 2, 3:14 PM
kevans added inline comments to D21877: libusb: let LIBUSB_DEBUG environment variable override libusb_set_debug.
Wed, Oct 2, 2:12 PM
kevans updated the summary of D21877: libusb: let LIBUSB_DEBUG environment variable override libusb_set_debug.
Wed, Oct 2, 1:47 PM
kevans created D21877: libusb: let LIBUSB_DEBUG environment variable override libusb_set_debug.
Wed, Oct 2, 1:43 PM
kevans committed rS353002: Unbreak etcupdate(8) and mergemaster(8) after r352950.
Unbreak etcupdate(8) and mergemaster(8) after r352950
Wed, Oct 2, 12:46 PM
kevans committed rS352952: shm_open2(2): completely unbreak.
shm_open2(2): completely unbreak
Wed, Oct 2, 2:37 AM
kevans committed rS352951: caroot: add @generated tags to extracted .pem.
caroot: add @generated tags to extracted .pem
Wed, Oct 2, 1:27 AM
kevans committed rS352950: [3/3] etcupdate and mergemaster support for certctl.
[3/3] etcupdate and mergemaster support for certctl
Wed, Oct 2, 1:06 AM
kevans closed D17389: WIP: Add support for certctl(8) to etcupdate and mergemaster.
Wed, Oct 2, 1:06 AM
kevans committed rS352949: [2/3] Add certctl(8).
[2/3] Add certctl(8)
Wed, Oct 2, 1:06 AM
kevans closed D16857: Introduce certctl(8).
Wed, Oct 2, 1:06 AM
kevans committed rS352948: [1/3] Initial infrastructure for SSL root bundle in base.
[1/3] Initial infrastructure for SSL root bundle in base
Wed, Oct 2, 1:06 AM
kevans closed D16856: Initial proposed patch for SSL root bundle in base.
Wed, Oct 2, 1:06 AM

Tue, Oct 1

kevans committed rS352935: Move httpd to simple_httpd....
Move httpd to simple_httpd...
Tue, Oct 1, 6:14 PM
kevans added inline comments to D21845: linuxolator: implement memfd_create syscall.
Tue, Oct 1, 4:57 PM
kevans added a comment to D21724: Move simple_httpd out of picobsd, add HTTPD option.

Wouldn't /usr/libexec be a better place for it, like for many other such daemons (like fingerd etc…)?

Tue, Oct 1, 4:52 PM
kevans abandoned D21681: mips: incredibly naive attempt at 8/16-bit atomics (set/clear/add/sub/cmpset/fcmpset).

Abandoning in favor of just implementing the atomics needed now in C in terms of atomic_*cmpset_32 (D21822)

Tue, Oct 1, 3:22 PM
kevans committed rS352929: Add a top-level makeman target.
Add a top-level makeman target
Tue, Oct 1, 3:10 PM
kevans committed rS352928: Regen src.conf(5) after r352927 (HTTPD option added).
Regen src.conf(5) after r352927 (HTTPD option added)
Tue, Oct 1, 2:56 PM
kevans committed rS352927: Move simple_httpd out of picobsd, add HTTPD option (default OFF).
Move simple_httpd out of picobsd, add HTTPD option (default OFF)
Tue, Oct 1, 2:55 PM
kevans closed D21724: Move simple_httpd out of picobsd, add HTTPD option.
Tue, Oct 1, 2:55 PM

Mon, Sep 30

kevans updated the diff for D21845: linuxolator: implement memfd_create syscall.

I waffled on things like this, and I think I ended up with a solution like this...

Mon, Sep 30, 9:46 PM
kevans updated the diff for D21822: Provide generic sub-word atomic *cmpset.

Stop mucking with sys/systm.h, drop all assertions and fix the s/HWORD/BYTE/ typo that snuck in.

Mon, Sep 30, 3:57 PM