IABSD.fr/src/sys
Branch :
Log
| Author | Commit | Date | CI | Message |
|---|---|---|---|---|
 |
dbfd42af | 2025-04-26 17:49:55 | Revert previous commit; it doesn't build. | |
 |
30411ea8 | 2025-04-26 13:59:31 | Fix i386 build. SOFTINTR_* definitions should be SOFTINTR_SOFT*. No functional changes. | |
 |
58cc3164 | 2025-04-26 13:58:08 | Run TCP input in parallel on multiple CPUs. Mark the protocol input function tcp_input() as MP-safe. Then it is called directly from the IP deliver loop with shared net lock. Do not enqueue TCP packets to wait for exclusive net lock. This results in more contention on the socket lock. Throughput optimization for that problem could be done later. tested by Mark Patruck; OK mvs@ | |
 |
b49d59c7 | 2025-04-26 11:10:28 | powerpc64: Use MI soft interrupt code Tested by gkoehler@ (previous version) and kettenis@ OK kettenis@ | |
|
993eaad3 | 2025-04-26 11:01:55 | riscv64: Use MI soft interrupt code Tested by and OK jca@ | |
|
92407eb4 | 2025-04-26 10:51:48 | Remove unused softintr.[ch] files The dust is possibly starting to settle after the initial MI soft interrupt migrations. Remove the now unused softintr.[ch] files, before me or someone else gets confused. | |
|
d4f884f6 | 2025-04-25 19:10:50 | psp(4) adds ioctl(2) to encrypt and measure state for AMD SEV-ES. When launching a SEV-ES enabled VM, vmd(8) will have to encrypt the initial state of the vCPU in the VM Save Area (VMSA) allocated by vmm(4). In addition the VMSA has to be measured by the PSP. Implement functions psp_encrypt_state() and psp_launch_update_vmsa() used by new ioctl PSP_IOC_ENCRYPT_STATE. from hshoexer@ | |
|
096a5827 | 2025-04-25 12:48:48 | Make arm64 use the MI soft interrupt code. ok visa kettenis | |
|
50f0e847 | 2025-04-25 12:47:37 | Make i386 use the MI soft interrupt code. ok visa | |
|
f3498c2c | 2025-04-23 21:54:12 | vmm(4) provides VMSA for AMD SEV-ES For SEV-ES the hypervisor has to provide a VM Save Area (VMSA). It will be used by the CPU to load and store encrypted guest CPU state. Also supply a helper function to retrieve the physical address of the VMSA. Non SEV-ES VMs are not affected by this change. from hshoexer@; OK dv@ | |
|
31338944 | 2025-04-23 21:40:22 | Polish the qcscm(4) turd a bit. ok deraadt@, mglocker@ | |
|
9094b91d | 2025-04-23 17:52:12 | Refactor TCP softlro extract header. The header of the tail mbuf has already been extraced. Do all header extraction in tcp_softlro_glue() and pass them down to tcp_softlro(). Check port number before address as IPv6 address comparison is expensive. The empty segment check has already been done in tcp_softlro_check(), remove it from tcp_softlro(). OK jan@ | |
|
9244c012 | 2025-04-23 15:08:05 | amd64: Use MI soft interrupt code Make amd64 use the MI soft interrupt code. File sys/arch/amd64/amd64/softintr.c is now unlinked from the build. It is kept in the source tree a little longer in case problems arise in the coming days. OK mvs@ | |
|
15f0531d | 2025-04-23 15:07:00 | Add MI high-level software interrupt dispatcher This provides a common subsystem for the high-level allocation, scheduling, and dispatching of soft interrupts. This will reduce duplication in the MD code, improve consistency, and hopefully ease maintenance and future development, once all architectures have been migrated. MP-readiness is built-in. When a soft interrupt is allocated with IPL_MPSAFE, the handler will be called without the kernel lock. Deallocation has an automatic barrier that blocks until the soft interrupt has stopped. As an extra feature, the dispatcher serializes each soft interrupt so that the handler does not execute on more than one CPU at a time. A typical MP-safe soft interrupt handler probably uses some kind of locking anyway, so this serialization should reduce needless spinning. This feature can be made optional later if necessary. Input from cheloha@ and kettenis@ OK mvs@ | |
|
1b07127f | 2025-04-22 22:34:27 | Fix whitespace introduced in previous commit. | |
|
e684669a | 2025-04-22 19:59:55 | Refactor TCP softlro checks. In TCP software LRO do all checks that affect only a single packet before looping over the enqueued packets. Fragment and TCP protocol checks can be removed, ether_extract_headers() already does that. After the check is successful, store the payload length in ph_mss. This is done later anyway. If this mbuf field contains a positive length, we know that the check has already been done. OK jan@ | |
 |
49d9704b | 2025-04-22 05:55:18 | drm/amdgpu: grab an additional reference on the gang fence v2 From Christian Koenig fb4c507bf24a77e210ed68924c48a1ff7eeff94f in linux-6.12.y/6.12.24 0d9a95099dcb05b5f4719c830d15bf4fdcad0dc2 in mainline linux | |
|
246c2791 | 2025-04-22 05:53:21 | drm/amdgpu: Fix the race condition for draining retry fault From Emily Deng e64be12f8401819662e608efa247638b61d023cd in linux-6.12.y/6.12.24 f844732e3ad9c4b78df7436232949b8d2096d1a6 in mainline linux | |
|
fb17fff5 | 2025-04-22 05:51:11 | drm/amdgpu: handle amdgpu_cgs_create_device() errors in amd_powerplay_create() From Wentao Liang dc4380f34613eaae997b3ed263bd1cb3d0fd0075 in linux-6.12.y/6.12.24 1435e895d4fc967d64e9f5bf81e992ac32f5ac76 in mainline linux | |
|
6b993d91 | 2025-04-22 05:49:27 | drm/amdkfd: debugfs hang_hws skip GPU with MES From Philip Yang 1a322b330dc0b775d1d7a84e55c752d9451bfe7d in linux-6.12.y/6.12.24 fe9d0061c413f8fb8c529b18b592b04170850ded in mainline linux | |
|
8873455d | 2025-04-22 05:48:06 | drm/amdkfd: Fix pqm_destroy_queue race with GPU reset From Philip Yang 10ce36501f1e6f54e046f7cd35683a87225e74b0 in linux-6.12.y/6.12.24 7919b4cad5545ed93778f11881ceee72e4dbed66 in mainline linux | |
|
90bb15b7 | 2025-04-22 05:46:28 | drm/amdkfd: Fix mode1 reset crash issue From Philip Yang ffd37d7d44d7e0b6e769d4fe6590e327f8cc3951 in linux-6.12.y/6.12.24 f0b4440cdc1807bb6ec3dce0d6de81170803569b in mainline linux | |
|
20a99bdd | 2025-04-22 05:44:52 | drm/amdkfd: clamp queue size to minimum From David Yat Sin 96757c085bd94495ebe4069807f9d35d6d2c75f8 in linux-6.12.y/6.12.24 e90711946b53590371ecce32e8fcc381a99d6333 in mainline linux | |
|
78f04f32 | 2025-04-22 05:43:17 | drm/amd/display: stop DML2 from removing pipes based on planes From Mike Katsnelson c9323cbc94d566dce7f2717b7b4f143250cbdac2 in linux-6.12.y/6.12.24 8adeff83a3b07fa6d0958ed51e1b38ba7469e448 in mainline linux | |
|
612010f0 | 2025-04-22 05:41:04 | drm/bridge: panel: forbid initializing a panel with unknown connector type From Luca Ceresoli b22cb42a5ee1d708ec7ac2bc8475399efb7a8a19 in linux-6.12.y/6.12.24 b296955b3a740ecc8b3b08e34fd64f1ceabb8fb4 in mainline linux | |
|
9b8498e3 | 2025-04-22 05:38:54 | drm/debugfs: fix printk format for bridge index From Luca Ceresoli f04612890c56d00134c50ca99891b89472a89eea in linux-6.12.y/6.12.24 72443c730b7a7b5670a921ea928e17b9b99bd934 in mainline linux | |
|
100e037b | 2025-04-22 05:37:01 | drm: panel-orientation-quirks: Add quirk for OneXPlayer Mini (Intel) From Andrew Wyatt ba5a998f84cdaa0e15c6ec7958fe3226b754e0f3 in linux-6.12.y/6.12.24 b24dcc183583fc360ae0f0899e286a68f46abbd0 in mainline linux | |
|
0ac00827 | 2025-04-22 05:35:18 | drm: panel-orientation-quirks: Add new quirk for GPD Win 2 From Andrew Wyatt a64e0974266e449963a11be2646aa0aa7b999f1f in linux-6.12.y/6.12.24 a860eb9c6ba6cdbf32e3e01a606556e5a90a2931 in mainline linux | |
|
b9fa4d95 | 2025-04-22 05:33:49 | drm: panel-orientation-quirks: Add quirk for AYA NEO Slide From Andrew Wyatt 5dd6fdb88953e0c9a60ce882e0e6a695d383486e in linux-6.12.y/6.12.24 132c89ef8872e602cfb909377815111d121fe8d7 in mainline linux | |
|
e2bff5fa | 2025-04-22 05:32:00 | drm: panel-orientation-quirks: Add quirks for AYA NEO Flip DS and KB From Andrew Wyatt 6fe4ed94ee82ff769907c16f26e99f27a925725a in linux-6.12.y/6.12.24 529741c331da1fbf54f86c6ec3a4558b9b0b16dc in mainline linux | |
 |
a2a57b37 | 2025-04-22 05:31:49 | the recent commit for qcscm_acpi breaks configure of systems with acpi, but without fdt. Move "device qcscm" to the MI file, to handle this ok mglocker miod | |
|
10255204 | 2025-04-22 05:30:14 | drm: panel-orientation-quirks: Add support for AYANEO 2S From Andrew Wyatt df33b535f0deb965fcf8555c1dc11db4b3bbc0ef in linux-6.12.y/6.12.24 eb8f1e3e8ee10cff591d4a47437dfd34d850d454 in mainline linux | |
|
924ccdfb | 2025-04-22 05:28:14 | drm/amdgpu: Unlocked unmap only clear page table leaves From Philip Yang 357ba4ed6980afe236b961d5bcedc47c22e10f93 in linux-6.12.y/6.12.24 23b645231eeffdaf44021debac881d2f26824150 in mainline linux | |
|
57218313 | 2025-04-22 05:25:09 | drm/amd/display: Update Cursor request mode to the beginning prefetch always From Zhikai Zhai c15a9c84494c1e215a0d26526992e18a86420c84 in linux-6.12.y/6.12.24 4a4077b4b63a8404efd6d37fc2926f03fb25bace in mainline linux | |
|
293edf9a | 2025-04-22 05:22:54 | drm/xe/bmg: Add new PCI IDs From Shekhar Chauhan 4a5f14246bc484db3d06969ddd9831e6bdcf9a29 in linux-6.12.y/6.12.24 fa8ffaae1b15236b8afb0fbbc04117ff7c900a83 in mainline linux | |
|
e0054a1c | 2025-04-22 05:20:52 | drm: allow encoder mode_set even when connectors change for crtc From Abhinav Kumar 044c1b3528413bdf02711df0e6fb37ef3c5adf6f in linux-6.12.y/6.12.24 7e182cb4f5567f53417b762ec0d679f0b6f0039d in mainline linux | |
|
27d75587 | 2025-04-22 05:18:45 | drm/i915/huc: Fix fence not released on early probe errors From Janusz Krzysztofik c5a906806162aea62dbe5d327760ce3b7117ca17 in linux-6.12.y/6.12.24 e3ea2eae70692a455e256787e4f54153fb739b90 in mainline linux | |
|
6aa66d75 | 2025-04-22 05:14:57 | drm/i915: Disable RPG during live selftest From Badal Nilawar 9ddc7edc558acef8dd4f645af3683b226e2b590c in linux-6.12.y/6.12.24 9d3d9776bd3bd9c32d460dfe6c3363134de578bc in mainline linux | |
|
01e0d42d | 2025-04-22 04:53:50 | remove unused P_ZOMBIE() define ok mpi@ | |
 |
c629b831 | 2025-04-21 21:11:04 | Make qcscm(4) attach at acpi(4). This lets qcom machines which use qcscm(4) also access EFI variables in ACPI mode. Some arm64/qcom machines, like for example the Samsung Galaxy Book4 Edge, require to set the EFI variables during the installation procedure in ACPI mode, so that the BIOS can successfully boot OpenBSD. As agreed with deraadt@, this will make it in to the 7.7 release to fix the installation procedure of the impacted arm64/qcom machines. Input from miod@. Tested by kettenis@, kirill@. Collaboration and ok kettenis@ | |
 |
e8eeea37 | 2025-04-21 14:46:18 | Return EPERM if uvm_map_immutable hits a MAP_INHERIT_ZERO entry. Discussed with deraadt@. Plan is to get this in early in release cycle. | |
|
f427bb86 | 2025-04-21 09:54:53 | Consistent naming for TCP softtso and softlro. If the hardware does not support TSO or LRO, an alternative implementation in software is used. Call these functions tcp_softlro_glue() and tcp_softtso_chop() and use the same order of paramters. OK jan@ | |
|
cfa610fa | 2025-04-20 09:04:10 | RK3528 support. ok jsg@ | |
 |
8c061c0b | 2025-04-19 19:35:32 | sys/uvideo: avoid one bcopy for the mmap'ed buffer Tested by ian@ OK and collaboration: mglocker@ | |
|
a84637d4 | 2025-04-19 04:12:36 | remove unused SO_NETPROC define tested in ports bulk by tb@, ok tb@ kirill@ deraadt@ beck@ | |
|
37ecd33d | 2025-04-17 17:23:17 | Describe software LRO and TSO counters consistently. The netstat -s -p tcp description was not quite right anymore. We have one counter each when we glue or chop TCP in software, and for offloading. The loopback pseudo device is special as it just passes large packets. Do not count such non-existing small packets. OK jan@ | |
 |
23635783 | 2025-04-17 12:42:50 | typo: property -> properly From Tim Kuijsten | |
|
a9792827 | 2025-04-17 12:01:26 | change some if statements to switch cases ok deraadt@ miod@ | |
|
a1855bdb | 2025-04-17 09:51:55 | sys/qwz: fix build with enabled QWZ_DEBUG OK: stsp@ | |
 |
d908cbc5 | 2025-04-16 17:17:06 | Add a software implementation of TCP Large Receive Offload. This diff adds a SoftLRO implementation in tcp_input.c as its counter part tcp_chopper() (which is a SoftTSO implementation) in tcp_output.c. We just use SoftLRO in ixl(4) for now, but default off. Because of some unclear bugs in ixl(4) and our network stack. The mbuf chain length produced by SoftLRO is limited to a maximum of 8. This avoids m_defrag() calls in which leads to races in ixl(4) tx/rx interrupts and mbuf handling. We also use the packet type field in the receive descriptors to differentiate between TCP and other Packets. So, we have two mbuf lists and non-TCP traffic it not influenced by SoftLRO. This it not necessary for all drivers, but helps to avoid a decrease of UDP bulk traffic. Thanks to all the testers: Mark Patruck, Yuichiro NAITO, Hrvoje Popovski jj@, lucas@ and bluhm@. With tweaks from dlg@ and bluhm@. ok bluhm@ | |
|
96fb1b48 | 2025-04-16 12:51:11 | Take socket lock in TCP input. In preparation to run tcp_input() in parallel, the socket has to be locked while processing incoming TCP packets. After inpcb lookup, in addition take the socket lock and increase the socket reference counter. The function in_pcbsolock_ref() upgrades a from a locked inpcb to a locked socket by using mutex and refcount. syn_cache_get() unlocks the listen socket and returns a locked socket, syn_cache_add() relies on socket lock. With this commit, exclusive net lock is still held. TCP thoughput gets slower by 6% due to the additional mutex and refcount. But I want to see if locking and refcount works, before switching tcp_input() to shared net lock. Running TCP in parallel will more than compensate the cost of locking. tested as part of parallel TCP input by Mark Patruck, Hrvoje Popovski OK mvs@ | |
 |
4c8bb646 | 2025-04-16 09:16:48 | Use pool caches for "struct vm_anon". ok bluhm@, kettenis@ | |
|
42337cd2 | 2025-04-15 12:14:06 | Release `sb_mtx' mutex(9) while doing sleeping m_copym(..., M_WAIT) in soreceive() and somove(). It is possible in both places. We copy only `len' bytes from the single mbuf(9) pointed by `m'. The `len' is always less than m->m_len. The m->m_len could only grow while `sb_mtx' is unlocked, but concurrent thread will not override our chunk of m->m_data. There is no difference with the lockless uiomove(mtod(m)). Reported-by: syzbot+6cac839a17bc8be499d7@syzkaller.appspotmail.com ok bluhm | |
|
bb834979 | 2025-04-15 06:44:37 | sys/video: avoid call video_stop multiple times OK: mpi@ | |
|
2766c3c8 | 2025-04-15 05:51:51 | remove unused vop_generic_kqfilter() ok mlarkin@ | |
 |
03c532ca | 2025-04-14 20:02:34 | pf: Count m_gethdr() failures in PFRES_MEMORY counter This requires passing the reason pointer down into pf_build_tcp(). ok bluhm@ | |
|
4c24f8f9 | 2025-04-14 12:53:54 | Remove unused flags. | |
|
5693b799 | 2025-04-14 09:15:24 | Set ps_mtx child lock for witness(4) This avoids duplicate lock error with ps_mtx when process_suspend_signal() calls prsignal(). OK mpi@ claudio@ | |
|
d9ea35f5 | 2025-04-14 09:14:51 | witness: Improve handling of lock nesting Add an option to specify a parent/child lock relation for a lock object. This lets witness(4) handle the nesting of locks that have the same type without using LO_DUPOK or LOP_DUPOK. An example: The states of objects o1 and o2 are protected by locks o1->o_mtx and o2->o_mtx, respectively. The objects are hierarchical and there is an operation that requires keeping them both locked at the same time. Assume the hierarchy mandates that o1 is locked first. The duplicate locking error can now be avoided by using WITNESS_SETCHILD(&o1->o_mtx.lock_object, &o2->o_mtx.lock_object) or WITNESS_SETPARENT(&o2->o_mtx.lock_object, &o1->o_mtx.lock_object) after the lock initializer. The choice between WITNESS_SETCHILD() and WITNESS_SETPARENT() depends on whether the object access pattern is many-to-one or one-to-many from locking perspective; the implementation allows only one predefined parent/child relation per lock. OK mpi@, claudio@ (earlier version) | |
|
8f39de42 | 2025-04-14 09:14:13 | Remove unnecessary firmware version check for gc v9_4_2 From Candice Li 0c121f2065e44ae055e2661f05b4cfe6f6d6c500 in linux-6.12.y/6.12.23 5b3c08ae9ed324743f5f7286940d45caeb656e6e in mainline linux | |
|
f460bca6 | 2025-04-14 09:11:59 | drm/amdgpu/gfx12: fix num_mec From Alex Deucher d3f0a68b2914061e20b004b488034010d83d0be1 in linux-6.12.y/6.12.23 dce8bd9137b88735dd0efc4e2693213d98c15913 in mainline linux | |
|
faf5dc02 | 2025-04-14 09:10:22 | drm/amdgpu/gfx11: fix num_mec From Alex Deucher 2ac69453e9e5c24ebb55abe15e075d59b6ca01da in linux-6.12.y/6.12.23 4161050d47e1b083a7e1b0b875c9907e1a6f1f1f in mainline linux | |
|
9ecdddf0 | 2025-04-14 09:09:07 | drm/amd: Keep display off while going into S4 From Mario Limonciello cf1b90486748fc7251794074aeacb779ac9befb2 in linux-6.12.y/6.12.23 4afacc9948e1f8fdbca401d259ae65ad93d298c0 in mainline linux | |
|
654bf9f9 | 2025-04-14 09:07:09 | drm/amd/display: avoid NPD when ASIC does not support DMUB From Thadeu Lima de Souza Cascardo 3453bcaf2ca92659346bf8504c2b52b3993fbd79 in linux-6.12.y/6.12.23 42d9d7bed270247f134190ba0cb05bbd072f58c2 in mainline linux | |
|
c5a62f9c | 2025-04-14 09:04:39 | drm/amd/display: fix type mismatch in CalculateDynamicMetadataParameters() From Vitaliy Shevtsov f42a78df889406efa7721db65a3c289d6afd0f15 in linux-6.12.y/6.12.23 c3c584c18c90a024a54716229809ba36424f9660 in mainline linux | |
|
482f6738 | 2025-04-14 09:02:01 | drm/amdkfd: Fix Circular Locking Dependency in 'svm_range_cpu_invalidate_pagetables' From Srinivasan Shanmugam c0ef1c8ef70bfa3487ea84f794d7a1f0fea0735d in linux-6.12.y/6.12.23 fddc45026311c05a5355fd34b9dc0a1d7eaef4a2 in mainline linux | |
|
df295510 | 2025-04-14 08:59:49 | drm/amd/display: fix an indent issue in DML21 From Aurabindo Pillai a99219bbd69a2e9421e626f6b908d919a37b30e1 in linux-6.12.y/6.12.23 a1addcf8499a566496847f1e36e1cf0b4ad72a26 in mainline linux | |
|
c8bc4fd0 | 2025-04-14 08:58:13 | drm/amdgpu/umsch: fix ucode check From Alex Deucher f2b099e945ee9961bf84120ccd770a982d68046a in linux-6.12.y/6.12.23 c917e39cbdcd9fff421184db6cc461cc58d52c17 in mainline linux | |
|
9319eced | 2025-04-14 08:56:32 | drm/amdgpu: refine smu send msg debug log format From Yang Wang 2dbf9e3efcc0abacddd1f945717158242444d460 in linux-6.12.y/6.12.23 8c6631234557515a7567c6251505a98e9793c8a6 in mainline linux | |
|
b20ad962 | 2025-04-14 08:54:47 | drm/dp_mst: Fix drm RAD print From Wayne Lin 69e90c1e33918877d3916a26b82febc13808d0da in linux-6.12.y/6.12.23 6bbce873a9c97cb12f5455c497be279ac58e707f in mainline linux | |
 |
69e74f77 | 2025-04-14 05:37:58 | Add support for the Realtek RTL8125D chip to the rge(4) driver and update microcode for RTL8125B. The RTL8125D chip can be found on motherboards like MSI PRO B840-P WIFI. | |
|
7d1180e9 | 2025-04-14 01:41:34 | re-enable POOL_DEBUG | |
|
875b602d | 2025-04-14 01:39:33 | drm/amd/display: Don't write DP_MSTM_CTRL after LT From Wayne Lin 2c1674fb52b2e1a1d7be78888f1a747e70c05474 in linux-6.12.y/6.12.22 bc068194f548ef1f230d96c4398046bf59165992 in mainline linux | |
|
b0164c94 | 2025-04-13 09:52:43 | remove unused b_active define; ok mpi@ miod@ | |
|
fe056491 | 2025-04-12 02:13:14 | now working on 7.7-current | |
 |
ade9dbe6 | 2025-04-10 07:06:44 | always call ifq_restart() if ice_txeof() makes space on a full Tx ring bluhm's udpbench tests have exposed a race where ice_txeof() makes space on a full (OACTIVE) Tx ring, yet for some reason the OACTIVE flag is not set while the interrupt handler is checking it. The interrupt handler would then fail to call ifq_restart(), and the driver would get stuck in OACTIVE state with no more traffic passing until ifconfig down/up. This patch is a workaround which makes all of bluhm's stress tests pass. I will keep trying to find a better fix, but this workaround is good enough for release. | |
|
d4602f96 | 2025-04-08 15:31:22 | Get rid of "...tcp(4) sockets rely on exclusive solock..." commentary. Since the shared solock() permanently landed to the TCP sockets area, this comment is not actual anymore. No functional changes. ok bluhm | |
|
b462fe85 | 2025-04-08 07:36:02 | place interrupt-, timeout-, and ifq-barriers into ice_down() | |
|
64f9882f | 2025-04-08 07:27:58 | allow up to 8 DMA segments per frame on ice(4), like xl(4) does Reduces use of m_defrag(), which was quite high according to testing done by bluhm@. In my testing, TCP Tx throughput increases from 4.5 Gbit/s to 6 Gbit/s. | |
|
f2dcb370 | 2025-04-07 17:37:31 | Disable lazy cr3 switching. The current implementation isn't compatible with the locking that's done to make the pmap mpsafe. ok sthen@, mpi@ | |
 |
98b93c7c | 2025-04-07 15:43:00 | Fix cpu idle percentage in top(1) on macppc I broke it in October 2024 (locore.S r1.66), when I missed that the depth of a clock interrupt on an idle cpu changed from 0 to 1. ok kettenis@ miod@ jca@ | |
|
d70fab9c | 2025-04-07 12:26:06 | use a signed type when testing OF_getproplen() return value found by smatch, ok miod@ tobhe@ | |
|
2dd7be35 | 2025-04-06 20:20:11 | Fix a (mostly) hypothetical race in pinsyscalls(2) by making it return an error if called in a multi-threaded process. The race would involve an alternate ld.so implementation that starts threads before loading libc.so, which is just a bad idea. Several ideas to fix the race have been bounced around but those either involve an additional per-syscall cost or some convoluted variable dependencies (that would still require an explicit barrier on alpha). That's simply not worth it. ok deraadt@ | |
|
84d0464d | 2025-04-06 09:46:30 | sys/uvideo: revert uvideo.c,1.256 for the release This perfromance optimization is required to support 60 FPS streaming, but it introduces a regression for ian@ use case of multimedia/motion. OK: mglocker@ | |
|
78c2114e | 2025-04-06 00:37:07 | use ansi style functions; build test on hppa and ok miod@ | |
|
61acf678 | 2025-04-04 12:50:07 | enable MSI-X interrupts for ice(4) Tx queues I missed doing this because the FreeBSD driver doesn't enable them, even though it contains code to enable them. Fixes ice(4) getting stuck in OACTIVE if traffic is sent while no return traffic is received, e.g. when doing UDP-only Tx tests. Issue reported by bluhm@ | |
|
68e43bfa | 2025-04-04 12:48:32 | clear the OACTIVE flag on tx queues when ixl(4) is reset ice(4) inherited this problem from ixl(4), so fix it here, too | |
|
6dbbcb15 | 2025-04-04 12:46:35 | clear the OACTIVE flag on tx queues when ice(4) is reset issue noticed by bluhm@ | |
|
e674ff4e | 2025-04-03 11:02:44 | sys/usb: fix potential buffer overflow from oversized USB chunks If a device announces it will send n bytes but then actually sends a chunk larger than n bytes, a kernel without DIAGNOSTIC appears to overwrite memory past the buffer boundary. Conversely, with DIAGNOSTIC enabled, this memory overwrite does not occur. OK: mpi@ | |
|
1337afad | 2025-04-01 09:12:39 | Apparently to read the counters we need to write the full DMA address before setting the command bit. The current code causes hangs on a Radxa Orion O6 board with onboard RTL8126 chips. ok dlg@, kevlo@ | |
|
fafc58a3 | 2025-04-01 08:43:33 | Use acpipci(4) on hypervisors If the hypervisor cpuid bit is set, use acpipci to attach PCI busses. As virtualization is not that old, we can assume that in VMs we don't need the quirk for old, broken ACPI. This solves problems with PCI BAR access and recent seabios versions on qemu. Agreement from @kettenis | |
|
5f0ad35b | 2025-04-01 08:35:31 | unbreak IPv6 neighbor discovery on ice(4) problem reported by bluhm@ | |
|
9abccaf1 | 2025-04-01 08:34:09 | restrict ice(4) firmware to features actually supported by the driver Avoids traffic stalls due to firmware trying to use multiple queues, which the driver does not handle yet. | |
|
2894c9c0 | 2025-04-01 08:32:16 | port ice(4) code for loading DDP firmware packages from freebsd Loading firmware is a prerequisite for performance features such as checksum offload and TSO. These features are not yet implemented by our version of this driver and will be added during the next release cycle. Firmware has been made available on firmware.openbsd.org by sthen@ | |
|
6cbbb384 | 2025-03-31 15:40:22 | pvclock: Use VM_PAGE_TO_PHYS as requested by kettenis@ | |
|
416e3503 | 2025-03-31 14:43:00 | pvclock: Map page unencrypted If SEV is enabled, we need to map the pvclock page as unencrypted / shared with the hypervisor. ok bluhm@ looks good to me hshoexer@ | |
|
909c4a09 | 2025-03-31 08:39:38 | remove outdated note to drop O* ioctl defines, they were removed in 2003 | |
|
a799c56d | 2025-03-31 05:27:49 | disable POOL_DEBUG for release ok deraadt@ | |
|
239cd4da | 2025-03-31 00:20:14 | drm/amdkfd: Fix user queue validation on Gfx7/8 From Philip Yang 731eccbd65aac026770e78279d551e1895572bd0 in linux-6.12.y/6.12.21 542c3bb836733a1325874310d54d25b4907ed10e in mainline linux | |
|
69dec186 | 2025-03-31 00:18:29 | drm/amdgpu: Fix JPEG video caps max size for navi1x and raven From David Rosca 3aa8e00fefcaadd33f40977225014b26ef3ae140 in linux-6.12.y/6.12.21 ec33964d9d88488fa954a03d476a8b811efc6e85 in mainline linux |