diff options
Diffstat (limited to 'drivers/dma/idxd')
-rw-r--r-- | drivers/dma/idxd/cdev.c | 100 | ||||
-rw-r--r-- | drivers/dma/idxd/debugfs.c | 4 | ||||
-rw-r--r-- | drivers/dma/idxd/device.c | 8 | ||||
-rw-r--r-- | drivers/dma/idxd/idxd.h | 5 | ||||
-rw-r--r-- | drivers/dma/idxd/init.c | 6 | ||||
-rw-r--r-- | drivers/dma/idxd/irq.c | 4 | ||||
-rw-r--r-- | drivers/dma/idxd/perfmon.c | 9 | ||||
-rw-r--r-- | drivers/dma/idxd/registers.h | 3 | ||||
-rw-r--r-- | drivers/dma/idxd/sysfs.c | 27 |
9 files changed, 133 insertions, 33 deletions
diff --git a/drivers/dma/idxd/cdev.c b/drivers/dma/idxd/cdev.c index 8078ab9acfbc..57f1bf2ab20b 100644 --- a/drivers/dma/idxd/cdev.c +++ b/drivers/dma/idxd/cdev.c @@ -342,7 +342,7 @@ static void idxd_cdev_evl_drain_pasid(struct idxd_wq *wq, u32 pasid) if (!evl) return; - spin_lock(&evl->lock); + mutex_lock(&evl->lock); status.bits = ioread64(idxd->reg_base + IDXD_EVLSTATUS_OFFSET); t = status.tail; h = status.head; @@ -354,9 +354,8 @@ static void idxd_cdev_evl_drain_pasid(struct idxd_wq *wq, u32 pasid) set_bit(h, evl->bmap); h = (h + 1) % size; } - spin_unlock(&evl->lock); - drain_workqueue(wq->wq); + mutex_unlock(&evl->lock); } static int idxd_cdev_release(struct inode *node, struct file *filep) @@ -401,6 +400,18 @@ static int idxd_cdev_mmap(struct file *filp, struct vm_area_struct *vma) int rc; dev_dbg(&pdev->dev, "%s called\n", __func__); + + /* + * Due to an erratum in some of the devices supported by the driver, + * direct user submission to the device can be unsafe. + * (See the INTEL-SA-01084 security advisory) + * + * For the devices that exhibit this behavior, require that the user + * has CAP_SYS_RAWIO capabilities. + */ + if (!idxd->user_submission_safe && !capable(CAP_SYS_RAWIO)) + return -EPERM; + rc = check_vma(wq, vma, __func__); if (rc < 0) return rc; @@ -415,6 +426,70 @@ static int idxd_cdev_mmap(struct file *filp, struct vm_area_struct *vma) vma->vm_page_prot); } +static int idxd_submit_user_descriptor(struct idxd_user_context *ctx, + struct dsa_hw_desc __user *udesc) +{ + struct idxd_wq *wq = ctx->wq; + struct idxd_dev *idxd_dev = &wq->idxd->idxd_dev; + const uint64_t comp_addr_align = is_dsa_dev(idxd_dev) ? 0x20 : 0x40; + void __iomem *portal = idxd_wq_portal_addr(wq); + struct dsa_hw_desc descriptor __aligned(64); + int rc; + + rc = copy_from_user(&descriptor, udesc, sizeof(descriptor)); + if (rc) + return -EFAULT; + + /* + * DSA devices are capable of indirect ("batch") command submission. + * On devices where direct user submissions are not safe, we cannot + * allow this since there is no good way for us to verify these + * indirect commands. + */ + if (is_dsa_dev(idxd_dev) && descriptor.opcode == DSA_OPCODE_BATCH && + !wq->idxd->user_submission_safe) + return -EINVAL; + /* + * As per the programming specification, the completion address must be + * aligned to 32 or 64 bytes. If this is violated the hardware + * engine can get very confused (security issue). + */ + if (!IS_ALIGNED(descriptor.completion_addr, comp_addr_align)) + return -EINVAL; + + if (wq_dedicated(wq)) + iosubmit_cmds512(portal, &descriptor, 1); + else { + descriptor.priv = 0; + descriptor.pasid = ctx->pasid; + rc = idxd_enqcmds(wq, portal, &descriptor); + if (rc < 0) + return rc; + } + + return 0; +} + +static ssize_t idxd_cdev_write(struct file *filp, const char __user *buf, size_t len, + loff_t *unused) +{ + struct dsa_hw_desc __user *udesc = (struct dsa_hw_desc __user *)buf; + struct idxd_user_context *ctx = filp->private_data; + ssize_t written = 0; + int i; + + for (i = 0; i < len/sizeof(struct dsa_hw_desc); i++) { + int rc = idxd_submit_user_descriptor(ctx, udesc + i); + + if (rc) + return written ? written : rc; + + written += sizeof(struct dsa_hw_desc); + } + + return written; +} + static __poll_t idxd_cdev_poll(struct file *filp, struct poll_table_struct *wait) { @@ -437,6 +512,7 @@ static const struct file_operations idxd_cdev_fops = { .open = idxd_cdev_open, .release = idxd_cdev_release, .mmap = idxd_cdev_mmap, + .write = idxd_cdev_write, .poll = idxd_cdev_poll, }; @@ -501,7 +577,6 @@ void idxd_wq_del_cdev(struct idxd_wq *wq) struct idxd_cdev *idxd_cdev; idxd_cdev = wq->idxd_cdev; - ida_destroy(&file_ida); wq->idxd_cdev = NULL; cdev_device_del(&idxd_cdev->cdev, cdev_dev(idxd_cdev)); put_device(cdev_dev(idxd_cdev)); @@ -517,6 +592,14 @@ static int idxd_user_drv_probe(struct idxd_dev *idxd_dev) if (idxd->state != IDXD_DEV_ENABLED) return -ENXIO; + mutex_lock(&wq->wq_lock); + + if (!idxd_wq_driver_name_match(wq, dev)) { + idxd->cmd_status = IDXD_SCMD_WQ_NO_DRV_NAME; + rc = -ENODEV; + goto wq_err; + } + /* * User type WQ is enabled only when SVA is enabled for two reasons: * - If no IOMMU or IOMMU Passthrough without SVA, userspace @@ -532,14 +615,7 @@ static int idxd_user_drv_probe(struct idxd_dev *idxd_dev) dev_dbg(&idxd->pdev->dev, "User type WQ cannot be enabled without SVA.\n"); - return -EOPNOTSUPP; - } - - mutex_lock(&wq->wq_lock); - - if (!idxd_wq_driver_name_match(wq, dev)) { - idxd->cmd_status = IDXD_SCMD_WQ_NO_DRV_NAME; - rc = -ENODEV; + rc = -EOPNOTSUPP; goto wq_err; } diff --git a/drivers/dma/idxd/debugfs.c b/drivers/dma/idxd/debugfs.c index f3f25ee676f3..ad4245cb301d 100644 --- a/drivers/dma/idxd/debugfs.c +++ b/drivers/dma/idxd/debugfs.c @@ -66,7 +66,7 @@ static int debugfs_evl_show(struct seq_file *s, void *d) if (!evl || !evl->log) return 0; - spin_lock(&evl->lock); + mutex_lock(&evl->lock); evl_status.bits = ioread64(idxd->reg_base + IDXD_EVLSTATUS_OFFSET); t = evl_status.tail; @@ -87,7 +87,7 @@ static int debugfs_evl_show(struct seq_file *s, void *d) dump_event_entry(idxd, s, i, &count, processed); } - spin_unlock(&evl->lock); + mutex_unlock(&evl->lock); return 0; } diff --git a/drivers/dma/idxd/device.c b/drivers/dma/idxd/device.c index ecfdf4a8f1f8..c41ef195eeb9 100644 --- a/drivers/dma/idxd/device.c +++ b/drivers/dma/idxd/device.c @@ -775,7 +775,7 @@ static int idxd_device_evl_setup(struct idxd_device *idxd) goto err_alloc; } - spin_lock(&evl->lock); + mutex_lock(&evl->lock); evl->log = addr; evl->dma = dma_addr; evl->log_size = size; @@ -796,7 +796,7 @@ static int idxd_device_evl_setup(struct idxd_device *idxd) gencfg.evl_en = 1; iowrite32(gencfg.bits, idxd->reg_base + IDXD_GENCFG_OFFSET); - spin_unlock(&evl->lock); + mutex_unlock(&evl->lock); return 0; err_alloc: @@ -819,7 +819,7 @@ static void idxd_device_evl_free(struct idxd_device *idxd) if (!gencfg.evl_en) return; - spin_lock(&evl->lock); + mutex_lock(&evl->lock); gencfg.evl_en = 0; iowrite32(gencfg.bits, idxd->reg_base + IDXD_GENCFG_OFFSET); @@ -836,7 +836,7 @@ static void idxd_device_evl_free(struct idxd_device *idxd) evl_dma = evl->dma; evl->log = NULL; evl->size = IDXD_EVL_SIZE_MIN; - spin_unlock(&evl->lock); + mutex_unlock(&evl->lock); dma_free_coherent(dev, evl_log_size, evl_log, evl_dma); } diff --git a/drivers/dma/idxd/idxd.h b/drivers/dma/idxd/idxd.h index a4099a1e2340..868b724a3b75 100644 --- a/drivers/dma/idxd/idxd.h +++ b/drivers/dma/idxd/idxd.h @@ -288,12 +288,13 @@ struct idxd_driver_data { int evl_cr_off; int cr_status_off; int cr_result_off; + bool user_submission_safe; load_device_defaults_fn_t load_device_defaults; }; struct idxd_evl { /* Lock to protect event log access. */ - spinlock_t lock; + struct mutex lock; void *log; dma_addr_t dma; /* Total size of event log = number of entries * entry size. */ @@ -374,6 +375,8 @@ struct idxd_device { struct dentry *dbgfs_dir; struct dentry *dbgfs_evl_file; + + bool user_submission_safe; }; static inline unsigned int evl_ent_size(struct idxd_device *idxd) diff --git a/drivers/dma/idxd/init.c b/drivers/dma/idxd/init.c index 4954adc6bb60..a7295943fa22 100644 --- a/drivers/dma/idxd/init.c +++ b/drivers/dma/idxd/init.c @@ -47,6 +47,7 @@ static struct idxd_driver_data idxd_driver_data[] = { .align = 32, .dev_type = &dsa_device_type, .evl_cr_off = offsetof(struct dsa_evl_entry, cr), + .user_submission_safe = false, /* See INTEL-SA-01084 security advisory */ .cr_status_off = offsetof(struct dsa_completion_record, status), .cr_result_off = offsetof(struct dsa_completion_record, result), }, @@ -57,6 +58,7 @@ static struct idxd_driver_data idxd_driver_data[] = { .align = 64, .dev_type = &iax_device_type, .evl_cr_off = offsetof(struct iax_evl_entry, cr), + .user_submission_safe = false, /* See INTEL-SA-01084 security advisory */ .cr_status_off = offsetof(struct iax_completion_record, status), .cr_result_off = offsetof(struct iax_completion_record, error_code), .load_device_defaults = idxd_load_iaa_device_defaults, @@ -354,7 +356,7 @@ static int idxd_init_evl(struct idxd_device *idxd) if (!evl) return -ENOMEM; - spin_lock_init(&evl->lock); + mutex_init(&evl->lock); evl->size = IDXD_EVL_SIZE_MIN; idxd_name = dev_name(idxd_confdev(idxd)); @@ -774,6 +776,8 @@ static int idxd_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id) dev_info(&pdev->dev, "Intel(R) Accelerator Device (v%x)\n", idxd->hw.version); + idxd->user_submission_safe = data->user_submission_safe; + return 0; err_dev_register: diff --git a/drivers/dma/idxd/irq.c b/drivers/dma/idxd/irq.c index 348aa21389a9..8dc029c86551 100644 --- a/drivers/dma/idxd/irq.c +++ b/drivers/dma/idxd/irq.c @@ -363,7 +363,7 @@ static void process_evl_entries(struct idxd_device *idxd) evl_status.bits = 0; evl_status.int_pending = 1; - spin_lock(&evl->lock); + mutex_lock(&evl->lock); /* Clear interrupt pending bit */ iowrite32(evl_status.bits_upper32, idxd->reg_base + IDXD_EVLSTATUS_OFFSET + sizeof(u32)); @@ -380,7 +380,7 @@ static void process_evl_entries(struct idxd_device *idxd) evl_status.head = h; iowrite32(evl_status.bits_lower32, idxd->reg_base + IDXD_EVLSTATUS_OFFSET); - spin_unlock(&evl->lock); + mutex_unlock(&evl->lock); } irqreturn_t idxd_misc_thread(int vec, void *data) diff --git a/drivers/dma/idxd/perfmon.c b/drivers/dma/idxd/perfmon.c index fdda6d604262..5e94247e1ea7 100644 --- a/drivers/dma/idxd/perfmon.c +++ b/drivers/dma/idxd/perfmon.c @@ -528,14 +528,11 @@ static int perf_event_cpu_offline(unsigned int cpu, struct hlist_node *node) return 0; target = cpumask_any_but(cpu_online_mask, cpu); - /* migrate events if there is a valid target */ - if (target < nr_cpu_ids) + if (target < nr_cpu_ids) { cpumask_set_cpu(target, &perfmon_dsa_cpu_mask); - else - target = -1; - - perf_pmu_migrate_context(&idxd_pmu->pmu, cpu, target); + perf_pmu_migrate_context(&idxd_pmu->pmu, cpu, target); + } return 0; } diff --git a/drivers/dma/idxd/registers.h b/drivers/dma/idxd/registers.h index 315c004f58e4..e16dbf9ab324 100644 --- a/drivers/dma/idxd/registers.h +++ b/drivers/dma/idxd/registers.h @@ -6,9 +6,6 @@ #include <uapi/linux/idxd.h> /* PCI Config */ -#define PCI_DEVICE_ID_INTEL_DSA_SPR0 0x0b25 -#define PCI_DEVICE_ID_INTEL_IAX_SPR0 0x0cfe - #define DEVICE_VERSION_1 0x100 #define DEVICE_VERSION_2 0x200 diff --git a/drivers/dma/idxd/sysfs.c b/drivers/dma/idxd/sysfs.c index 7f28f01be672..f706eae0e76b 100644 --- a/drivers/dma/idxd/sysfs.c +++ b/drivers/dma/idxd/sysfs.c @@ -1197,12 +1197,35 @@ static ssize_t wq_enqcmds_retries_store(struct device *dev, struct device_attrib static struct device_attribute dev_attr_wq_enqcmds_retries = __ATTR(enqcmds_retries, 0644, wq_enqcmds_retries_show, wq_enqcmds_retries_store); +static ssize_t op_cap_show_common(struct device *dev, char *buf, unsigned long *opcap_bmap) +{ + ssize_t pos; + int i; + + pos = 0; + for (i = IDXD_MAX_OPCAP_BITS/64 - 1; i >= 0; i--) { + unsigned long val = opcap_bmap[i]; + + /* On systems where direct user submissions are not safe, we need to clear out + * the BATCH capability from the capability mask in sysfs since we cannot support + * that command on such systems. + */ + if (i == DSA_OPCODE_BATCH/64 && !confdev_to_idxd(dev)->user_submission_safe) + clear_bit(DSA_OPCODE_BATCH % 64, &val); + + pos += sysfs_emit_at(buf, pos, "%*pb", 64, &val); + pos += sysfs_emit_at(buf, pos, "%c", i == 0 ? '\n' : ','); + } + + return pos; +} + static ssize_t wq_op_config_show(struct device *dev, struct device_attribute *attr, char *buf) { struct idxd_wq *wq = confdev_to_wq(dev); - return sysfs_emit(buf, "%*pb\n", IDXD_MAX_OPCAP_BITS, wq->opcap_bmap); + return op_cap_show_common(dev, buf, wq->opcap_bmap); } static int idxd_verify_supported_opcap(struct idxd_device *idxd, unsigned long *opmask) @@ -1455,7 +1478,7 @@ static ssize_t op_cap_show(struct device *dev, { struct idxd_device *idxd = confdev_to_idxd(dev); - return sysfs_emit(buf, "%*pb\n", IDXD_MAX_OPCAP_BITS, idxd->opcap_bmap); + return op_cap_show_common(dev, buf, idxd->opcap_bmap); } static DEVICE_ATTR_RO(op_cap); |