Mercurial > illumos > git > illumos-joyent
changeset 25214:5b4f2c1d6e77
[illumos-gate merge]
commit eb8e4816e886312ff129c1a9213ad86d0fe5ffb6
12935 loader: setting vdev size based on label asize is not working
commit b38fc3ed80ecaf16c46433e690e6f17495c5e179
12922 bhyve vmm ops could have stronger guardrail
commit 88628b1bc8bd723915686a0f84bd0461ec80e590
12913 i40e want configurable num_rx_groups
| author | Jerry Jelinek <jerry.jelinek@joyent.com> |
|---|---|
| date | Tue, 14 Jul 2020 11:45:08 +0000 |
| parents | c854183fa07c (current diff) 721689034089 (diff) |
| children | 91529019146f |
| files | usr/src/boot/lib/libstand/zfs/zfsimpl.c usr/src/man/man7d/i40e.7d usr/src/uts/common/io/i40e/i40e_gld.c usr/src/uts/common/io/i40e/i40e_main.c usr/src/uts/common/io/i40e/i40e_sw.h usr/src/uts/i86pc/io/vmm/vmm.c |
| diffstat | 7 files changed, 148 insertions(+), 112 deletions(-) [+] |
line wrap: on
line diff
--- a/usr/src/boot/Makefile.version Mon Jul 13 14:13:11 2020 +0000 +++ b/usr/src/boot/Makefile.version Tue Jul 14 11:45:08 2020 +0000 @@ -33,4 +33,4 @@ # Use date like formatting here, YYYY.MM.DD.XX, without leading zeroes. # The version is processed from left to right, the version number can only # be increased. -BOOT_VERSION = $(LOADER_VERSION)-2020.06.26.1 +BOOT_VERSION = $(LOADER_VERSION)-2020.07.07.1
--- a/usr/src/boot/lib/libstand/zfs/zfsimpl.c Mon Jul 13 14:13:11 2020 +0000 +++ b/usr/src/boot/lib/libstand/zfs/zfsimpl.c Tue Jul 14 11:45:08 2020 +0000 @@ -1835,6 +1835,55 @@ return (vdev_read_phys(vd, &bp, buf, off, size)); } +static uint64_t +vdev_get_label_asize(unsigned char *nvl) +{ + unsigned char *vdevs; + uint64_t asize; + const char *type; + int len; + + asize = 0; + /* Get vdev tree */ + if (nvlist_find(nvl, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST, + NULL, &vdevs, NULL) != 0) + return (asize); + + /* + * Get vdev type. We will calculate asize for raidz, mirror and disk. + * For raidz, the asize is raw size of all children. + */ + if (nvlist_find(vdevs, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, + NULL, &type, &len) != 0) + goto done; + + if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 && + memcmp(type, VDEV_TYPE_DISK, len) != 0 && + memcmp(type, VDEV_TYPE_RAIDZ, len) != 0) + goto done; + + if (nvlist_find(vdevs, ZPOOL_CONFIG_ASIZE, DATA_TYPE_UINT64, + NULL, &asize, NULL) != 0) + goto done; + + if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) { + unsigned char *kids; + int nkids; + + if (nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, + DATA_TYPE_NVLIST_ARRAY, &nkids, &kids, NULL) != 0) { + asize = 0; + goto done; + } + + asize /= nkids; + } + + asize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; +done: + return (asize); +} + static unsigned char * vdev_label_read_config(vdev_t *vd, uint64_t txg) { @@ -1882,10 +1931,9 @@ * Use asize from pool config. We need this * because we can get bad value from BIOS. */ - if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE, - DATA_TYPE_UINT64, NULL, &asize, NULL) == 0) { - vd->v_psize = asize + - VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; + asize = vdev_get_label_asize(nvl); + if (asize != 0) { + vd->v_psize = asize; } } }
--- a/usr/src/man/man7d/i40e.7d Mon Jul 13 14:13:11 2020 +0000 +++ b/usr/src/man/man7d/i40e.7d Tue Jul 14 11:45:08 2020 +0000 @@ -10,8 +10,9 @@ .\" .\" .\" Copyright (c) 2018 Joyent, Inc. +.\" Copyright 2020 Ryan Zezeski .\" -.Dd April 15, 2020 +.Dd July 1, 2020 .Dt I40E 7D .Os .Sh NAME @@ -157,6 +158,25 @@ The default is always to enable them. It is not recommended to to disable them. .Ed +.It Sy rx_num_groups +.Bd -filled -compact +Minimum: +.Sy 1 | +Maximum: +.Sy 32 +.Ed +.Bd -filled +The +.Sy rx_num_groups +property determines the number of receive mac groups provided by the driver. +Each group can handle all unicast traffic for a single MAC address, more groups +means more unicast traffic that can be steered by hardware. +However, more groups also means more demand for kernel memory. +If you are not making heavy use of VNICs, or do not need the efficiency gains +of hardware steering, then reducing this number can reduce kernel memory +taken by +.Nm i40e. +.Ed .It Sy rx_ring_size .Bd -filled -compact Minimum:
--- a/usr/src/uts/common/io/i40e/i40e_gld.c Mon Jul 13 14:13:11 2020 +0000 +++ b/usr/src/uts/common/io/i40e/i40e_gld.c Tue Jul 14 11:45:08 2020 +0000 @@ -13,6 +13,7 @@ * Copyright 2015 OmniTI Computer Consulting, Inc. All rights reserved. * Copyright (c) 2018, Joyent, Inc. * Copyright 2017 Tegile Systems, Inc. All rights reserved. + * Copyright 2020 Ryan Zezeski */ /* @@ -565,7 +566,7 @@ infop->mgi_addmac = i40e_group_add_mac; infop->mgi_remmac = i40e_group_remove_mac; - ASSERT(i40e->i40e_num_rx_groups <= I40E_GROUP_MAX); + ASSERT3U(i40e->i40e_num_rx_groups, <=, I40E_MAX_NUM_RX_GROUPS); infop->mgi_count = i40e->i40e_num_trqpairs_per_vsi; }
--- a/usr/src/uts/common/io/i40e/i40e_main.c Mon Jul 13 14:13:11 2020 +0000 +++ b/usr/src/uts/common/io/i40e/i40e_main.c Tue Jul 14 11:45:08 2020 +0000 @@ -14,6 +14,7 @@ * Copyright 2019 Joyent, Inc. * Copyright 2017 Tegile Systems, Inc. All rights reserved. * Copyright 2020 RackTop Systems, Inc. + * Copyright 2020 Ryan Zezeski */ /* @@ -1220,7 +1221,7 @@ } if (i40e->i40e_num_rx_groups == 0) { - i40e->i40e_num_rx_groups = I40E_GROUP_MAX; + i40e->i40e_num_rx_groups = I40E_DEF_NUM_RX_GROUPS; } } @@ -1589,6 +1590,10 @@ i40e->i40e_tx_ring_size - I40E_TX_MAX_COOKIE, I40E_DEF_TX_BLOCK_THRESH); + i40e->i40e_num_rx_groups = i40e_get_prop(i40e, "rx_num_groups", + I40E_MIN_NUM_RX_GROUPS, I40E_MAX_NUM_RX_GROUPS, + I40E_DEF_NUM_RX_GROUPS); + i40e->i40e_rx_ring_size = i40e_get_prop(i40e, "rx_ring_size", I40E_MIN_RX_RING_SIZE, I40E_MAX_RX_RING_SIZE, I40E_DEF_RX_RING_SIZE); @@ -1776,7 +1781,6 @@ } i40e->i40e_intr_type = 0; - i40e->i40e_num_rx_groups = I40E_GROUP_MAX; /* * We need to determine the number of queue pairs per traffic
--- a/usr/src/uts/common/io/i40e/i40e_sw.h Mon Jul 13 14:13:11 2020 +0000 +++ b/usr/src/uts/common/io/i40e/i40e_sw.h Tue Jul 14 11:45:08 2020 +0000 @@ -13,6 +13,7 @@ * Copyright 2015 OmniTI Computer Consulting, Inc. All rights reserved. * Copyright 2019 Joyent, Inc. * Copyright 2017 Tegile Systems, Inc. All rights reserved. + * Copyright 2020 Ryan Zezeski */ /* @@ -89,6 +90,16 @@ #define I40E_MAX_TX_RING_SIZE 4096 #define I40E_DEF_TX_RING_SIZE 1024 +/* + * Place an artificial limit on the max number of groups. The X710 + * series supports up to 384 VSIs to be partitioned across PFs as the + * driver sees fit. But until we support more interrupts this seems + * like a good place to start. + */ +#define I40E_MIN_NUM_RX_GROUPS 1 +#define I40E_MAX_NUM_RX_GROUPS 32 +#define I40E_DEF_NUM_RX_GROUPS 16 + #define I40E_MIN_RX_RING_SIZE 64 #define I40E_MAX_RX_RING_SIZE 4096 #define I40E_DEF_RX_RING_SIZE 1024 @@ -269,14 +280,6 @@ */ #define I40E_DDI_PROP_LEN 64 -/* - * Place an artificial limit on the max number of groups. The X710 - * series supports up to 384 VSIs to be partitioned across PFs as the - * driver sees fit. But until we support more interrupts this seems - * like a good place to start. - */ -#define I40E_GROUP_MAX 32 - #define I40E_GROUP_NOMSIX 1 #define I40E_TRQPAIR_NOMSIX 1 @@ -834,7 +837,7 @@ /* * Device state, switch information, and resources. */ - i40e_vsi_t i40e_vsis[I40E_GROUP_MAX]; + i40e_vsi_t i40e_vsis[I40E_MAX_NUM_RX_GROUPS]; uint16_t i40e_mac_seid; /* SEID of physical MAC */ uint16_t i40e_veb_seid; /* switch atop MAC (SEID) */ uint16_t i40e_vsi_avail; /* VSIs avail to this PF */
--- a/usr/src/uts/i86pc/io/vmm/vmm.c Mon Jul 13 14:13:11 2020 +0000 +++ b/usr/src/uts/i86pc/io/vmm/vmm.c Tue Jul 14 11:45:08 2020 +0000 @@ -39,6 +39,7 @@ * * Copyright 2015 Pluribus Networks Inc. * Copyright 2018 Joyent, Inc. + * Copyright 2020 Oxide Computer Company */ #include <sys/cdefs.h> @@ -203,35 +204,56 @@ static int vmm_initialized; -static struct vmm_ops *ops; -#define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) -#define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) -#define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) - -#define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) + +static void +nullop_panic(void) +{ + panic("null vmm operation call"); +} + +/* Do not allow use of an un-set `ops` to do anything but panic */ +static struct vmm_ops vmm_ops_null = { + .init = (vmm_init_func_t)nullop_panic, + .cleanup = (vmm_cleanup_func_t)nullop_panic, + .resume = (vmm_resume_func_t)nullop_panic, + .vminit = (vmi_init_func_t)nullop_panic, + .vmrun = (vmi_run_func_t)nullop_panic, + .vmcleanup = (vmi_cleanup_func_t)nullop_panic, + .vmgetreg = (vmi_get_register_t)nullop_panic, + .vmsetreg = (vmi_set_register_t)nullop_panic, + .vmgetdesc = (vmi_get_desc_t)nullop_panic, + .vmsetdesc = (vmi_set_desc_t)nullop_panic, + .vmgetcap = (vmi_get_cap_t)nullop_panic, + .vmsetcap = (vmi_set_cap_t)nullop_panic, + .vmspace_alloc = (vmi_vmspace_alloc)nullop_panic, + .vmspace_free = (vmi_vmspace_free)nullop_panic, + .vlapic_init = (vmi_vlapic_init)nullop_panic, + .vlapic_cleanup = (vmi_vlapic_cleanup)nullop_panic, + .vmsavectx = (vmi_savectx)nullop_panic, + .vmrestorectx = (vmi_restorectx)nullop_panic, +}; + +static struct vmm_ops *ops = &vmm_ops_null; + +#define VMM_INIT(num) ((*ops->init)(num)) +#define VMM_CLEANUP() ((*ops->cleanup)()) +#define VMM_RESUME() ((*ops->resume)()) + +#define VMINIT(vm, pmap) ((*ops->vminit)(vm, pmap)) #define VMRUN(vmi, vcpu, rip, pmap, evinfo) \ - (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, evinfo) : ENXIO) -#define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) -#define VMSPACE_ALLOC(min, max) \ - (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) -#define VMSPACE_FREE(vmspace) \ - (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) -#define VMGETREG(vmi, vcpu, num, retval) \ - (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) -#define VMSETREG(vmi, vcpu, num, val) \ - (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) -#define VMGETDESC(vmi, vcpu, num, desc) \ - (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) -#define VMSETDESC(vmi, vcpu, num, desc) \ - (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) -#define VMGETCAP(vmi, vcpu, num, retval) \ - (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) -#define VMSETCAP(vmi, vcpu, num, val) \ - (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) -#define VLAPIC_INIT(vmi, vcpu) \ - (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) -#define VLAPIC_CLEANUP(vmi, vlapic) \ - (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) + ((*ops->vmrun)(vmi, vcpu, rip, pmap, evinfo) ) +#define VMCLEANUP(vmi) ((*ops->vmcleanup)(vmi) ) +#define VMSPACE_ALLOC(min, max) ((*ops->vmspace_alloc)(min, max)) +#define VMSPACE_FREE(vmspace) ((*ops->vmspace_free)(vmspace)) + +#define VMGETREG(vmi, vcpu, num, rv) ((*ops->vmgetreg)(vmi, vcpu, num, rv)) +#define VMSETREG(vmi, vcpu, num, val) ((*ops->vmsetreg)(vmi, vcpu, num, val)) +#define VMGETDESC(vmi, vcpu, num, dsc) ((*ops->vmgetdesc)(vmi, vcpu, num, dsc)) +#define VMSETDESC(vmi, vcpu, num, dsc) ((*ops->vmsetdesc)(vmi, vcpu, num, dsc)) +#define VMGETCAP(vmi, vcpu, num, rv) ((*ops->vmgetcap)(vmi, vcpu, num, rv)) +#define VMSETCAP(vmi, vcpu, num, val) ((*ops->vmsetcap)(vmi, vcpu, num, val)) +#define VLAPIC_INIT(vmi, vcpu) ((*ops->vlapic_init)(vmi, vcpu)) +#define VLAPIC_CLEANUP(vmi, vlapic) ((*ops->vlapic_cleanup)(vmi, vlapic)) #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) #define fpu_stop_emulating() clts() @@ -380,14 +402,6 @@ return (&vcpu->exitinfo); } -#ifdef __FreeBSD__ -static void -vmm_resume(void) -{ - VMM_RESUME(); -} -#endif - static int vmm_init(void) { @@ -423,67 +437,13 @@ return (VMM_INIT(vmm_ipinum)); } -#ifdef __FreeBSD__ - -static int -vmm_handler(module_t mod, int what, void *arg) -{ - int error; - - switch (what) { - case MOD_LOAD: - vmmdev_init(); - error = vmm_init(); - if (error == 0) - vmm_initialized = 1; - break; - case MOD_UNLOAD: - error = vmmdev_cleanup(); - if (error == 0) { - vmm_resume_p = NULL; - iommu_cleanup(); -#ifdef __FreeBSD__ - if (vmm_ipinum != IPI_AST) - lapic_ipi_free(vmm_ipinum); -#endif - error = VMM_CLEANUP(); - /* - * Something bad happened - prevent new - * VMs from being created - */ - if (error) - vmm_initialized = 0; - } - break; - default: - error = 0; - break; - } - return (error); -} - -static moduledata_t vmm_kmod = { - "vmm", - vmm_handler, - NULL -}; - -/* - * vmm initialization has the following dependencies: - * - * - VT-x initialization requires smp_rendezvous() and therefore must happen - * after SMP is fully functional (after SI_SUB_SMP). - */ -DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); -MODULE_VERSION(vmm, 1); - -#else /* __FreeBSD__ */ - int vmm_mod_load() { int error; + VERIFY(vmm_initialized == 0); + error = vmm_init(); if (error == 0) vmm_initialized = 1; @@ -496,6 +456,8 @@ { int error; + VERIFY(vmm_initialized == 1); + iommu_cleanup(); error = VMM_CLEANUP(); if (error) @@ -505,8 +467,6 @@ return (0); } -#endif /* __FreeBSD__ */ - static void vm_init(struct vm *vm, bool create) {