diff options
Diffstat (limited to 'sys/dev')
| -rw-r--r-- | sys/dev/mlx5/mlx5_en/en_hw_tls.h | 3 | ||||
| -rw-r--r-- | sys/dev/mlx5/mlx5_en/mlx5_en_hw_tls.c | 53 | ||||
| -rw-r--r-- | sys/dev/mlx5/mlx5_en/mlx5_en_main.c | 3 | ||||
| -rw-r--r-- | sys/dev/mmc/mmc_fdt_helpers.c | 11 | ||||
| -rw-r--r-- | sys/dev/mmc/mmcsd.c | 8 | ||||
| -rw-r--r-- | sys/dev/nvme/nvme_ns.c | 14 | ||||
| -rw-r--r-- | sys/dev/thunderbolt/tb_pcib.c | 14 | ||||
| -rw-r--r-- | sys/dev/virtio/virtqueue.c | 3 | ||||
| -rw-r--r-- | sys/dev/vmm/vmm_dev.c | 91 | ||||
| -rw-r--r-- | sys/dev/vmm/vmm_dev.h | 19 | ||||
| -rw-r--r-- | sys/dev/vmm/vmm_param.h | 33 |
11 files changed, 220 insertions, 32 deletions
diff --git a/sys/dev/mlx5/mlx5_en/en_hw_tls.h b/sys/dev/mlx5/mlx5_en/en_hw_tls.h index d637314e040e..cd57d2ac5f72 100644 --- a/sys/dev/mlx5/mlx5_en/en_hw_tls.h +++ b/sys/dev/mlx5/mlx5_en/en_hw_tls.h @@ -82,6 +82,8 @@ struct mlx5e_tls { struct sysctl_ctx_list ctx; struct mlx5e_tls_stats stats; struct workqueue_struct *wq; + struct workqueue_struct *prealloc_wq; + struct work_struct prealloc_work; uma_zone_t zone; uint32_t max_resources; /* max number of resources */ int zone_max; @@ -92,6 +94,7 @@ struct mlx5e_tls { int mlx5e_tls_init(struct mlx5e_priv *); void mlx5e_tls_cleanup(struct mlx5e_priv *); int mlx5e_sq_tls_xmit(struct mlx5e_sq *, struct mlx5e_xmit_args *, struct mbuf **); +void mlx5e_tls_prealloc_tags(struct mlx5e_priv *priv); if_snd_tag_alloc_t mlx5e_tls_snd_tag_alloc; diff --git a/sys/dev/mlx5/mlx5_en/mlx5_en_hw_tls.c b/sys/dev/mlx5/mlx5_en/mlx5_en_hw_tls.c index 6c83de5f3580..851316ccfcd7 100644 --- a/sys/dev/mlx5/mlx5_en/mlx5_en_hw_tls.c +++ b/sys/dev/mlx5/mlx5_en/mlx5_en_hw_tls.c @@ -80,23 +80,39 @@ static const char *mlx5e_tls_stats_desc[] = { }; static void mlx5e_tls_work(struct work_struct *); +static void mlx5e_tls_prealloc_work(struct work_struct *); /* - * Expand the tls tag UMA zone in a sleepable context + * Expand the tls tag UMA zone in an async context */ static void -mlx5e_prealloc_tags(struct mlx5e_priv *priv, int nitems) +mlx5e_tls_prealloc_work(struct work_struct *work) { + struct mlx5e_priv *priv; + struct mlx5e_tls *ptls; struct mlx5e_tls_tag **tags; - int i; + int i, nitems; + + ptls = container_of(work, struct mlx5e_tls, prealloc_work); + priv = container_of(ptls, struct mlx5e_priv, tls); + nitems = ptls->zone_max; tags = malloc(sizeof(tags[0]) * nitems, - M_MLX5E_TLS, M_WAITOK); - for (i = 0; i < nitems; i++) - tags[i] = uma_zalloc(priv->tls.zone, M_WAITOK); + M_MLX5E_TLS, M_WAITOK | M_ZERO); + for (i = 0; i < nitems; i++) { + tags[i] = uma_zalloc(priv->tls.zone, M_NOWAIT); + /* + * If the allocation fails, its likely we are competing + * with real consumers of tags and the zone is full, + * so exit the loop, and release the tags like we would + * if we allocated all "nitems" + */ + if (tags[i] == NULL) + break; + } __compiler_membar(); - for (i = 0; i < nitems; i++) + for (i = 0; i < nitems && tags[i] != NULL; i++) uma_zfree(priv->tls.zone, tags[i]); free(tags, M_MLX5E_TLS); } @@ -244,8 +260,6 @@ mlx5e_tls_init(struct mlx5e_priv *priv) } uma_zone_set_max(ptls->zone, ptls->zone_max); - if (prealloc_tags != 0) - mlx5e_prealloc_tags(priv, ptls->zone_max); for (x = 0; x != MLX5E_TLS_STATS_NUM; x++) ptls->stats.arg[x] = counter_u64_alloc(M_WAITOK); @@ -271,6 +285,23 @@ mlx5e_tls_init(struct mlx5e_priv *priv) } void +mlx5e_tls_prealloc_tags(struct mlx5e_priv *priv) +{ + struct mlx5e_tls *ptls = &priv->tls; + int prealloc_tags = 0; + + if (ptls->prealloc_wq != NULL) + return; + + TUNABLE_INT_FETCH("hw.mlx5.tls_prealloc_tags", &prealloc_tags); + if (prealloc_tags == 0) + return; + ptls->prealloc_wq = create_singlethread_workqueue("mlx5-tls-prealloc_wq"); + INIT_WORK(&ptls->prealloc_work, mlx5e_tls_prealloc_work); + queue_work(ptls->prealloc_wq, &ptls->prealloc_work); +} + +void mlx5e_tls_cleanup(struct mlx5e_priv *priv) { struct mlx5e_tls *ptls = &priv->tls; @@ -280,6 +311,10 @@ mlx5e_tls_cleanup(struct mlx5e_priv *priv) return; ptls->init = 0; + if (ptls->prealloc_wq != NULL) { + flush_workqueue(ptls->prealloc_wq); + destroy_workqueue(ptls->prealloc_wq); + } flush_workqueue(ptls->wq); sysctl_ctx_free(&ptls->ctx); uma_zdestroy(ptls->zone); diff --git a/sys/dev/mlx5/mlx5_en/mlx5_en_main.c b/sys/dev/mlx5/mlx5_en/mlx5_en_main.c index f83506bda1aa..ee9c53bb0a60 100644 --- a/sys/dev/mlx5/mlx5_en/mlx5_en_main.c +++ b/sys/dev/mlx5/mlx5_en/mlx5_en_main.c @@ -3335,6 +3335,9 @@ mlx5e_open_locked(if_t ifp) mlx5e_update_carrier(priv); + if ((if_getcapenable(ifp) & (IFCAP_TXTLS4 | IFCAP_TXTLS6)) != 0) + mlx5e_tls_prealloc_tags(priv); + return (0); err_close_channels: diff --git a/sys/dev/mmc/mmc_fdt_helpers.c b/sys/dev/mmc/mmc_fdt_helpers.c index aed85dab55f4..980785464a00 100644 --- a/sys/dev/mmc/mmc_fdt_helpers.c +++ b/sys/dev/mmc/mmc_fdt_helpers.c @@ -160,6 +160,17 @@ cd_setup(struct mmc_helper *helper, phandle_t node) } /* + * If the device has no card-detection, treat it as non-removable. + * This could be improved by polling for detection. + */ + if (helper->props & MMC_PROP_BROKEN_CD) { + helper->cd_disabled = true; + if (bootverbose) + device_printf(dev, "Broken card-detect\n"); + return; + } + + /* * If there is no cd-gpios property, then presumably the hardware * PRESENT_STATE register and interrupts will reflect card state * properly, and there's nothing more for us to do. Our get_present() diff --git a/sys/dev/mmc/mmcsd.c b/sys/dev/mmc/mmcsd.c index 5b9cb93c7b31..f2965048b285 100644 --- a/sys/dev/mmc/mmcsd.c +++ b/sys/dev/mmc/mmcsd.c @@ -1422,7 +1422,7 @@ mmcsd_task(void *arg) struct mmcsd_softc *sc; struct bio *bp; device_t dev, mmcbus; - int bio_error, err, sz; + int abio_error, err, sz; part = arg; sc = part->sc; @@ -1430,7 +1430,7 @@ mmcsd_task(void *arg) mmcbus = sc->mmcbus; while (1) { - bio_error = 0; + abio_error = 0; MMCSD_DISK_LOCK(part); do { if (part->running == 0) @@ -1475,11 +1475,11 @@ mmcsd_task(void *arg) } else if (bp->bio_cmd == BIO_DELETE) block = mmcsd_delete(part, bp); else - bio_error = EOPNOTSUPP; + abio_error = EOPNOTSUPP; release: MMCBUS_RELEASE_BUS(mmcbus, dev); if (block < end) { - bp->bio_error = (bio_error == 0) ? EIO : bio_error; + bp->bio_error = (abio_error == 0) ? EIO : abio_error; bp->bio_resid = (end - block) * sz; bp->bio_flags |= BIO_ERROR; } else diff --git a/sys/dev/nvme/nvme_ns.c b/sys/dev/nvme/nvme_ns.c index f4a588373c98..17684cc14ba2 100644 --- a/sys/dev/nvme/nvme_ns.c +++ b/sys/dev/nvme/nvme_ns.c @@ -45,7 +45,7 @@ #include "nvme_private.h" #include "nvme_linux.h" -static void nvme_bio_child_inbed(struct bio *parent, int bio_error); +static void nvme_bio_child_inbed(struct bio *parent, int abio_error); static void nvme_bio_child_done(void *arg, const struct nvme_completion *cpl); static uint32_t nvme_get_num_segments(uint64_t addr, uint64_t size, @@ -275,14 +275,14 @@ nvme_ns_bio_done(void *arg, const struct nvme_completion *status) } static void -nvme_bio_child_inbed(struct bio *parent, int bio_error) +nvme_bio_child_inbed(struct bio *parent, int abio_error) { struct nvme_completion parent_cpl; int children, inbed; - if (bio_error != 0) { + if (abio_error != 0) { parent->bio_flags |= BIO_ERROR; - parent->bio_error = bio_error; + parent->bio_error = abio_error; } /* @@ -309,12 +309,12 @@ nvme_bio_child_done(void *arg, const struct nvme_completion *cpl) { struct bio *child = arg; struct bio *parent; - int bio_error; + int abio_error; parent = child->bio_parent; g_destroy_bio(child); - bio_error = nvme_completion_is_error(cpl) ? EIO : 0; - nvme_bio_child_inbed(parent, bio_error); + abio_error = nvme_completion_is_error(cpl) ? EIO : 0; + nvme_bio_child_inbed(parent, abio_error); } static uint32_t diff --git a/sys/dev/thunderbolt/tb_pcib.c b/sys/dev/thunderbolt/tb_pcib.c index 00738984ad1c..bc4fc1ce00ec 100644 --- a/sys/dev/thunderbolt/tb_pcib.c +++ b/sys/dev/thunderbolt/tb_pcib.c @@ -557,8 +557,20 @@ static int tb_pci_probe(device_t dev) { struct tb_pcib_ident *n; + device_t parent; + devclass_t dc; - if ((n = tb_pcib_find_ident(device_get_parent(dev))) != NULL) { + /* + * This driver is only valid if the parent device is a PCI-PCI + * bridge. To determine that, check if the grandparent is a + * PCI bus. + */ + parent = device_get_parent(dev); + dc = device_get_devclass(device_get_parent(parent)); + if (strcmp(devclass_get_name(dc), "pci") != 0) + return (ENXIO); + + if ((n = tb_pcib_find_ident(parent)) != NULL) { switch (n->flags & TB_GEN_MASK) { case TB_GEN_TB1: device_set_desc(dev, "Thunderbolt 1 Link"); diff --git a/sys/dev/virtio/virtqueue.c b/sys/dev/virtio/virtqueue.c index cc7a233d60ee..41e01549c8b2 100644 --- a/sys/dev/virtio/virtqueue.c +++ b/sys/dev/virtio/virtqueue.c @@ -580,7 +580,8 @@ virtqueue_dequeue(struct virtqueue *vq, uint32_t *len) void *cookie; uint16_t used_idx, desc_idx; - if (vq->vq_used_cons_idx == vq_htog16(vq, vq->vq_ring.used->idx)) + if (vq->vq_used_cons_idx == + vq_htog16(vq, atomic_load_16(&vq->vq_ring.used->idx))) return (NULL); used_idx = vq->vq_used_cons_idx++ & (vq->vq_nentries - 1); diff --git a/sys/dev/vmm/vmm_dev.c b/sys/dev/vmm/vmm_dev.c index ebbceb25b69e..d6543bf6534e 100644 --- a/sys/dev/vmm/vmm_dev.c +++ b/sys/dev/vmm/vmm_dev.c @@ -14,9 +14,11 @@ #include <sys/kernel.h> #include <sys/malloc.h> #include <sys/mman.h> +#include <sys/module.h> #include <sys/priv.h> #include <sys/proc.h> #include <sys/queue.h> +#include <sys/smp.h> #include <sys/sx.h> #include <sys/sysctl.h> #include <sys/ucred.h> @@ -78,6 +80,8 @@ struct vmmdev_softc { int flags; }; +static bool vmm_initialized = false; + static SLIST_HEAD(, vmmdev_softc) head; static unsigned pr_allow_flag; @@ -88,6 +92,10 @@ static MALLOC_DEFINE(M_VMMDEV, "vmmdev", "vmmdev"); SYSCTL_DECL(_hw_vmm); +u_int vm_maxcpu; +SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, + &vm_maxcpu, 0, "Maximum number of vCPUs"); + static void devmem_destroy(void *arg); static int devmem_create_cdev(struct vmmdev_softc *sc, int id, char *devmem); @@ -619,20 +627,16 @@ vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag, } error = domainset_populate(&domain, mask, mseg->ds_policy, mseg->ds_mask_size); - if (error) { - free(mask, M_VMMDEV); + free(mask, M_VMMDEV); + if (error) break; - } domainset = domainset_create(&domain); if (domainset == NULL) { error = EINVAL; - free(mask, M_VMMDEV); break; } - free(mask, M_VMMDEV); } error = alloc_memseg(sc, mseg, sizeof(mseg->name), domainset); - break; } case VM_GET_MEMSEG: @@ -985,6 +989,9 @@ vmmdev_create(const char *name, struct ucred *cred) struct vm *vm; int error; + if (name == NULL || strlen(name) > VM_MAX_NAMELEN) + return (EINVAL); + sx_xlock(&vmmdev_mtx); sc = vmmdev_lookup(name, cred); if (sc != NULL) { @@ -1025,6 +1032,9 @@ sysctl_vmm_create(SYSCTL_HANDLER_ARGS) char *buf; int error, buflen; + if (!vmm_initialized) + return (ENXIO); + error = vmm_priv_check(req->td->td_ucred); if (error != 0) return (error); @@ -1110,7 +1120,7 @@ static struct cdevsw vmmctlsw = { .d_ioctl = vmmctl_ioctl, }; -int +static int vmmdev_init(void) { int error; @@ -1126,7 +1136,7 @@ vmmdev_init(void) return (error); } -int +static int vmmdev_cleanup(void) { sx_xlock(&vmmdev_mtx); @@ -1144,6 +1154,71 @@ vmmdev_cleanup(void) } static int +vmm_handler(module_t mod, int what, void *arg) +{ + int error; + + switch (what) { + case MOD_LOAD: + error = vmmdev_init(); + if (error != 0) + break; + + vm_maxcpu = mp_ncpus; + TUNABLE_INT_FETCH("hw.vmm.maxcpu", &vm_maxcpu); + if (vm_maxcpu > VM_MAXCPU) { + printf("vmm: vm_maxcpu clamped to %u\n", VM_MAXCPU); + vm_maxcpu = VM_MAXCPU; + } + if (vm_maxcpu == 0) + vm_maxcpu = 1; + + error = vmm_modinit(); + if (error == 0) + vmm_initialized = true; + else { + error = vmmdev_cleanup(); + KASSERT(error == 0, + ("%s: vmmdev_cleanup failed: %d", __func__, error)); + } + break; + case MOD_UNLOAD: + error = vmmdev_cleanup(); + if (error == 0 && vmm_initialized) { + error = vmm_modcleanup(); + if (error) { + /* + * Something bad happened - prevent new + * VMs from being created + */ + vmm_initialized = false; + } + } + break; + default: + error = 0; + break; + } + return (error); +} + +static moduledata_t vmm_kmod = { + "vmm", + vmm_handler, + NULL +}; + +/* + * vmm initialization has the following dependencies: + * + * - Initialization requires smp_rendezvous() and therefore must happen + * after SMP is fully functional (after SI_SUB_SMP). + * - vmm device initialization requires an initialized devfs. + */ +DECLARE_MODULE(vmm, vmm_kmod, MAX(SI_SUB_SMP, SI_SUB_DEVFS) + 1, SI_ORDER_ANY); +MODULE_VERSION(vmm, 1); + +static int devmem_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t len, struct vm_object **objp, int nprot) { diff --git a/sys/dev/vmm/vmm_dev.h b/sys/dev/vmm/vmm_dev.h index 2881a7063565..f14176c8afad 100644 --- a/sys/dev/vmm/vmm_dev.h +++ b/sys/dev/vmm/vmm_dev.h @@ -11,15 +11,19 @@ #include <sys/types.h> #include <sys/ioccom.h> + #include <machine/vmm_dev.h> +#include <dev/vmm/vmm_param.h> + #ifdef _KERNEL struct thread; struct vm; struct vcpu; -int vmmdev_init(void); -int vmmdev_cleanup(void); +int vmm_modinit(void); +int vmm_modcleanup(void); + int vmmdev_machdep_ioctl(struct vm *vm, struct vcpu *vcpu, u_long cmd, caddr_t data, int fflag, struct thread *td); @@ -53,6 +57,17 @@ struct vmmdev_ioctl { extern const struct vmmdev_ioctl vmmdev_machdep_ioctls[]; extern const size_t vmmdev_machdep_ioctl_count; +/* + * Upper limit on vm_maxcpu. Limited by use of uint16_t types for CPU counts as + * well as range of vpid values for VT-x on amd64 and by the capacity of + * cpuset_t masks. The call to new_unrhdr() in vpid_init() in vmx.c requires + * 'vm_maxcpu + 1 <= 0xffff', hence the '- 1' below. + */ +#define VM_MAXCPU MIN(0xffff - 1, CPU_SETSIZE) + +/* Maximum number of vCPUs in a single VM. */ +extern u_int vm_maxcpu; + #endif /* _KERNEL */ struct vmmctl_vm_create { diff --git a/sys/dev/vmm/vmm_param.h b/sys/dev/vmm/vmm_param.h new file mode 100644 index 000000000000..a5040eb0f58c --- /dev/null +++ b/sys/dev/vmm/vmm_param.h @@ -0,0 +1,33 @@ +/*- + * SPDX-License-Identifier: BSD-2-Clause + * + * Copyright (c) 2011 NetApp, Inc. + * All rights reserved. + */ + +#ifndef _DEV_VMM_PARAM_H_ +#define _DEV_VMM_PARAM_H_ + +/* + * The VM name has to fit into the pathname length constraints of devfs, + * governed primarily by SPECNAMELEN. The length is the total number of + * characters in the full path, relative to the mount point and not + * including any leading '/' characters. + * A prefix and a suffix are added to the name specified by the user. + * The prefix is usually "vmm/" or "vmm.io/", but can be a few characters + * longer for future use. + * The suffix is a string that identifies a bootrom image or some similar + * image that is attached to the VM. A separator character gets added to + * the suffix automatically when generating the full path, so it must be + * accounted for, reducing the effective length by 1. + * The effective length of a VM name is 229 bytes for FreeBSD 13 and 37 + * bytes for FreeBSD 12. A minimum length is set for safety and supports + * a SPECNAMELEN as small as 32 on old systems. + */ +#define VM_MAX_PREFIXLEN 10 +#define VM_MAX_SUFFIXLEN 15 +#define VM_MIN_NAMELEN 6 +#define VM_MAX_NAMELEN \ + (SPECNAMELEN - VM_MAX_PREFIXLEN - VM_MAX_SUFFIXLEN - 1) + +#endif /* !_DEV_VMM_PARAM_H_ */ |