Date: Thu, 23 May 2019 15:00:31 +0000 From: "aleksandr.fedorov_itglobal.com (Aleksandr Fedorov)" <phabric-noreply@FreeBSD.org> To: Phabricator <phabric-noreply@FreeBSD.org> Cc: freebsd-virtualization@freebsd.org Subject: [Differential] D20276: [bhyve][virtio-net] Allow guest VM's to set JUMBO MTU in case of using the VALE switch. Message-ID: <b949fb0b7c886d7b0693b92839d80858@localhost.localdomain> In-Reply-To: <differential-rev-PHID-DREV-pueoqfdkfh54jyuh6emz-req@reviews.freebsd.org> References: <differential-rev-PHID-DREV-pueoqfdkfh54jyuh6emz-req@reviews.freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
aleksandr.fedorov_itglobal.com added inline comments. INLINE COMMENTS > v.maffione_gmail.com wrote in pci_virtio_net.c:496 > In theory this NIOCRXSYNC is not needed, because poll() or kqueue_wait() calls NIOCRXSYNC internally. This works perfectly with poll(), at least. As far as I know bhyve uses kqueue to wait on the netmap file descriptor. What happens if you remove this ioctl()? I agree, it seems there are no need to call ioctl(..., NIOCRXSYNC, ...). And I observed a some throughput increase (~300-500 MBit/s 1500 MTU), when RX packet occupied one netmap buffer. This is explained by the fact that additional syscall was gone. But I need some time to investigate how this affects performance with a large MTU. CHANGES SINCE LAST ACTION https://reviews.freebsd.org/D20276/new/ REVISION DETAIL https://reviews.freebsd.org/D20276 EMAIL PREFERENCES https://reviews.freebsd.org/settings/panel/emailpreferences/ To: aleksandr.fedorov_itglobal.com, #bhyve, jhb, rgrimes, krion, v.maffione_gmail.com Cc: mizhka_gmail.com, novel, olevole_olevole.ru, freebsd-virtualization-list, evgueni.gavrilov_itglobal.com, bcran
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?b949fb0b7c886d7b0693b92839d80858>