1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright 2020-2021 Amazon.com, Inc. or its affiliates. All Rights Reserved.
7 * DOC: Enclave lifetime management driver for Nitro Enclaves (NE).
8 * Nitro is a hypervisor that has been developed by Amazon.
11 #include <linux/anon_inodes.h>
12 #include <linux/capability.h>
13 #include <linux/cpu.h>
14 #include <linux/device.h>
15 #include <linux/file.h>
16 #include <linux/hugetlb.h>
17 #include <linux/limits.h>
18 #include <linux/list.h>
19 #include <linux/miscdevice.h>
21 #include <linux/mman.h>
22 #include <linux/module.h>
23 #include <linux/mutex.h>
24 #include <linux/nitro_enclaves.h>
25 #include <linux/pci.h>
26 #include <linux/poll.h>
27 #include <linux/range.h>
28 #include <linux/slab.h>
29 #include <linux/types.h>
30 #include <uapi/linux/vm_sockets.h>
32 #include "ne_misc_dev.h"
33 #include "ne_pci_dev.h"
36 * NE_CPUS_SIZE - Size for max 128 CPUs, for now, in a cpu-list string, comma
37 * separated. The NE CPU pool includes CPUs from a single NUMA
40 #define NE_CPUS_SIZE (512)
43 * NE_EIF_LOAD_OFFSET - The offset where to copy the Enclave Image Format (EIF)
44 * image in enclave memory.
46 #define NE_EIF_LOAD_OFFSET (8 * 1024UL * 1024UL)
49 * NE_MIN_ENCLAVE_MEM_SIZE - The minimum memory size an enclave can be launched
52 #define NE_MIN_ENCLAVE_MEM_SIZE (64 * 1024UL * 1024UL)
55 * NE_MIN_MEM_REGION_SIZE - The minimum size of an enclave memory region.
57 #define NE_MIN_MEM_REGION_SIZE (2 * 1024UL * 1024UL)
60 * NE_PARENT_VM_CID - The CID for the vsock device of the primary / parent VM.
62 #define NE_PARENT_VM_CID (3)
64 static long ne_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
66 static const struct file_operations ne_fops = {
68 .llseek = noop_llseek,
69 .unlocked_ioctl = ne_ioctl,
72 static struct miscdevice ne_misc_dev = {
73 .minor = MISC_DYNAMIC_MINOR,
74 .name = "nitro_enclaves",
79 struct ne_devs ne_devs = {
80 .ne_misc_dev = &ne_misc_dev,
84 * TODO: Update logic to create new sysfs entries instead of using
85 * a kernel parameter e.g. if multiple sysfs files needed.
87 static int ne_set_kernel_param(const char *val, const struct kernel_param *kp);
89 static const struct kernel_param_ops ne_cpu_pool_ops = {
90 .get = param_get_string,
91 .set = ne_set_kernel_param,
94 static char ne_cpus[NE_CPUS_SIZE];
95 static struct kparam_string ne_cpus_arg = {
96 .maxlen = sizeof(ne_cpus),
100 module_param_cb(ne_cpus, &ne_cpu_pool_ops, &ne_cpus_arg, 0644);
101 /* https://www.kernel.org/doc/html/latest/admin-guide/kernel-parameters.html#cpu-lists */
102 MODULE_PARM_DESC(ne_cpus, "<cpu-list> - CPU pool used for Nitro Enclaves");
105 * struct ne_cpu_pool - CPU pool used for Nitro Enclaves.
106 * @avail_threads_per_core: Available full CPU cores to be dedicated to
107 * enclave(s). The cpumasks from the array, indexed
108 * by core id, contain all the threads from the
109 * available cores, that are not set for created
110 * enclave(s). The full CPU cores are part of the
112 * @mutex: Mutex for the access to the NE CPU pool.
113 * @nr_parent_vm_cores : The size of the available threads per core array.
114 * The total number of CPU cores available on the
115 * primary / parent VM.
116 * @nr_threads_per_core: The number of threads that a full CPU core has.
117 * @numa_node: NUMA node of the CPUs in the pool.
120 cpumask_var_t *avail_threads_per_core;
122 unsigned int nr_parent_vm_cores;
123 unsigned int nr_threads_per_core;
127 static struct ne_cpu_pool ne_cpu_pool;
130 * struct ne_phys_contig_mem_regions - Contiguous physical memory regions.
131 * @num: The number of regions that currently has.
132 * @regions: The array of physical memory regions.
134 struct ne_phys_contig_mem_regions {
136 struct range *regions;
140 * ne_check_enclaves_created() - Verify if at least one enclave has been created.
141 * @void: No parameters provided.
143 * Context: Process context.
145 * * True if at least one enclave is created.
148 static bool ne_check_enclaves_created(void)
150 struct ne_pci_dev *ne_pci_dev = ne_devs.ne_pci_dev;
156 mutex_lock(&ne_pci_dev->enclaves_list_mutex);
158 if (!list_empty(&ne_pci_dev->enclaves_list))
161 mutex_unlock(&ne_pci_dev->enclaves_list_mutex);
167 * ne_setup_cpu_pool() - Set the NE CPU pool after handling sanity checks such
168 * as not sharing CPU cores with the primary / parent VM
169 * or not using CPU 0, which should remain available for
170 * the primary / parent VM. Offline the CPUs from the
171 * pool after the checks passed.
172 * @ne_cpu_list: The CPU list used for setting NE CPU pool.
174 * Context: Process context.
177 * * Negative return value on failure.
179 static int ne_setup_cpu_pool(const char *ne_cpu_list)
182 unsigned int cpu = 0;
183 cpumask_var_t cpu_pool;
184 unsigned int cpu_sibling = 0;
189 if (!zalloc_cpumask_var(&cpu_pool, GFP_KERNEL))
192 mutex_lock(&ne_cpu_pool.mutex);
194 rc = cpulist_parse(ne_cpu_list, cpu_pool);
196 pr_err("%s: Error in cpulist parse [rc=%d]\n", ne_misc_dev.name, rc);
198 goto free_pool_cpumask;
201 cpu = cpumask_any(cpu_pool);
202 if (cpu >= nr_cpu_ids) {
203 pr_err("%s: No CPUs available in CPU pool\n", ne_misc_dev.name);
207 goto free_pool_cpumask;
211 * Check if the CPUs are online, to further get info about them
212 * e.g. numa node, core id, siblings.
214 for_each_cpu(cpu, cpu_pool)
215 if (cpu_is_offline(cpu)) {
216 pr_err("%s: CPU %d is offline, has to be online to get its metadata\n",
217 ne_misc_dev.name, cpu);
221 goto free_pool_cpumask;
225 * Check if the CPUs from the NE CPU pool are from the same NUMA node.
227 for_each_cpu(cpu, cpu_pool)
229 numa_node = cpu_to_node(cpu);
231 pr_err("%s: Invalid NUMA node %d\n",
232 ne_misc_dev.name, numa_node);
236 goto free_pool_cpumask;
239 if (numa_node != cpu_to_node(cpu)) {
240 pr_err("%s: CPUs with different NUMA nodes\n",
245 goto free_pool_cpumask;
250 * Check if CPU 0 and its siblings are included in the provided CPU pool
251 * They should remain available for the primary / parent VM.
253 if (cpumask_test_cpu(0, cpu_pool)) {
254 pr_err("%s: CPU 0 has to remain available\n", ne_misc_dev.name);
258 goto free_pool_cpumask;
261 for_each_cpu(cpu_sibling, topology_sibling_cpumask(0)) {
262 if (cpumask_test_cpu(cpu_sibling, cpu_pool)) {
263 pr_err("%s: CPU sibling %d for CPU 0 is in CPU pool\n",
264 ne_misc_dev.name, cpu_sibling);
268 goto free_pool_cpumask;
273 * Check if CPU siblings are included in the provided CPU pool. The
274 * expectation is that full CPU cores are made available in the CPU pool
277 for_each_cpu(cpu, cpu_pool) {
278 for_each_cpu(cpu_sibling, topology_sibling_cpumask(cpu)) {
279 if (!cpumask_test_cpu(cpu_sibling, cpu_pool)) {
280 pr_err("%s: CPU %d is not in CPU pool\n",
281 ne_misc_dev.name, cpu_sibling);
285 goto free_pool_cpumask;
290 /* Calculate the number of threads from a full CPU core. */
291 cpu = cpumask_any(cpu_pool);
292 for_each_cpu(cpu_sibling, topology_sibling_cpumask(cpu))
293 ne_cpu_pool.nr_threads_per_core++;
295 ne_cpu_pool.nr_parent_vm_cores = nr_cpu_ids / ne_cpu_pool.nr_threads_per_core;
297 ne_cpu_pool.avail_threads_per_core = kcalloc(ne_cpu_pool.nr_parent_vm_cores,
298 sizeof(*ne_cpu_pool.avail_threads_per_core),
300 if (!ne_cpu_pool.avail_threads_per_core) {
303 goto free_pool_cpumask;
306 for (i = 0; i < ne_cpu_pool.nr_parent_vm_cores; i++)
307 if (!zalloc_cpumask_var(&ne_cpu_pool.avail_threads_per_core[i], GFP_KERNEL)) {
310 goto free_cores_cpumask;
314 * Split the NE CPU pool in threads per core to keep the CPU topology
315 * after offlining the CPUs.
317 for_each_cpu(cpu, cpu_pool) {
318 core_id = topology_core_id(cpu);
319 if (core_id < 0 || core_id >= ne_cpu_pool.nr_parent_vm_cores) {
320 pr_err("%s: Invalid core id %d for CPU %d\n",
321 ne_misc_dev.name, core_id, cpu);
328 cpumask_set_cpu(cpu, ne_cpu_pool.avail_threads_per_core[core_id]);
332 * CPUs that are given to enclave(s) should not be considered online
333 * by Linux anymore, as the hypervisor will degrade them to floating.
334 * The physical CPUs (full cores) are carved out of the primary / parent
335 * VM and given to the enclave VM. The same number of vCPUs would run
336 * on less pCPUs for the primary / parent VM.
338 * We offline them here, to not degrade performance and expose correct
339 * topology to Linux and user space.
341 for_each_cpu(cpu, cpu_pool) {
342 rc = remove_cpu(cpu);
344 pr_err("%s: CPU %d is not offlined [rc=%d]\n",
345 ne_misc_dev.name, cpu, rc);
351 free_cpumask_var(cpu_pool);
353 ne_cpu_pool.numa_node = numa_node;
355 mutex_unlock(&ne_cpu_pool.mutex);
360 for_each_cpu(cpu, cpu_pool)
363 for (i = 0; i < ne_cpu_pool.nr_parent_vm_cores; i++)
364 cpumask_clear(ne_cpu_pool.avail_threads_per_core[i]);
366 for (i = 0; i < ne_cpu_pool.nr_parent_vm_cores; i++)
367 free_cpumask_var(ne_cpu_pool.avail_threads_per_core[i]);
368 kfree(ne_cpu_pool.avail_threads_per_core);
370 free_cpumask_var(cpu_pool);
371 ne_cpu_pool.nr_parent_vm_cores = 0;
372 ne_cpu_pool.nr_threads_per_core = 0;
373 ne_cpu_pool.numa_node = -1;
374 mutex_unlock(&ne_cpu_pool.mutex);
380 * ne_teardown_cpu_pool() - Online the CPUs from the NE CPU pool and cleanup the
382 * @void: No parameters provided.
384 * Context: Process context.
386 static void ne_teardown_cpu_pool(void)
388 unsigned int cpu = 0;
392 mutex_lock(&ne_cpu_pool.mutex);
394 if (!ne_cpu_pool.nr_parent_vm_cores) {
395 mutex_unlock(&ne_cpu_pool.mutex);
400 for (i = 0; i < ne_cpu_pool.nr_parent_vm_cores; i++) {
401 for_each_cpu(cpu, ne_cpu_pool.avail_threads_per_core[i]) {
404 pr_err("%s: CPU %d is not onlined [rc=%d]\n",
405 ne_misc_dev.name, cpu, rc);
408 cpumask_clear(ne_cpu_pool.avail_threads_per_core[i]);
410 free_cpumask_var(ne_cpu_pool.avail_threads_per_core[i]);
413 kfree(ne_cpu_pool.avail_threads_per_core);
414 ne_cpu_pool.nr_parent_vm_cores = 0;
415 ne_cpu_pool.nr_threads_per_core = 0;
416 ne_cpu_pool.numa_node = -1;
418 mutex_unlock(&ne_cpu_pool.mutex);
422 * ne_set_kernel_param() - Set the NE CPU pool value via the NE kernel parameter.
423 * @val: NE CPU pool string value.
424 * @kp : NE kernel parameter associated with the NE CPU pool.
426 * Context: Process context.
429 * * Negative return value on failure.
431 static int ne_set_kernel_param(const char *val, const struct kernel_param *kp)
433 char error_val[] = "";
436 if (!capable(CAP_SYS_ADMIN))
439 if (ne_check_enclaves_created()) {
440 pr_err("%s: The CPU pool is used by enclave(s)\n", ne_misc_dev.name);
445 ne_teardown_cpu_pool();
447 rc = ne_setup_cpu_pool(val);
449 pr_err("%s: Error in setup CPU pool [rc=%d]\n", ne_misc_dev.name, rc);
451 param_set_copystring(error_val, kp);
456 rc = param_set_copystring(val, kp);
458 pr_err("%s: Error in param set copystring [rc=%d]\n", ne_misc_dev.name, rc);
460 ne_teardown_cpu_pool();
462 param_set_copystring(error_val, kp);
471 * ne_donated_cpu() - Check if the provided CPU is already used by the enclave.
472 * @ne_enclave : Private data associated with the current enclave.
473 * @cpu: CPU to check if already used.
475 * Context: Process context. This function is called with the ne_enclave mutex held.
477 * * True if the provided CPU is already used by the enclave.
480 static bool ne_donated_cpu(struct ne_enclave *ne_enclave, unsigned int cpu)
482 if (cpumask_test_cpu(cpu, ne_enclave->vcpu_ids))
489 * ne_get_unused_core_from_cpu_pool() - Get the id of a full core from the
491 * @void: No parameters provided.
493 * Context: Process context. This function is called with the ne_enclave and
494 * ne_cpu_pool mutexes held.
497 * * -1 if no CPU core available in the pool.
499 static int ne_get_unused_core_from_cpu_pool(void)
504 for (i = 0; i < ne_cpu_pool.nr_parent_vm_cores; i++)
505 if (!cpumask_empty(ne_cpu_pool.avail_threads_per_core[i])) {
515 * ne_set_enclave_threads_per_core() - Set the threads of the provided core in
516 * the enclave data structure.
517 * @ne_enclave : Private data associated with the current enclave.
518 * @core_id: Core id to get its threads from the NE CPU pool.
519 * @vcpu_id: vCPU id part of the provided core.
521 * Context: Process context. This function is called with the ne_enclave and
522 * ne_cpu_pool mutexes held.
525 * * Negative return value on failure.
527 static int ne_set_enclave_threads_per_core(struct ne_enclave *ne_enclave,
528 int core_id, u32 vcpu_id)
530 unsigned int cpu = 0;
532 if (core_id < 0 && vcpu_id == 0) {
533 dev_err_ratelimited(ne_misc_dev.this_device,
534 "No CPUs available in NE CPU pool\n");
536 return -NE_ERR_NO_CPUS_AVAIL_IN_POOL;
540 dev_err_ratelimited(ne_misc_dev.this_device,
541 "CPU %d is not in NE CPU pool\n", vcpu_id);
543 return -NE_ERR_VCPU_NOT_IN_CPU_POOL;
546 if (core_id >= ne_enclave->nr_parent_vm_cores) {
547 dev_err_ratelimited(ne_misc_dev.this_device,
548 "Invalid core id %d - ne_enclave\n", core_id);
550 return -NE_ERR_VCPU_INVALID_CPU_CORE;
553 for_each_cpu(cpu, ne_cpu_pool.avail_threads_per_core[core_id])
554 cpumask_set_cpu(cpu, ne_enclave->threads_per_core[core_id]);
556 cpumask_clear(ne_cpu_pool.avail_threads_per_core[core_id]);
562 * ne_get_cpu_from_cpu_pool() - Get a CPU from the NE CPU pool, either from the
563 * remaining sibling(s) of a CPU core or the first
564 * sibling of a new CPU core.
565 * @ne_enclave : Private data associated with the current enclave.
566 * @vcpu_id: vCPU to get from the NE CPU pool.
568 * Context: Process context. This function is called with the ne_enclave mutex held.
571 * * Negative return value on failure.
573 static int ne_get_cpu_from_cpu_pool(struct ne_enclave *ne_enclave, u32 *vcpu_id)
576 unsigned int cpu = 0;
581 * If previously allocated a thread of a core to this enclave, first
582 * check remaining sibling(s) for new CPU allocations, so that full
583 * CPU cores are used for the enclave.
585 for (i = 0; i < ne_enclave->nr_parent_vm_cores; i++)
586 for_each_cpu(cpu, ne_enclave->threads_per_core[i])
587 if (!ne_donated_cpu(ne_enclave, cpu)) {
593 mutex_lock(&ne_cpu_pool.mutex);
596 * If no remaining siblings, get a core from the NE CPU pool and keep
597 * track of all the threads in the enclave threads per core data structure.
599 core_id = ne_get_unused_core_from_cpu_pool();
601 rc = ne_set_enclave_threads_per_core(ne_enclave, core_id, *vcpu_id);
605 *vcpu_id = cpumask_any(ne_enclave->threads_per_core[core_id]);
610 mutex_unlock(&ne_cpu_pool.mutex);
616 * ne_get_vcpu_core_from_cpu_pool() - Get from the NE CPU pool the id of the
617 * core associated with the provided vCPU.
618 * @vcpu_id: Provided vCPU id to get its associated core id.
620 * Context: Process context. This function is called with the ne_enclave and
621 * ne_cpu_pool mutexes held.
624 * * -1 if the provided vCPU is not in the pool.
626 static int ne_get_vcpu_core_from_cpu_pool(u32 vcpu_id)
631 for (i = 0; i < ne_cpu_pool.nr_parent_vm_cores; i++)
632 if (cpumask_test_cpu(vcpu_id, ne_cpu_pool.avail_threads_per_core[i])) {
642 * ne_check_cpu_in_cpu_pool() - Check if the given vCPU is in the available CPUs
644 * @ne_enclave : Private data associated with the current enclave.
645 * @vcpu_id: ID of the vCPU to check if available in the NE CPU pool.
647 * Context: Process context. This function is called with the ne_enclave mutex held.
650 * * Negative return value on failure.
652 static int ne_check_cpu_in_cpu_pool(struct ne_enclave *ne_enclave, u32 vcpu_id)
658 if (ne_donated_cpu(ne_enclave, vcpu_id)) {
659 dev_err_ratelimited(ne_misc_dev.this_device,
660 "CPU %d already used\n", vcpu_id);
662 return -NE_ERR_VCPU_ALREADY_USED;
666 * If previously allocated a thread of a core to this enclave, but not
667 * the full core, first check remaining sibling(s).
669 for (i = 0; i < ne_enclave->nr_parent_vm_cores; i++)
670 if (cpumask_test_cpu(vcpu_id, ne_enclave->threads_per_core[i]))
673 mutex_lock(&ne_cpu_pool.mutex);
676 * If no remaining siblings, get from the NE CPU pool the core
677 * associated with the vCPU and keep track of all the threads in the
678 * enclave threads per core data structure.
680 core_id = ne_get_vcpu_core_from_cpu_pool(vcpu_id);
682 rc = ne_set_enclave_threads_per_core(ne_enclave, core_id, vcpu_id);
689 mutex_unlock(&ne_cpu_pool.mutex);
695 * ne_add_vcpu_ioctl() - Add a vCPU to the slot associated with the current
697 * @ne_enclave : Private data associated with the current enclave.
698 * @vcpu_id: ID of the CPU to be associated with the given slot,
701 * Context: Process context. This function is called with the ne_enclave mutex held.
704 * * Negative return value on failure.
706 static int ne_add_vcpu_ioctl(struct ne_enclave *ne_enclave, u32 vcpu_id)
708 struct ne_pci_dev_cmd_reply cmd_reply = {};
709 struct pci_dev *pdev = ne_devs.ne_pci_dev->pdev;
711 struct slot_add_vcpu_req slot_add_vcpu_req = {};
713 if (ne_enclave->mm != current->mm)
716 slot_add_vcpu_req.slot_uid = ne_enclave->slot_uid;
717 slot_add_vcpu_req.vcpu_id = vcpu_id;
719 rc = ne_do_request(pdev, SLOT_ADD_VCPU,
720 &slot_add_vcpu_req, sizeof(slot_add_vcpu_req),
721 &cmd_reply, sizeof(cmd_reply));
723 dev_err_ratelimited(ne_misc_dev.this_device,
724 "Error in slot add vCPU [rc=%d]\n", rc);
729 cpumask_set_cpu(vcpu_id, ne_enclave->vcpu_ids);
731 ne_enclave->nr_vcpus++;
737 * ne_sanity_check_user_mem_region() - Sanity check the user space memory
738 * region received during the set user
739 * memory region ioctl call.
740 * @ne_enclave : Private data associated with the current enclave.
741 * @mem_region : User space memory region to be sanity checked.
743 * Context: Process context. This function is called with the ne_enclave mutex held.
746 * * Negative return value on failure.
748 static int ne_sanity_check_user_mem_region(struct ne_enclave *ne_enclave,
749 struct ne_user_memory_region mem_region)
751 struct ne_mem_region *ne_mem_region = NULL;
753 if (ne_enclave->mm != current->mm)
756 if (mem_region.memory_size & (NE_MIN_MEM_REGION_SIZE - 1)) {
757 dev_err_ratelimited(ne_misc_dev.this_device,
758 "User space memory size is not multiple of 2 MiB\n");
760 return -NE_ERR_INVALID_MEM_REGION_SIZE;
763 if (!IS_ALIGNED(mem_region.userspace_addr, NE_MIN_MEM_REGION_SIZE)) {
764 dev_err_ratelimited(ne_misc_dev.this_device,
765 "User space address is not 2 MiB aligned\n");
767 return -NE_ERR_UNALIGNED_MEM_REGION_ADDR;
770 if ((mem_region.userspace_addr & (NE_MIN_MEM_REGION_SIZE - 1)) ||
771 !access_ok((void __user *)(unsigned long)mem_region.userspace_addr,
772 mem_region.memory_size)) {
773 dev_err_ratelimited(ne_misc_dev.this_device,
774 "Invalid user space address range\n");
776 return -NE_ERR_INVALID_MEM_REGION_ADDR;
779 list_for_each_entry(ne_mem_region, &ne_enclave->mem_regions_list,
780 mem_region_list_entry) {
781 u64 memory_size = ne_mem_region->memory_size;
782 u64 userspace_addr = ne_mem_region->userspace_addr;
784 if ((userspace_addr <= mem_region.userspace_addr &&
785 mem_region.userspace_addr < (userspace_addr + memory_size)) ||
786 (mem_region.userspace_addr <= userspace_addr &&
787 (mem_region.userspace_addr + mem_region.memory_size) > userspace_addr)) {
788 dev_err_ratelimited(ne_misc_dev.this_device,
789 "User space memory region already used\n");
791 return -NE_ERR_MEM_REGION_ALREADY_USED;
799 * ne_sanity_check_user_mem_region_page() - Sanity check a page from the user space
800 * memory region received during the set
801 * user memory region ioctl call.
802 * @ne_enclave : Private data associated with the current enclave.
803 * @mem_region_page: Page from the user space memory region to be sanity checked.
805 * Context: Process context. This function is called with the ne_enclave mutex held.
808 * * Negative return value on failure.
810 static int ne_sanity_check_user_mem_region_page(struct ne_enclave *ne_enclave,
811 struct page *mem_region_page)
813 if (!PageHuge(mem_region_page)) {
814 dev_err_ratelimited(ne_misc_dev.this_device,
815 "Not a hugetlbfs page\n");
817 return -NE_ERR_MEM_NOT_HUGE_PAGE;
820 if (page_size(mem_region_page) & (NE_MIN_MEM_REGION_SIZE - 1)) {
821 dev_err_ratelimited(ne_misc_dev.this_device,
822 "Page size not multiple of 2 MiB\n");
824 return -NE_ERR_INVALID_PAGE_SIZE;
827 if (ne_enclave->numa_node != page_to_nid(mem_region_page)) {
828 dev_err_ratelimited(ne_misc_dev.this_device,
829 "Page is not from NUMA node %d\n",
830 ne_enclave->numa_node);
832 return -NE_ERR_MEM_DIFFERENT_NUMA_NODE;
839 * ne_sanity_check_phys_mem_region() - Sanity check the start address and the size
840 * of a physical memory region.
841 * @phys_mem_region_paddr : Physical start address of the region to be sanity checked.
842 * @phys_mem_region_size : Length of the region to be sanity checked.
844 * Context: Process context. This function is called with the ne_enclave mutex held.
847 * * Negative return value on failure.
849 static int ne_sanity_check_phys_mem_region(u64 phys_mem_region_paddr,
850 u64 phys_mem_region_size)
852 if (phys_mem_region_size & (NE_MIN_MEM_REGION_SIZE - 1)) {
853 dev_err_ratelimited(ne_misc_dev.this_device,
854 "Physical mem region size is not multiple of 2 MiB\n");
859 if (!IS_ALIGNED(phys_mem_region_paddr, NE_MIN_MEM_REGION_SIZE)) {
860 dev_err_ratelimited(ne_misc_dev.this_device,
861 "Physical mem region address is not 2 MiB aligned\n");
870 * ne_merge_phys_contig_memory_regions() - Add a memory region and merge the adjacent
871 * regions if they are physically contiguous.
872 * @phys_contig_regions : Private data associated with the contiguous physical memory regions.
873 * @page_paddr : Physical start address of the region to be added.
874 * @page_size : Length of the region to be added.
876 * Context: Process context. This function is called with the ne_enclave mutex held.
879 * * Negative return value on failure.
882 ne_merge_phys_contig_memory_regions(struct ne_phys_contig_mem_regions *phys_contig_regions,
883 u64 page_paddr, u64 page_size)
885 unsigned long num = phys_contig_regions->num;
888 rc = ne_sanity_check_phys_mem_region(page_paddr, page_size);
892 /* Physically contiguous, just merge */
893 if (num && (phys_contig_regions->regions[num - 1].end + 1) == page_paddr) {
894 phys_contig_regions->regions[num - 1].end += page_size;
896 phys_contig_regions->regions[num].start = page_paddr;
897 phys_contig_regions->regions[num].end = page_paddr + page_size - 1;
898 phys_contig_regions->num++;
905 * ne_set_user_memory_region_ioctl() - Add user space memory region to the slot
906 * associated with the current enclave.
907 * @ne_enclave : Private data associated with the current enclave.
908 * @mem_region : User space memory region to be associated with the given slot.
910 * Context: Process context. This function is called with the ne_enclave mutex held.
913 * * Negative return value on failure.
915 static int ne_set_user_memory_region_ioctl(struct ne_enclave *ne_enclave,
916 struct ne_user_memory_region mem_region)
920 unsigned long max_nr_pages = 0;
921 unsigned long memory_size = 0;
922 struct ne_mem_region *ne_mem_region = NULL;
923 struct pci_dev *pdev = ne_devs.ne_pci_dev->pdev;
924 struct ne_phys_contig_mem_regions phys_contig_mem_regions = {};
927 rc = ne_sanity_check_user_mem_region(ne_enclave, mem_region);
931 ne_mem_region = kzalloc(sizeof(*ne_mem_region), GFP_KERNEL);
935 max_nr_pages = mem_region.memory_size / NE_MIN_MEM_REGION_SIZE;
937 ne_mem_region->pages = kcalloc(max_nr_pages, sizeof(*ne_mem_region->pages),
939 if (!ne_mem_region->pages) {
942 goto free_mem_region;
945 phys_contig_mem_regions.regions = kcalloc(max_nr_pages,
946 sizeof(*phys_contig_mem_regions.regions),
948 if (!phys_contig_mem_regions.regions) {
951 goto free_mem_region;
955 i = ne_mem_region->nr_pages;
957 if (i == max_nr_pages) {
958 dev_err_ratelimited(ne_misc_dev.this_device,
959 "Reached max nr of pages in the pages data struct\n");
966 gup_rc = get_user_pages_unlocked(mem_region.userspace_addr + memory_size, 1,
967 ne_mem_region->pages + i, FOLL_GET);
972 dev_err_ratelimited(ne_misc_dev.this_device,
973 "Error in get user pages [rc=%d]\n", rc);
978 rc = ne_sanity_check_user_mem_region_page(ne_enclave, ne_mem_region->pages[i]);
982 rc = ne_merge_phys_contig_memory_regions(&phys_contig_mem_regions,
983 page_to_phys(ne_mem_region->pages[i]),
984 page_size(ne_mem_region->pages[i]));
988 memory_size += page_size(ne_mem_region->pages[i]);
990 ne_mem_region->nr_pages++;
991 } while (memory_size < mem_region.memory_size);
993 if ((ne_enclave->nr_mem_regions + phys_contig_mem_regions.num) >
994 ne_enclave->max_mem_regions) {
995 dev_err_ratelimited(ne_misc_dev.this_device,
996 "Reached max memory regions %lld\n",
997 ne_enclave->max_mem_regions);
999 rc = -NE_ERR_MEM_MAX_REGIONS;
1004 for (i = 0; i < phys_contig_mem_regions.num; i++) {
1005 u64 phys_region_addr = phys_contig_mem_regions.regions[i].start;
1006 u64 phys_region_size = range_len(&phys_contig_mem_regions.regions[i]);
1008 rc = ne_sanity_check_phys_mem_region(phys_region_addr, phys_region_size);
1013 ne_mem_region->memory_size = mem_region.memory_size;
1014 ne_mem_region->userspace_addr = mem_region.userspace_addr;
1016 list_add(&ne_mem_region->mem_region_list_entry, &ne_enclave->mem_regions_list);
1018 for (i = 0; i < phys_contig_mem_regions.num; i++) {
1019 struct ne_pci_dev_cmd_reply cmd_reply = {};
1020 struct slot_add_mem_req slot_add_mem_req = {};
1022 slot_add_mem_req.slot_uid = ne_enclave->slot_uid;
1023 slot_add_mem_req.paddr = phys_contig_mem_regions.regions[i].start;
1024 slot_add_mem_req.size = range_len(&phys_contig_mem_regions.regions[i]);
1026 rc = ne_do_request(pdev, SLOT_ADD_MEM,
1027 &slot_add_mem_req, sizeof(slot_add_mem_req),
1028 &cmd_reply, sizeof(cmd_reply));
1030 dev_err_ratelimited(ne_misc_dev.this_device,
1031 "Error in slot add mem [rc=%d]\n", rc);
1033 kfree(phys_contig_mem_regions.regions);
1036 * Exit here without put pages as memory regions may
1037 * already been added.
1042 ne_enclave->mem_size += slot_add_mem_req.size;
1043 ne_enclave->nr_mem_regions++;
1046 kfree(phys_contig_mem_regions.regions);
1051 for (i = 0; i < ne_mem_region->nr_pages; i++)
1052 put_page(ne_mem_region->pages[i]);
1054 kfree(phys_contig_mem_regions.regions);
1055 kfree(ne_mem_region->pages);
1056 kfree(ne_mem_region);
1062 * ne_start_enclave_ioctl() - Trigger enclave start after the enclave resources,
1063 * such as memory and CPU, have been set.
1064 * @ne_enclave : Private data associated with the current enclave.
1065 * @enclave_start_info : Enclave info that includes enclave cid and flags.
1067 * Context: Process context. This function is called with the ne_enclave mutex held.
1070 * * Negative return value on failure.
1072 static int ne_start_enclave_ioctl(struct ne_enclave *ne_enclave,
1073 struct ne_enclave_start_info *enclave_start_info)
1075 struct ne_pci_dev_cmd_reply cmd_reply = {};
1076 unsigned int cpu = 0;
1077 struct enclave_start_req enclave_start_req = {};
1079 struct pci_dev *pdev = ne_devs.ne_pci_dev->pdev;
1082 if (!ne_enclave->nr_mem_regions) {
1083 dev_err_ratelimited(ne_misc_dev.this_device,
1084 "Enclave has no mem regions\n");
1086 return -NE_ERR_NO_MEM_REGIONS_ADDED;
1089 if (ne_enclave->mem_size < NE_MIN_ENCLAVE_MEM_SIZE) {
1090 dev_err_ratelimited(ne_misc_dev.this_device,
1091 "Enclave memory is less than %ld\n",
1092 NE_MIN_ENCLAVE_MEM_SIZE);
1094 return -NE_ERR_ENCLAVE_MEM_MIN_SIZE;
1097 if (!ne_enclave->nr_vcpus) {
1098 dev_err_ratelimited(ne_misc_dev.this_device,
1099 "Enclave has no vCPUs\n");
1101 return -NE_ERR_NO_VCPUS_ADDED;
1104 for (i = 0; i < ne_enclave->nr_parent_vm_cores; i++)
1105 for_each_cpu(cpu, ne_enclave->threads_per_core[i])
1106 if (!cpumask_test_cpu(cpu, ne_enclave->vcpu_ids)) {
1107 dev_err_ratelimited(ne_misc_dev.this_device,
1108 "Full CPU cores not used\n");
1110 return -NE_ERR_FULL_CORES_NOT_USED;
1113 enclave_start_req.enclave_cid = enclave_start_info->enclave_cid;
1114 enclave_start_req.flags = enclave_start_info->flags;
1115 enclave_start_req.slot_uid = ne_enclave->slot_uid;
1117 rc = ne_do_request(pdev, ENCLAVE_START,
1118 &enclave_start_req, sizeof(enclave_start_req),
1119 &cmd_reply, sizeof(cmd_reply));
1121 dev_err_ratelimited(ne_misc_dev.this_device,
1122 "Error in enclave start [rc=%d]\n", rc);
1127 ne_enclave->state = NE_STATE_RUNNING;
1129 enclave_start_info->enclave_cid = cmd_reply.enclave_cid;
1135 * ne_enclave_ioctl() - Ioctl function provided by the enclave file.
1136 * @file: File associated with this ioctl function.
1137 * @cmd: The command that is set for the ioctl call.
1138 * @arg: The argument that is provided for the ioctl call.
1140 * Context: Process context.
1143 * * Negative return value on failure.
1145 static long ne_enclave_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1147 struct ne_enclave *ne_enclave = file->private_data;
1154 if (copy_from_user(&vcpu_id, (void __user *)arg, sizeof(vcpu_id)))
1157 mutex_lock(&ne_enclave->enclave_info_mutex);
1159 if (ne_enclave->state != NE_STATE_INIT) {
1160 dev_err_ratelimited(ne_misc_dev.this_device,
1161 "Enclave is not in init state\n");
1163 mutex_unlock(&ne_enclave->enclave_info_mutex);
1165 return -NE_ERR_NOT_IN_INIT_STATE;
1168 if (vcpu_id >= (ne_enclave->nr_parent_vm_cores *
1169 ne_enclave->nr_threads_per_core)) {
1170 dev_err_ratelimited(ne_misc_dev.this_device,
1171 "vCPU id higher than max CPU id\n");
1173 mutex_unlock(&ne_enclave->enclave_info_mutex);
1175 return -NE_ERR_INVALID_VCPU;
1179 /* Use the CPU pool for choosing a CPU for the enclave. */
1180 rc = ne_get_cpu_from_cpu_pool(ne_enclave, &vcpu_id);
1182 dev_err_ratelimited(ne_misc_dev.this_device,
1183 "Error in get CPU from pool [rc=%d]\n",
1186 mutex_unlock(&ne_enclave->enclave_info_mutex);
1191 /* Check if the provided vCPU is available in the NE CPU pool. */
1192 rc = ne_check_cpu_in_cpu_pool(ne_enclave, vcpu_id);
1194 dev_err_ratelimited(ne_misc_dev.this_device,
1195 "Error in check CPU %d in pool [rc=%d]\n",
1198 mutex_unlock(&ne_enclave->enclave_info_mutex);
1204 rc = ne_add_vcpu_ioctl(ne_enclave, vcpu_id);
1206 mutex_unlock(&ne_enclave->enclave_info_mutex);
1211 mutex_unlock(&ne_enclave->enclave_info_mutex);
1213 if (copy_to_user((void __user *)arg, &vcpu_id, sizeof(vcpu_id)))
1219 case NE_GET_IMAGE_LOAD_INFO: {
1220 struct ne_image_load_info image_load_info = {};
1222 if (copy_from_user(&image_load_info, (void __user *)arg, sizeof(image_load_info)))
1225 mutex_lock(&ne_enclave->enclave_info_mutex);
1227 if (ne_enclave->state != NE_STATE_INIT) {
1228 dev_err_ratelimited(ne_misc_dev.this_device,
1229 "Enclave is not in init state\n");
1231 mutex_unlock(&ne_enclave->enclave_info_mutex);
1233 return -NE_ERR_NOT_IN_INIT_STATE;
1236 mutex_unlock(&ne_enclave->enclave_info_mutex);
1238 if (!image_load_info.flags ||
1239 image_load_info.flags >= NE_IMAGE_LOAD_MAX_FLAG_VAL) {
1240 dev_err_ratelimited(ne_misc_dev.this_device,
1241 "Incorrect flag in enclave image load info\n");
1243 return -NE_ERR_INVALID_FLAG_VALUE;
1246 if (image_load_info.flags == NE_EIF_IMAGE)
1247 image_load_info.memory_offset = NE_EIF_LOAD_OFFSET;
1249 if (copy_to_user((void __user *)arg, &image_load_info, sizeof(image_load_info)))
1255 case NE_SET_USER_MEMORY_REGION: {
1256 struct ne_user_memory_region mem_region = {};
1259 if (copy_from_user(&mem_region, (void __user *)arg, sizeof(mem_region)))
1262 if (mem_region.flags >= NE_MEMORY_REGION_MAX_FLAG_VAL) {
1263 dev_err_ratelimited(ne_misc_dev.this_device,
1264 "Incorrect flag for user memory region\n");
1266 return -NE_ERR_INVALID_FLAG_VALUE;
1269 mutex_lock(&ne_enclave->enclave_info_mutex);
1271 if (ne_enclave->state != NE_STATE_INIT) {
1272 dev_err_ratelimited(ne_misc_dev.this_device,
1273 "Enclave is not in init state\n");
1275 mutex_unlock(&ne_enclave->enclave_info_mutex);
1277 return -NE_ERR_NOT_IN_INIT_STATE;
1280 rc = ne_set_user_memory_region_ioctl(ne_enclave, mem_region);
1282 mutex_unlock(&ne_enclave->enclave_info_mutex);
1287 mutex_unlock(&ne_enclave->enclave_info_mutex);
1292 case NE_START_ENCLAVE: {
1293 struct ne_enclave_start_info enclave_start_info = {};
1296 if (copy_from_user(&enclave_start_info, (void __user *)arg,
1297 sizeof(enclave_start_info)))
1300 if (enclave_start_info.flags >= NE_ENCLAVE_START_MAX_FLAG_VAL) {
1301 dev_err_ratelimited(ne_misc_dev.this_device,
1302 "Incorrect flag in enclave start info\n");
1304 return -NE_ERR_INVALID_FLAG_VALUE;
1308 * Do not use well-known CIDs - 0, 1, 2 - for enclaves.
1309 * VMADDR_CID_ANY = -1U
1310 * VMADDR_CID_HYPERVISOR = 0
1311 * VMADDR_CID_LOCAL = 1
1312 * VMADDR_CID_HOST = 2
1313 * Note: 0 is used as a placeholder to auto-generate an enclave CID.
1314 * http://man7.org/linux/man-pages/man7/vsock.7.html
1316 if (enclave_start_info.enclave_cid > 0 &&
1317 enclave_start_info.enclave_cid <= VMADDR_CID_HOST) {
1318 dev_err_ratelimited(ne_misc_dev.this_device,
1319 "Well-known CID value, not to be used for enclaves\n");
1321 return -NE_ERR_INVALID_ENCLAVE_CID;
1324 if (enclave_start_info.enclave_cid == U32_MAX) {
1325 dev_err_ratelimited(ne_misc_dev.this_device,
1326 "Well-known CID value, not to be used for enclaves\n");
1328 return -NE_ERR_INVALID_ENCLAVE_CID;
1332 * Do not use the CID of the primary / parent VM for enclaves.
1334 if (enclave_start_info.enclave_cid == NE_PARENT_VM_CID) {
1335 dev_err_ratelimited(ne_misc_dev.this_device,
1336 "CID of the parent VM, not to be used for enclaves\n");
1338 return -NE_ERR_INVALID_ENCLAVE_CID;
1341 /* 64-bit CIDs are not yet supported for the vsock device. */
1342 if (enclave_start_info.enclave_cid > U32_MAX) {
1343 dev_err_ratelimited(ne_misc_dev.this_device,
1344 "64-bit CIDs not yet supported for the vsock device\n");
1346 return -NE_ERR_INVALID_ENCLAVE_CID;
1349 mutex_lock(&ne_enclave->enclave_info_mutex);
1351 if (ne_enclave->state != NE_STATE_INIT) {
1352 dev_err_ratelimited(ne_misc_dev.this_device,
1353 "Enclave is not in init state\n");
1355 mutex_unlock(&ne_enclave->enclave_info_mutex);
1357 return -NE_ERR_NOT_IN_INIT_STATE;
1360 rc = ne_start_enclave_ioctl(ne_enclave, &enclave_start_info);
1362 mutex_unlock(&ne_enclave->enclave_info_mutex);
1367 mutex_unlock(&ne_enclave->enclave_info_mutex);
1369 if (copy_to_user((void __user *)arg, &enclave_start_info,
1370 sizeof(enclave_start_info)))
1384 * ne_enclave_remove_all_mem_region_entries() - Remove all memory region entries
1385 * from the enclave data structure.
1386 * @ne_enclave : Private data associated with the current enclave.
1388 * Context: Process context. This function is called with the ne_enclave mutex held.
1390 static void ne_enclave_remove_all_mem_region_entries(struct ne_enclave *ne_enclave)
1392 unsigned long i = 0;
1393 struct ne_mem_region *ne_mem_region = NULL;
1394 struct ne_mem_region *ne_mem_region_tmp = NULL;
1396 list_for_each_entry_safe(ne_mem_region, ne_mem_region_tmp,
1397 &ne_enclave->mem_regions_list,
1398 mem_region_list_entry) {
1399 list_del(&ne_mem_region->mem_region_list_entry);
1401 for (i = 0; i < ne_mem_region->nr_pages; i++)
1402 put_page(ne_mem_region->pages[i]);
1404 kfree(ne_mem_region->pages);
1406 kfree(ne_mem_region);
1411 * ne_enclave_remove_all_vcpu_id_entries() - Remove all vCPU id entries from
1412 * the enclave data structure.
1413 * @ne_enclave : Private data associated with the current enclave.
1415 * Context: Process context. This function is called with the ne_enclave mutex held.
1417 static void ne_enclave_remove_all_vcpu_id_entries(struct ne_enclave *ne_enclave)
1419 unsigned int cpu = 0;
1422 mutex_lock(&ne_cpu_pool.mutex);
1424 for (i = 0; i < ne_enclave->nr_parent_vm_cores; i++) {
1425 for_each_cpu(cpu, ne_enclave->threads_per_core[i])
1426 /* Update the available NE CPU pool. */
1427 cpumask_set_cpu(cpu, ne_cpu_pool.avail_threads_per_core[i]);
1429 free_cpumask_var(ne_enclave->threads_per_core[i]);
1432 mutex_unlock(&ne_cpu_pool.mutex);
1434 kfree(ne_enclave->threads_per_core);
1436 free_cpumask_var(ne_enclave->vcpu_ids);
1440 * ne_pci_dev_remove_enclave_entry() - Remove the enclave entry from the data
1441 * structure that is part of the NE PCI
1442 * device private data.
1443 * @ne_enclave : Private data associated with the current enclave.
1444 * @ne_pci_dev : Private data associated with the PCI device.
1446 * Context: Process context. This function is called with the ne_pci_dev enclave
1449 static void ne_pci_dev_remove_enclave_entry(struct ne_enclave *ne_enclave,
1450 struct ne_pci_dev *ne_pci_dev)
1452 struct ne_enclave *ne_enclave_entry = NULL;
1453 struct ne_enclave *ne_enclave_entry_tmp = NULL;
1455 list_for_each_entry_safe(ne_enclave_entry, ne_enclave_entry_tmp,
1456 &ne_pci_dev->enclaves_list, enclave_list_entry) {
1457 if (ne_enclave_entry->slot_uid == ne_enclave->slot_uid) {
1458 list_del(&ne_enclave_entry->enclave_list_entry);
1466 * ne_enclave_release() - Release function provided by the enclave file.
1467 * @inode: Inode associated with this file release function.
1468 * @file: File associated with this release function.
1470 * Context: Process context.
1473 * * Negative return value on failure.
1475 static int ne_enclave_release(struct inode *inode, struct file *file)
1477 struct ne_pci_dev_cmd_reply cmd_reply = {};
1478 struct enclave_stop_req enclave_stop_request = {};
1479 struct ne_enclave *ne_enclave = file->private_data;
1480 struct ne_pci_dev *ne_pci_dev = ne_devs.ne_pci_dev;
1481 struct pci_dev *pdev = ne_pci_dev->pdev;
1483 struct slot_free_req slot_free_req = {};
1489 * Early exit in case there is an error in the enclave creation logic
1490 * and fput() is called on the cleanup path.
1492 if (!ne_enclave->slot_uid)
1496 * Acquire the enclave list mutex before the enclave mutex
1497 * in order to avoid deadlocks with @ref ne_event_work_handler.
1499 mutex_lock(&ne_pci_dev->enclaves_list_mutex);
1500 mutex_lock(&ne_enclave->enclave_info_mutex);
1502 if (ne_enclave->state != NE_STATE_INIT && ne_enclave->state != NE_STATE_STOPPED) {
1503 enclave_stop_request.slot_uid = ne_enclave->slot_uid;
1505 rc = ne_do_request(pdev, ENCLAVE_STOP,
1506 &enclave_stop_request, sizeof(enclave_stop_request),
1507 &cmd_reply, sizeof(cmd_reply));
1509 dev_err_ratelimited(ne_misc_dev.this_device,
1510 "Error in enclave stop [rc=%d]\n", rc);
1515 memset(&cmd_reply, 0, sizeof(cmd_reply));
1518 slot_free_req.slot_uid = ne_enclave->slot_uid;
1520 rc = ne_do_request(pdev, SLOT_FREE,
1521 &slot_free_req, sizeof(slot_free_req),
1522 &cmd_reply, sizeof(cmd_reply));
1524 dev_err_ratelimited(ne_misc_dev.this_device,
1525 "Error in slot free [rc=%d]\n", rc);
1530 ne_pci_dev_remove_enclave_entry(ne_enclave, ne_pci_dev);
1531 ne_enclave_remove_all_mem_region_entries(ne_enclave);
1532 ne_enclave_remove_all_vcpu_id_entries(ne_enclave);
1534 mutex_unlock(&ne_enclave->enclave_info_mutex);
1535 mutex_unlock(&ne_pci_dev->enclaves_list_mutex);
1542 mutex_unlock(&ne_enclave->enclave_info_mutex);
1543 mutex_unlock(&ne_pci_dev->enclaves_list_mutex);
1549 * ne_enclave_poll() - Poll functionality used for enclave out-of-band events.
1550 * @file: File associated with this poll function.
1551 * @wait: Poll table data structure.
1553 * Context: Process context.
1557 static __poll_t ne_enclave_poll(struct file *file, poll_table *wait)
1560 struct ne_enclave *ne_enclave = file->private_data;
1562 poll_wait(file, &ne_enclave->eventq, wait);
1564 if (ne_enclave->has_event)
1570 static const struct file_operations ne_enclave_fops = {
1571 .owner = THIS_MODULE,
1572 .llseek = noop_llseek,
1573 .poll = ne_enclave_poll,
1574 .unlocked_ioctl = ne_enclave_ioctl,
1575 .release = ne_enclave_release,
1579 * ne_create_vm_ioctl() - Alloc slot to be associated with an enclave. Create
1580 * enclave file descriptor to be further used for enclave
1581 * resources handling e.g. memory regions and CPUs.
1582 * @ne_pci_dev : Private data associated with the PCI device.
1583 * @slot_uid: User pointer to store the generated unique slot id
1584 * associated with an enclave to.
1586 * Context: Process context. This function is called with the ne_pci_dev enclave
1589 * * Enclave fd on success.
1590 * * Negative return value on failure.
1592 static int ne_create_vm_ioctl(struct ne_pci_dev *ne_pci_dev, u64 __user *slot_uid)
1594 struct ne_pci_dev_cmd_reply cmd_reply = {};
1595 int enclave_fd = -1;
1596 struct file *enclave_file = NULL;
1598 struct ne_enclave *ne_enclave = NULL;
1599 struct pci_dev *pdev = ne_pci_dev->pdev;
1601 struct slot_alloc_req slot_alloc_req = {};
1603 mutex_lock(&ne_cpu_pool.mutex);
1605 for (i = 0; i < ne_cpu_pool.nr_parent_vm_cores; i++)
1606 if (!cpumask_empty(ne_cpu_pool.avail_threads_per_core[i]))
1609 if (i == ne_cpu_pool.nr_parent_vm_cores) {
1610 dev_err_ratelimited(ne_misc_dev.this_device,
1611 "No CPUs available in CPU pool\n");
1613 mutex_unlock(&ne_cpu_pool.mutex);
1615 return -NE_ERR_NO_CPUS_AVAIL_IN_POOL;
1618 mutex_unlock(&ne_cpu_pool.mutex);
1620 ne_enclave = kzalloc(sizeof(*ne_enclave), GFP_KERNEL);
1624 mutex_lock(&ne_cpu_pool.mutex);
1626 ne_enclave->nr_parent_vm_cores = ne_cpu_pool.nr_parent_vm_cores;
1627 ne_enclave->nr_threads_per_core = ne_cpu_pool.nr_threads_per_core;
1628 ne_enclave->numa_node = ne_cpu_pool.numa_node;
1630 mutex_unlock(&ne_cpu_pool.mutex);
1632 ne_enclave->threads_per_core = kcalloc(ne_enclave->nr_parent_vm_cores,
1633 sizeof(*ne_enclave->threads_per_core),
1635 if (!ne_enclave->threads_per_core) {
1638 goto free_ne_enclave;
1641 for (i = 0; i < ne_enclave->nr_parent_vm_cores; i++)
1642 if (!zalloc_cpumask_var(&ne_enclave->threads_per_core[i], GFP_KERNEL)) {
1648 if (!zalloc_cpumask_var(&ne_enclave->vcpu_ids, GFP_KERNEL)) {
1654 enclave_fd = get_unused_fd_flags(O_CLOEXEC);
1655 if (enclave_fd < 0) {
1658 dev_err_ratelimited(ne_misc_dev.this_device,
1659 "Error in getting unused fd [rc=%d]\n", rc);
1664 enclave_file = anon_inode_getfile("ne-vm", &ne_enclave_fops, ne_enclave, O_RDWR);
1665 if (IS_ERR(enclave_file)) {
1666 rc = PTR_ERR(enclave_file);
1668 dev_err_ratelimited(ne_misc_dev.this_device,
1669 "Error in anon inode get file [rc=%d]\n", rc);
1674 rc = ne_do_request(pdev, SLOT_ALLOC,
1675 &slot_alloc_req, sizeof(slot_alloc_req),
1676 &cmd_reply, sizeof(cmd_reply));
1678 dev_err_ratelimited(ne_misc_dev.this_device,
1679 "Error in slot alloc [rc=%d]\n", rc);
1684 init_waitqueue_head(&ne_enclave->eventq);
1685 ne_enclave->has_event = false;
1686 mutex_init(&ne_enclave->enclave_info_mutex);
1687 ne_enclave->max_mem_regions = cmd_reply.mem_regions;
1688 INIT_LIST_HEAD(&ne_enclave->mem_regions_list);
1689 ne_enclave->mm = current->mm;
1690 ne_enclave->slot_uid = cmd_reply.slot_uid;
1691 ne_enclave->state = NE_STATE_INIT;
1693 list_add(&ne_enclave->enclave_list_entry, &ne_pci_dev->enclaves_list);
1695 if (copy_to_user(slot_uid, &ne_enclave->slot_uid, sizeof(ne_enclave->slot_uid))) {
1697 * As we're holding the only reference to 'enclave_file', fput()
1698 * will call ne_enclave_release() which will do a proper cleanup
1699 * of all so far allocated resources, leaving only the unused fd
1703 put_unused_fd(enclave_fd);
1708 fd_install(enclave_fd, enclave_file);
1715 put_unused_fd(enclave_fd);
1717 free_cpumask_var(ne_enclave->vcpu_ids);
1718 for (i = 0; i < ne_enclave->nr_parent_vm_cores; i++)
1719 free_cpumask_var(ne_enclave->threads_per_core[i]);
1720 kfree(ne_enclave->threads_per_core);
1728 * ne_ioctl() - Ioctl function provided by the NE misc device.
1729 * @file: File associated with this ioctl function.
1730 * @cmd: The command that is set for the ioctl call.
1731 * @arg: The argument that is provided for the ioctl call.
1733 * Context: Process context.
1735 * * Ioctl result (e.g. enclave file descriptor) on success.
1736 * * Negative return value on failure.
1738 static long ne_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1741 case NE_CREATE_VM: {
1742 int enclave_fd = -1;
1743 struct ne_pci_dev *ne_pci_dev = ne_devs.ne_pci_dev;
1744 u64 __user *slot_uid = (void __user *)arg;
1746 mutex_lock(&ne_pci_dev->enclaves_list_mutex);
1747 enclave_fd = ne_create_vm_ioctl(ne_pci_dev, slot_uid);
1748 mutex_unlock(&ne_pci_dev->enclaves_list_mutex);
1760 #if defined(CONFIG_NITRO_ENCLAVES_MISC_DEV_TEST)
1761 #include "ne_misc_dev_test.c"
1764 static int __init ne_init(void)
1766 mutex_init(&ne_cpu_pool.mutex);
1768 return pci_register_driver(&ne_pci_driver);
1771 static void __exit ne_exit(void)
1773 pci_unregister_driver(&ne_pci_driver);
1775 ne_teardown_cpu_pool();
1778 module_init(ne_init);
1779 module_exit(ne_exit);
1781 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
1782 MODULE_DESCRIPTION("Nitro Enclaves Driver");
1783 MODULE_LICENSE("GPL v2");