wireguard-openbsd/sys/arch/arm64/dev/agtimer.c, branch jd/queueboosts WireGuard implementation for the OpenBSD kernel https://git.zx2c4.com/wireguard-openbsd/atom/sys/arch/arm64/dev/agtimer.c?h=jd%2Fqueueboosts 2020-06-05T22:14:25Z Allow userland access to the virtual counter. 2020-06-05T22:14:25Z kettenis kettenis@openbsd.org 2020-06-05T22:14:25Z urn:sha1:b0e77709c0dd5e58378837906c8281d687b7a4d8 ok patrick@, deraadt@ Add workaround for Cortex-A73 errata 858921. Pointed out by drahn@ who 2019-10-05T11:27:02Z kettenis kettenis@openbsd.org 2019-10-05T11:27:02Z urn:sha1:8493aee131ebd973330c4281be861eebb72c1001 also came up with the initial implementation. ok drahn@, jsg@ Use MAXCPUS as the number of elements for the array of per-cpu data. 2018-08-11T10:41:08Z kettenis kettenis@openbsd.org 2018-08-11T10:41:08Z urn:sha1:2b9b0c2aa9b7f81fa356431a7e0fcb178ee35198 Fixes machines with more than 8 cores. ok jsg@, patrick@ Add MULTIPROCESSOR support to the interrupt controller drivers. This makes 2018-01-31T10:52:12Z kettenis kettenis@openbsd.org 2018-01-31T10:52:12Z urn:sha1:d7e3db9cb74dfa9506722c07883ed0b6b05b346d the secondary CPUs receive clock interrupts. Based on diffs from drahn@. ok patrick@ Switch arm64 generic timer to use virtual timer instead of physical 2017-03-26T18:27:55Z drahn drahn@openbsd.org 2017-03-26T18:27:55Z urn:sha1:512509efbef8aa3c3a0f644c6fd338d15e79fa75 timer. virtual timer will always be present where physical timer may be disabled by hypervisor. Other OSes use virtual timer. ok patrick@ Make sure that the timer control registers are written before any 2017-02-18T00:47:18Z patrick patrick@openbsd.org 2017-02-18T00:47:18Z urn:sha1:53432af2b559cb01dba45633b65ca4d02810f654 further code is executed. Additionally, make sure the counter is read only after all previous code has executed. Bump variables related to arithmetic operations to 64-bit. Especially 2017-02-18T00:34:14Z patrick patrick@openbsd.org 2017-02-18T00:34:14Z urn:sha1:520d84a6e407b7ae378a54a83f634a2d808053ee the bump of usec and csec is of concern since a timer frequency of around 187MHz overflows in a 32-bit only calculation, as seen on the AMD Seattle SoC. Since we are running a 64-bit architecture, doing 64-bit arithmetic operations doesn't hurt us as much as on the 32-bit ARMv7 port. Initialize the generic timer early so that its delay function can be 2017-02-18T00:19:33Z patrick patrick@openbsd.org 2017-02-18T00:19:33Z urn:sha1:c3d704d48a347c9aee6fd3fa492ac325de1a5363 used early, similar to the armv7 implementation. The default frequency we chose for the generic timer does not always ring 2017-02-07T21:51:03Z patrick patrick@openbsd.org 2017-02-07T21:51:03Z urn:sha1:3fe8269783fbdf1117d4936b0b3e4178898d0709 true. Instead, unless overwritten by the device tree, we should ask the generic timer for its frequency. This fixes time on my AMD Seattle and should improve time management on QEMU as well. Also attach to "arm,armv7-timer". 2017-01-23T10:15:53Z kettenis kettenis@openbsd.org 2017-01-23T10:15:53Z urn:sha1:777fb14f988c0a88a4786c9ff984b431bea64d09 ok patrick@