xref: /openbmc/qemu/hw/ppc/spapr_nvdimm.c (revision b8012ecf)
1 /*
2  * QEMU PAPR Storage Class Memory Interfaces
3  *
4  * Copyright (c) 2019-2020, IBM Corporation.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #include "qemu/osdep.h"
25 #include "qemu/cutils.h"
26 #include "qapi/error.h"
27 #include "hw/ppc/spapr_drc.h"
28 #include "hw/ppc/spapr_nvdimm.h"
29 #include "hw/mem/nvdimm.h"
30 #include "qemu/nvdimm-utils.h"
31 #include "hw/ppc/fdt.h"
32 #include "qemu/range.h"
33 #include "hw/ppc/spapr_numa.h"
34 #include "block/thread-pool.h"
35 #include "migration/vmstate.h"
36 #include "qemu/pmem.h"
37 #include "hw/qdev-properties.h"
38 
39 /* DIMM health bitmap bitmap indicators. Taken from kernel's papr_scm.c */
40 /* SCM device is unable to persist memory contents */
41 #define PAPR_PMEM_UNARMED PPC_BIT(0)
42 
43 /*
44  * The nvdimm size should be aligned to SCM block size.
45  * The SCM block size should be aligned to SPAPR_MEMORY_BLOCK_SIZE
46  * in order to have SCM regions not to overlap with dimm memory regions.
47  * The SCM devices can have variable block sizes. For now, fixing the
48  * block size to the minimum value.
49  */
50 #define SPAPR_MINIMUM_SCM_BLOCK_SIZE SPAPR_MEMORY_BLOCK_SIZE
51 
52 /* Have an explicit check for alignment */
53 QEMU_BUILD_BUG_ON(SPAPR_MINIMUM_SCM_BLOCK_SIZE % SPAPR_MEMORY_BLOCK_SIZE);
54 
55 #define TYPE_SPAPR_NVDIMM "spapr-nvdimm"
56 OBJECT_DECLARE_TYPE(SpaprNVDIMMDevice, SPAPRNVDIMMClass, SPAPR_NVDIMM)
57 
58 struct SPAPRNVDIMMClass {
59     /* private */
60     NVDIMMClass parent_class;
61 
62     /* public */
63     void (*realize)(NVDIMMDevice *dimm, Error **errp);
64     void (*unrealize)(NVDIMMDevice *dimm, Error **errp);
65 };
66 
67 bool spapr_nvdimm_validate(HotplugHandler *hotplug_dev, NVDIMMDevice *nvdimm,
68                            uint64_t size, Error **errp)
69 {
70     const MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
71     const MachineState *ms = MACHINE(hotplug_dev);
72     PCDIMMDevice *dimm = PC_DIMM(nvdimm);
73     MemoryRegion *mr = host_memory_backend_get_memory(dimm->hostmem);
74     g_autofree char *uuidstr = NULL;
75     QemuUUID uuid;
76     int ret;
77 
78     if (!mc->nvdimm_supported) {
79         error_setg(errp, "NVDIMM hotplug not supported for this machine");
80         return false;
81     }
82 
83     if (!ms->nvdimms_state->is_enabled) {
84         error_setg(errp, "nvdimm device found but 'nvdimm=off' was set");
85         return false;
86     }
87 
88     if (object_property_get_int(OBJECT(nvdimm), NVDIMM_LABEL_SIZE_PROP,
89                                 &error_abort) == 0) {
90         error_setg(errp, "PAPR requires NVDIMM devices to have label-size set");
91         return false;
92     }
93 
94     if (size % SPAPR_MINIMUM_SCM_BLOCK_SIZE) {
95         error_setg(errp, "PAPR requires NVDIMM memory size (excluding label)"
96                    " to be a multiple of %" PRIu64 "MB",
97                    SPAPR_MINIMUM_SCM_BLOCK_SIZE / MiB);
98         return false;
99     }
100 
101     uuidstr = object_property_get_str(OBJECT(nvdimm), NVDIMM_UUID_PROP,
102                                       &error_abort);
103     ret = qemu_uuid_parse(uuidstr, &uuid);
104     g_assert(!ret);
105 
106     if (qemu_uuid_is_null(&uuid)) {
107         error_setg(errp, "NVDIMM device requires the uuid to be set");
108         return false;
109     }
110 
111     if (object_dynamic_cast(OBJECT(nvdimm), TYPE_SPAPR_NVDIMM) &&
112         (memory_region_get_fd(mr) < 0)) {
113         error_setg(errp, "spapr-nvdimm device requires the "
114                    "memdev %s to be of memory-backend-file type",
115                    object_get_canonical_path_component(OBJECT(dimm->hostmem)));
116         return false;
117     }
118 
119     return true;
120 }
121 
122 
123 void spapr_add_nvdimm(DeviceState *dev, uint64_t slot)
124 {
125     SpaprDrc *drc;
126     bool hotplugged = spapr_drc_hotplugged(dev);
127 
128     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PMEM, slot);
129     g_assert(drc);
130 
131     /*
132      * pc_dimm_get_free_slot() provided a free slot at pre-plug. The
133      * corresponding DRC is thus assumed to be attachable.
134      */
135     spapr_drc_attach(drc, dev);
136 
137     if (hotplugged) {
138         spapr_hotplug_req_add_by_index(drc);
139     }
140 }
141 
142 static int spapr_dt_nvdimm(SpaprMachineState *spapr, void *fdt,
143                            int parent_offset, NVDIMMDevice *nvdimm)
144 {
145     int child_offset;
146     char *buf;
147     SpaprDrc *drc;
148     uint32_t drc_idx;
149     uint32_t node = object_property_get_uint(OBJECT(nvdimm), PC_DIMM_NODE_PROP,
150                                              &error_abort);
151     uint64_t slot = object_property_get_uint(OBJECT(nvdimm), PC_DIMM_SLOT_PROP,
152                                              &error_abort);
153     uint64_t lsize = nvdimm->label_size;
154     uint64_t size = object_property_get_int(OBJECT(nvdimm), PC_DIMM_SIZE_PROP,
155                                             NULL);
156 
157     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PMEM, slot);
158     g_assert(drc);
159 
160     drc_idx = spapr_drc_index(drc);
161 
162     buf = g_strdup_printf("ibm,pmemory@%x", drc_idx);
163     child_offset = fdt_add_subnode(fdt, parent_offset, buf);
164     g_free(buf);
165 
166     _FDT(child_offset);
167 
168     _FDT((fdt_setprop_cell(fdt, child_offset, "reg", drc_idx)));
169     _FDT((fdt_setprop_string(fdt, child_offset, "compatible", "ibm,pmemory")));
170     _FDT((fdt_setprop_string(fdt, child_offset, "device_type", "ibm,pmemory")));
171 
172     spapr_numa_write_associativity_dt(spapr, fdt, child_offset, node);
173 
174     buf = qemu_uuid_unparse_strdup(&nvdimm->uuid);
175     _FDT((fdt_setprop_string(fdt, child_offset, "ibm,unit-guid", buf)));
176     g_free(buf);
177 
178     _FDT((fdt_setprop_cell(fdt, child_offset, "ibm,my-drc-index", drc_idx)));
179 
180     _FDT((fdt_setprop_u64(fdt, child_offset, "ibm,block-size",
181                           SPAPR_MINIMUM_SCM_BLOCK_SIZE)));
182     _FDT((fdt_setprop_u64(fdt, child_offset, "ibm,number-of-blocks",
183                           size / SPAPR_MINIMUM_SCM_BLOCK_SIZE)));
184     _FDT((fdt_setprop_cell(fdt, child_offset, "ibm,metadata-size", lsize)));
185 
186     _FDT((fdt_setprop_string(fdt, child_offset, "ibm,pmem-application",
187                              "operating-system")));
188     _FDT(fdt_setprop(fdt, child_offset, "ibm,cache-flush-required", NULL, 0));
189 
190     if (object_dynamic_cast(OBJECT(nvdimm), TYPE_SPAPR_NVDIMM)) {
191         bool is_pmem = false, pmem_override = false;
192         PCDIMMDevice *dimm = PC_DIMM(nvdimm);
193         HostMemoryBackend *hostmem = dimm->hostmem;
194 
195         is_pmem = object_property_get_bool(OBJECT(hostmem), "pmem", NULL);
196         pmem_override = object_property_get_bool(OBJECT(nvdimm),
197                                                  "pmem-override", NULL);
198         if (!is_pmem || pmem_override) {
199             _FDT(fdt_setprop(fdt, child_offset, "ibm,hcall-flush-required",
200                              NULL, 0));
201         }
202     }
203 
204     return child_offset;
205 }
206 
207 int spapr_pmem_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
208                            void *fdt, int *fdt_start_offset, Error **errp)
209 {
210     NVDIMMDevice *nvdimm = NVDIMM(drc->dev);
211 
212     *fdt_start_offset = spapr_dt_nvdimm(spapr, fdt, 0, nvdimm);
213 
214     return 0;
215 }
216 
217 void spapr_dt_persistent_memory(SpaprMachineState *spapr, void *fdt)
218 {
219     int offset = fdt_subnode_offset(fdt, 0, "ibm,persistent-memory");
220     GSList *iter, *nvdimms = nvdimm_get_device_list();
221 
222     if (offset < 0) {
223         offset = fdt_add_subnode(fdt, 0, "ibm,persistent-memory");
224         _FDT(offset);
225         _FDT((fdt_setprop_cell(fdt, offset, "#address-cells", 0x1)));
226         _FDT((fdt_setprop_cell(fdt, offset, "#size-cells", 0x0)));
227         _FDT((fdt_setprop_string(fdt, offset, "device_type",
228                                  "ibm,persistent-memory")));
229     }
230 
231     /* Create DT entries for cold plugged NVDIMM devices */
232     for (iter = nvdimms; iter; iter = iter->next) {
233         NVDIMMDevice *nvdimm = iter->data;
234 
235         spapr_dt_nvdimm(spapr, fdt, offset, nvdimm);
236     }
237     g_slist_free(nvdimms);
238 
239     return;
240 }
241 
242 static target_ulong h_scm_read_metadata(PowerPCCPU *cpu,
243                                         SpaprMachineState *spapr,
244                                         target_ulong opcode,
245                                         target_ulong *args)
246 {
247     uint32_t drc_index = args[0];
248     uint64_t offset = args[1];
249     uint64_t len = args[2];
250     SpaprDrc *drc = spapr_drc_by_index(drc_index);
251     NVDIMMDevice *nvdimm;
252     NVDIMMClass *ddc;
253     uint64_t data = 0;
254     uint8_t buf[8] = { 0 };
255 
256     if (!drc || !drc->dev ||
257         spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
258         return H_PARAMETER;
259     }
260 
261     if (len != 1 && len != 2 &&
262         len != 4 && len != 8) {
263         return H_P3;
264     }
265 
266     nvdimm = NVDIMM(drc->dev);
267     if ((offset + len < offset) ||
268         (nvdimm->label_size < len + offset)) {
269         return H_P2;
270     }
271 
272     ddc = NVDIMM_GET_CLASS(nvdimm);
273     ddc->read_label_data(nvdimm, buf, len, offset);
274 
275     switch (len) {
276     case 1:
277         data = ldub_p(buf);
278         break;
279     case 2:
280         data = lduw_be_p(buf);
281         break;
282     case 4:
283         data = ldl_be_p(buf);
284         break;
285     case 8:
286         data = ldq_be_p(buf);
287         break;
288     default:
289         g_assert_not_reached();
290     }
291 
292     args[0] = data;
293 
294     return H_SUCCESS;
295 }
296 
297 static target_ulong h_scm_write_metadata(PowerPCCPU *cpu,
298                                          SpaprMachineState *spapr,
299                                          target_ulong opcode,
300                                          target_ulong *args)
301 {
302     uint32_t drc_index = args[0];
303     uint64_t offset = args[1];
304     uint64_t data = args[2];
305     uint64_t len = args[3];
306     SpaprDrc *drc = spapr_drc_by_index(drc_index);
307     NVDIMMDevice *nvdimm;
308     NVDIMMClass *ddc;
309     uint8_t buf[8] = { 0 };
310 
311     if (!drc || !drc->dev ||
312         spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
313         return H_PARAMETER;
314     }
315 
316     if (len != 1 && len != 2 &&
317         len != 4 && len != 8) {
318         return H_P4;
319     }
320 
321     nvdimm = NVDIMM(drc->dev);
322     if ((offset + len < offset) ||
323         (nvdimm->label_size < len + offset)) {
324         return H_P2;
325     }
326 
327     switch (len) {
328     case 1:
329         if (data & 0xffffffffffffff00) {
330             return H_P2;
331         }
332         stb_p(buf, data);
333         break;
334     case 2:
335         if (data & 0xffffffffffff0000) {
336             return H_P2;
337         }
338         stw_be_p(buf, data);
339         break;
340     case 4:
341         if (data & 0xffffffff00000000) {
342             return H_P2;
343         }
344         stl_be_p(buf, data);
345         break;
346     case 8:
347         stq_be_p(buf, data);
348         break;
349     default:
350             g_assert_not_reached();
351     }
352 
353     ddc = NVDIMM_GET_CLASS(nvdimm);
354     ddc->write_label_data(nvdimm, buf, len, offset);
355 
356     return H_SUCCESS;
357 }
358 
359 static target_ulong h_scm_bind_mem(PowerPCCPU *cpu, SpaprMachineState *spapr,
360                                    target_ulong opcode, target_ulong *args)
361 {
362     uint32_t drc_index = args[0];
363     uint64_t starting_idx = args[1];
364     uint64_t no_of_scm_blocks_to_bind = args[2];
365     uint64_t target_logical_mem_addr = args[3];
366     uint64_t continue_token = args[4];
367     uint64_t size;
368     uint64_t total_no_of_scm_blocks;
369     SpaprDrc *drc = spapr_drc_by_index(drc_index);
370     hwaddr addr;
371     NVDIMMDevice *nvdimm;
372 
373     if (!drc || !drc->dev ||
374         spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
375         return H_PARAMETER;
376     }
377 
378     /*
379      * Currently continue token should be zero qemu has already bound
380      * everything and this hcall doesnt return H_BUSY.
381      */
382     if (continue_token > 0) {
383         return H_P5;
384     }
385 
386     /* Currently qemu assigns the address. */
387     if (target_logical_mem_addr != 0xffffffffffffffff) {
388         return H_OVERLAP;
389     }
390 
391     nvdimm = NVDIMM(drc->dev);
392 
393     size = object_property_get_uint(OBJECT(nvdimm),
394                                     PC_DIMM_SIZE_PROP, &error_abort);
395 
396     total_no_of_scm_blocks = size / SPAPR_MINIMUM_SCM_BLOCK_SIZE;
397 
398     if (starting_idx > total_no_of_scm_blocks) {
399         return H_P2;
400     }
401 
402     if (((starting_idx + no_of_scm_blocks_to_bind) < starting_idx) ||
403         ((starting_idx + no_of_scm_blocks_to_bind) > total_no_of_scm_blocks)) {
404         return H_P3;
405     }
406 
407     addr = object_property_get_uint(OBJECT(nvdimm),
408                                     PC_DIMM_ADDR_PROP, &error_abort);
409 
410     addr += starting_idx * SPAPR_MINIMUM_SCM_BLOCK_SIZE;
411 
412     /* Already bound, Return target logical address in R5 */
413     args[1] = addr;
414     args[2] = no_of_scm_blocks_to_bind;
415 
416     return H_SUCCESS;
417 }
418 
419 typedef struct SpaprNVDIMMDeviceFlushState {
420     uint64_t continue_token;
421     int64_t hcall_ret;
422     uint32_t drcidx;
423 
424     QLIST_ENTRY(SpaprNVDIMMDeviceFlushState) node;
425 } SpaprNVDIMMDeviceFlushState;
426 
427 typedef struct SpaprNVDIMMDevice SpaprNVDIMMDevice;
428 struct SpaprNVDIMMDevice {
429     /* private */
430     NVDIMMDevice parent_obj;
431 
432     bool hcall_flush_required;
433     uint64_t nvdimm_flush_token;
434     QLIST_HEAD(, SpaprNVDIMMDeviceFlushState) pending_nvdimm_flush_states;
435     QLIST_HEAD(, SpaprNVDIMMDeviceFlushState) completed_nvdimm_flush_states;
436 
437     /* public */
438 
439     /*
440      * The 'on' value for this property forced the qemu to enable the hcall
441      * flush for the nvdimm device even if the backend is a pmem
442      */
443     bool pmem_override;
444 };
445 
446 static int flush_worker_cb(void *opaque)
447 {
448     SpaprNVDIMMDeviceFlushState *state = opaque;
449     SpaprDrc *drc = spapr_drc_by_index(state->drcidx);
450     PCDIMMDevice *dimm = PC_DIMM(drc->dev);
451     HostMemoryBackend *backend = MEMORY_BACKEND(dimm->hostmem);
452     int backend_fd = memory_region_get_fd(&backend->mr);
453 
454     if (object_property_get_bool(OBJECT(backend), "pmem", NULL)) {
455         MemoryRegion *mr = host_memory_backend_get_memory(dimm->hostmem);
456         void *ptr = memory_region_get_ram_ptr(mr);
457         size_t size = object_property_get_uint(OBJECT(dimm), PC_DIMM_SIZE_PROP,
458                                                NULL);
459 
460         /* flush pmem backend */
461         pmem_persist(ptr, size);
462     } else {
463         /* flush raw backing image */
464         if (qemu_fdatasync(backend_fd) < 0) {
465             error_report("papr_scm: Could not sync nvdimm to backend file: %s",
466                          strerror(errno));
467             return H_HARDWARE;
468         }
469     }
470 
471     return H_SUCCESS;
472 }
473 
474 static void spapr_nvdimm_flush_completion_cb(void *opaque, int hcall_ret)
475 {
476     SpaprNVDIMMDeviceFlushState *state = opaque;
477     SpaprDrc *drc = spapr_drc_by_index(state->drcidx);
478     SpaprNVDIMMDevice *s_nvdimm = SPAPR_NVDIMM(drc->dev);
479 
480     state->hcall_ret = hcall_ret;
481     QLIST_REMOVE(state, node);
482     QLIST_INSERT_HEAD(&s_nvdimm->completed_nvdimm_flush_states, state, node);
483 }
484 
485 static int spapr_nvdimm_flush_post_load(void *opaque, int version_id)
486 {
487     SpaprNVDIMMDevice *s_nvdimm = (SpaprNVDIMMDevice *)opaque;
488     SpaprNVDIMMDeviceFlushState *state;
489     ThreadPool *pool = aio_get_thread_pool(qemu_get_aio_context());
490     HostMemoryBackend *backend = MEMORY_BACKEND(PC_DIMM(s_nvdimm)->hostmem);
491     bool is_pmem = object_property_get_bool(OBJECT(backend), "pmem", NULL);
492     bool pmem_override = object_property_get_bool(OBJECT(s_nvdimm),
493                                                   "pmem-override", NULL);
494     bool dest_hcall_flush_required = pmem_override || !is_pmem;
495 
496     if (!s_nvdimm->hcall_flush_required && dest_hcall_flush_required) {
497         error_report("The file backend for the spapr-nvdimm device %s at "
498                      "source is a pmem, use pmem=on and pmem-override=off to "
499                      "continue.", DEVICE(s_nvdimm)->id);
500         return -EINVAL;
501     }
502     if (s_nvdimm->hcall_flush_required && !dest_hcall_flush_required) {
503         error_report("The guest expects hcall-flush support for the "
504                      "spapr-nvdimm device %s, use pmem_override=on to "
505                      "continue.", DEVICE(s_nvdimm)->id);
506         return -EINVAL;
507     }
508 
509     QLIST_FOREACH(state, &s_nvdimm->pending_nvdimm_flush_states, node) {
510         thread_pool_submit_aio(pool, flush_worker_cb, state,
511                                spapr_nvdimm_flush_completion_cb, state);
512     }
513 
514     return 0;
515 }
516 
517 static const VMStateDescription vmstate_spapr_nvdimm_flush_state = {
518      .name = "spapr_nvdimm_flush_state",
519      .version_id = 1,
520      .minimum_version_id = 1,
521      .fields = (VMStateField[]) {
522          VMSTATE_UINT64(continue_token, SpaprNVDIMMDeviceFlushState),
523          VMSTATE_INT64(hcall_ret, SpaprNVDIMMDeviceFlushState),
524          VMSTATE_UINT32(drcidx, SpaprNVDIMMDeviceFlushState),
525          VMSTATE_END_OF_LIST()
526      },
527 };
528 
529 const VMStateDescription vmstate_spapr_nvdimm_states = {
530     .name = "spapr_nvdimm_states",
531     .version_id = 1,
532     .minimum_version_id = 1,
533     .post_load = spapr_nvdimm_flush_post_load,
534     .fields = (VMStateField[]) {
535         VMSTATE_BOOL(hcall_flush_required, SpaprNVDIMMDevice),
536         VMSTATE_UINT64(nvdimm_flush_token, SpaprNVDIMMDevice),
537         VMSTATE_QLIST_V(completed_nvdimm_flush_states, SpaprNVDIMMDevice, 1,
538                         vmstate_spapr_nvdimm_flush_state,
539                         SpaprNVDIMMDeviceFlushState, node),
540         VMSTATE_QLIST_V(pending_nvdimm_flush_states, SpaprNVDIMMDevice, 1,
541                         vmstate_spapr_nvdimm_flush_state,
542                         SpaprNVDIMMDeviceFlushState, node),
543         VMSTATE_END_OF_LIST()
544     },
545 };
546 
547 /*
548  * Assign a token and reserve it for the new flush state.
549  */
550 static SpaprNVDIMMDeviceFlushState *spapr_nvdimm_init_new_flush_state(
551                                                 SpaprNVDIMMDevice *spapr_nvdimm)
552 {
553     SpaprNVDIMMDeviceFlushState *state;
554 
555     state = g_malloc0(sizeof(*state));
556 
557     spapr_nvdimm->nvdimm_flush_token++;
558     /* Token zero is presumed as no job pending. Assert on overflow to zero */
559     g_assert(spapr_nvdimm->nvdimm_flush_token != 0);
560 
561     state->continue_token = spapr_nvdimm->nvdimm_flush_token;
562 
563     QLIST_INSERT_HEAD(&spapr_nvdimm->pending_nvdimm_flush_states, state, node);
564 
565     return state;
566 }
567 
568 /*
569  * spapr_nvdimm_finish_flushes
570  *      Waits for all pending flush requests to complete
571  *      their execution and free the states
572  */
573 void spapr_nvdimm_finish_flushes(void)
574 {
575     SpaprNVDIMMDeviceFlushState *state, *next;
576     GSList *list, *nvdimms;
577 
578     /*
579      * Called on reset path, the main loop thread which calls
580      * the pending BHs has gotten out running in the reset path,
581      * finally reaching here. Other code path being guest
582      * h_client_architecture_support, thats early boot up.
583      */
584     nvdimms = nvdimm_get_device_list();
585     for (list = nvdimms; list; list = list->next) {
586         NVDIMMDevice *nvdimm = list->data;
587         if (object_dynamic_cast(OBJECT(nvdimm), TYPE_SPAPR_NVDIMM)) {
588             SpaprNVDIMMDevice *s_nvdimm = SPAPR_NVDIMM(nvdimm);
589             while (!QLIST_EMPTY(&s_nvdimm->pending_nvdimm_flush_states)) {
590                 aio_poll(qemu_get_aio_context(), true);
591             }
592 
593             QLIST_FOREACH_SAFE(state, &s_nvdimm->completed_nvdimm_flush_states,
594                                node, next) {
595                 QLIST_REMOVE(state, node);
596                 g_free(state);
597             }
598         }
599     }
600     g_slist_free(nvdimms);
601 }
602 
603 /*
604  * spapr_nvdimm_get_flush_status
605  *      Fetches the status of the hcall worker and returns
606  *      H_LONG_BUSY_ORDER_10_MSEC if the worker is still running.
607  */
608 static int spapr_nvdimm_get_flush_status(SpaprNVDIMMDevice *s_nvdimm,
609                                          uint64_t token)
610 {
611     SpaprNVDIMMDeviceFlushState *state, *node;
612 
613     QLIST_FOREACH(state, &s_nvdimm->pending_nvdimm_flush_states, node) {
614         if (state->continue_token == token) {
615             return H_LONG_BUSY_ORDER_10_MSEC;
616         }
617     }
618 
619     QLIST_FOREACH_SAFE(state, &s_nvdimm->completed_nvdimm_flush_states,
620                        node, node) {
621         if (state->continue_token == token) {
622             int ret = state->hcall_ret;
623             QLIST_REMOVE(state, node);
624             g_free(state);
625             return ret;
626         }
627     }
628 
629     /* If not found in complete list too, invalid token */
630     return H_P2;
631 }
632 
633 /*
634  * H_SCM_FLUSH
635  * Input: drc_index, continue-token
636  * Out: continue-token
637  * Return Value: H_SUCCESS, H_Parameter, H_P2, H_LONG_BUSY_ORDER_10_MSEC,
638  *               H_UNSUPPORTED
639  *
640  * Given a DRC Index Flush the data to backend NVDIMM device. The hcall returns
641  * H_LONG_BUSY_ORDER_10_MSEC when the flush takes longer time and the hcall
642  * needs to be issued multiple times in order to be completely serviced. The
643  * continue-token from the output to be passed in the argument list of
644  * subsequent hcalls until the hcall is completely serviced at which point
645  * H_SUCCESS or other error is returned.
646  */
647 static target_ulong h_scm_flush(PowerPCCPU *cpu, SpaprMachineState *spapr,
648                                 target_ulong opcode, target_ulong *args)
649 {
650     int ret;
651     uint32_t drc_index = args[0];
652     uint64_t continue_token = args[1];
653     SpaprDrc *drc = spapr_drc_by_index(drc_index);
654     PCDIMMDevice *dimm;
655     HostMemoryBackend *backend = NULL;
656     SpaprNVDIMMDeviceFlushState *state;
657     ThreadPool *pool = aio_get_thread_pool(qemu_get_aio_context());
658     int fd;
659 
660     if (!drc || !drc->dev ||
661         spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
662         return H_PARAMETER;
663     }
664 
665     dimm = PC_DIMM(drc->dev);
666     if (!object_dynamic_cast(OBJECT(dimm), TYPE_SPAPR_NVDIMM)) {
667         return H_PARAMETER;
668     }
669     if (continue_token == 0) {
670         bool is_pmem = false, pmem_override = false;
671         backend = MEMORY_BACKEND(dimm->hostmem);
672         fd = memory_region_get_fd(&backend->mr);
673 
674         if (fd < 0) {
675             return H_UNSUPPORTED;
676         }
677 
678         is_pmem = object_property_get_bool(OBJECT(backend), "pmem", NULL);
679         pmem_override = object_property_get_bool(OBJECT(dimm),
680                                                 "pmem-override", NULL);
681         if (is_pmem && !pmem_override) {
682             return H_UNSUPPORTED;
683         }
684 
685         state = spapr_nvdimm_init_new_flush_state(SPAPR_NVDIMM(dimm));
686         if (!state) {
687             return H_HARDWARE;
688         }
689 
690         state->drcidx = drc_index;
691 
692         thread_pool_submit_aio(pool, flush_worker_cb, state,
693                                spapr_nvdimm_flush_completion_cb, state);
694 
695         continue_token = state->continue_token;
696     }
697 
698     ret = spapr_nvdimm_get_flush_status(SPAPR_NVDIMM(dimm), continue_token);
699     if (H_IS_LONG_BUSY(ret)) {
700         args[0] = continue_token;
701     }
702 
703     return ret;
704 }
705 
706 static target_ulong h_scm_unbind_mem(PowerPCCPU *cpu, SpaprMachineState *spapr,
707                                      target_ulong opcode, target_ulong *args)
708 {
709     uint32_t drc_index = args[0];
710     uint64_t starting_scm_logical_addr = args[1];
711     uint64_t no_of_scm_blocks_to_unbind = args[2];
712     uint64_t continue_token = args[3];
713     uint64_t size_to_unbind;
714     Range blockrange = range_empty;
715     Range nvdimmrange = range_empty;
716     SpaprDrc *drc = spapr_drc_by_index(drc_index);
717     NVDIMMDevice *nvdimm;
718     uint64_t size, addr;
719 
720     if (!drc || !drc->dev ||
721         spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
722         return H_PARAMETER;
723     }
724 
725     /* continue_token should be zero as this hcall doesn't return H_BUSY. */
726     if (continue_token > 0) {
727         return H_P4;
728     }
729 
730     /* Check if starting_scm_logical_addr is block aligned */
731     if (!QEMU_IS_ALIGNED(starting_scm_logical_addr,
732                          SPAPR_MINIMUM_SCM_BLOCK_SIZE)) {
733         return H_P2;
734     }
735 
736     size_to_unbind = no_of_scm_blocks_to_unbind * SPAPR_MINIMUM_SCM_BLOCK_SIZE;
737     if (no_of_scm_blocks_to_unbind == 0 || no_of_scm_blocks_to_unbind !=
738                                size_to_unbind / SPAPR_MINIMUM_SCM_BLOCK_SIZE) {
739         return H_P3;
740     }
741 
742     nvdimm = NVDIMM(drc->dev);
743     size = object_property_get_int(OBJECT(nvdimm), PC_DIMM_SIZE_PROP,
744                                    &error_abort);
745     addr = object_property_get_int(OBJECT(nvdimm), PC_DIMM_ADDR_PROP,
746                                    &error_abort);
747 
748     range_init_nofail(&nvdimmrange, addr, size);
749     range_init_nofail(&blockrange, starting_scm_logical_addr, size_to_unbind);
750 
751     if (!range_contains_range(&nvdimmrange, &blockrange)) {
752         return H_P3;
753     }
754 
755     args[1] = no_of_scm_blocks_to_unbind;
756 
757     /* let unplug take care of actual unbind */
758     return H_SUCCESS;
759 }
760 
761 #define H_UNBIND_SCOPE_ALL 0x1
762 #define H_UNBIND_SCOPE_DRC 0x2
763 
764 static target_ulong h_scm_unbind_all(PowerPCCPU *cpu, SpaprMachineState *spapr,
765                                      target_ulong opcode, target_ulong *args)
766 {
767     uint64_t target_scope = args[0];
768     uint32_t drc_index = args[1];
769     uint64_t continue_token = args[2];
770     NVDIMMDevice *nvdimm;
771     uint64_t size;
772     uint64_t no_of_scm_blocks_unbound = 0;
773 
774     /* continue_token should be zero as this hcall doesn't return H_BUSY. */
775     if (continue_token > 0) {
776         return H_P4;
777     }
778 
779     if (target_scope == H_UNBIND_SCOPE_DRC) {
780         SpaprDrc *drc = spapr_drc_by_index(drc_index);
781 
782         if (!drc || !drc->dev ||
783             spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
784             return H_P2;
785         }
786 
787         nvdimm = NVDIMM(drc->dev);
788         size = object_property_get_int(OBJECT(nvdimm), PC_DIMM_SIZE_PROP,
789                                        &error_abort);
790 
791         no_of_scm_blocks_unbound = size / SPAPR_MINIMUM_SCM_BLOCK_SIZE;
792     } else if (target_scope ==  H_UNBIND_SCOPE_ALL) {
793         GSList *list, *nvdimms;
794 
795         nvdimms = nvdimm_get_device_list();
796         for (list = nvdimms; list; list = list->next) {
797             nvdimm = list->data;
798             size = object_property_get_int(OBJECT(nvdimm), PC_DIMM_SIZE_PROP,
799                                            &error_abort);
800 
801             no_of_scm_blocks_unbound += size / SPAPR_MINIMUM_SCM_BLOCK_SIZE;
802         }
803         g_slist_free(nvdimms);
804     } else {
805         return H_PARAMETER;
806     }
807 
808     args[1] = no_of_scm_blocks_unbound;
809 
810     /* let unplug take care of actual unbind */
811     return H_SUCCESS;
812 }
813 
814 static target_ulong h_scm_health(PowerPCCPU *cpu, SpaprMachineState *spapr,
815                                  target_ulong opcode, target_ulong *args)
816 {
817 
818     NVDIMMDevice *nvdimm;
819     uint64_t hbitmap = 0;
820     uint32_t drc_index = args[0];
821     SpaprDrc *drc = spapr_drc_by_index(drc_index);
822     const uint64_t hbitmap_mask = PAPR_PMEM_UNARMED;
823 
824 
825     /* Ensure that the drc is valid & is valid PMEM dimm and is plugged in */
826     if (!drc || !drc->dev ||
827         spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
828         return H_PARAMETER;
829     }
830 
831     nvdimm = NVDIMM(drc->dev);
832 
833     /* Update if the nvdimm is unarmed and send its status via health bitmaps */
834     if (object_property_get_bool(OBJECT(nvdimm), NVDIMM_UNARMED_PROP, NULL)) {
835         hbitmap |= PAPR_PMEM_UNARMED;
836     }
837 
838     /* Update the out args with health bitmap/mask */
839     args[0] = hbitmap;
840     args[1] = hbitmap_mask;
841 
842     return H_SUCCESS;
843 }
844 
845 static void spapr_scm_register_types(void)
846 {
847     /* qemu/scm specific hcalls */
848     spapr_register_hypercall(H_SCM_READ_METADATA, h_scm_read_metadata);
849     spapr_register_hypercall(H_SCM_WRITE_METADATA, h_scm_write_metadata);
850     spapr_register_hypercall(H_SCM_BIND_MEM, h_scm_bind_mem);
851     spapr_register_hypercall(H_SCM_UNBIND_MEM, h_scm_unbind_mem);
852     spapr_register_hypercall(H_SCM_UNBIND_ALL, h_scm_unbind_all);
853     spapr_register_hypercall(H_SCM_HEALTH, h_scm_health);
854     spapr_register_hypercall(H_SCM_FLUSH, h_scm_flush);
855 }
856 
857 type_init(spapr_scm_register_types)
858 
859 static void spapr_nvdimm_realize(NVDIMMDevice *dimm, Error **errp)
860 {
861     SpaprNVDIMMDevice *s_nvdimm = SPAPR_NVDIMM(dimm);
862     HostMemoryBackend *backend = MEMORY_BACKEND(PC_DIMM(dimm)->hostmem);
863     bool is_pmem = object_property_get_bool(OBJECT(backend),  "pmem", NULL);
864     bool pmem_override = object_property_get_bool(OBJECT(dimm), "pmem-override",
865                                              NULL);
866     if (!is_pmem || pmem_override) {
867         s_nvdimm->hcall_flush_required = true;
868     }
869 
870     vmstate_register(NULL, VMSTATE_INSTANCE_ID_ANY,
871                      &vmstate_spapr_nvdimm_states, dimm);
872 }
873 
874 static void spapr_nvdimm_unrealize(NVDIMMDevice *dimm)
875 {
876     vmstate_unregister(NULL, &vmstate_spapr_nvdimm_states, dimm);
877 }
878 
879 static Property spapr_nvdimm_properties[] = {
880 #ifdef CONFIG_LIBPMEM
881     DEFINE_PROP_BOOL("pmem-override", SpaprNVDIMMDevice, pmem_override, false),
882 #endif
883     DEFINE_PROP_END_OF_LIST(),
884 };
885 
886 static void spapr_nvdimm_class_init(ObjectClass *oc, void *data)
887 {
888     DeviceClass *dc = DEVICE_CLASS(oc);
889     NVDIMMClass *nvc = NVDIMM_CLASS(oc);
890 
891     nvc->realize = spapr_nvdimm_realize;
892     nvc->unrealize = spapr_nvdimm_unrealize;
893 
894     device_class_set_props(dc, spapr_nvdimm_properties);
895 }
896 
897 static void spapr_nvdimm_init(Object *obj)
898 {
899     SpaprNVDIMMDevice *s_nvdimm = SPAPR_NVDIMM(obj);
900 
901     s_nvdimm->hcall_flush_required = false;
902     QLIST_INIT(&s_nvdimm->pending_nvdimm_flush_states);
903     QLIST_INIT(&s_nvdimm->completed_nvdimm_flush_states);
904 }
905 
906 static TypeInfo spapr_nvdimm_info = {
907     .name          = TYPE_SPAPR_NVDIMM,
908     .parent        = TYPE_NVDIMM,
909     .class_init    = spapr_nvdimm_class_init,
910     .class_size    = sizeof(SPAPRNVDIMMClass),
911     .instance_size = sizeof(SpaprNVDIMMDevice),
912     .instance_init = spapr_nvdimm_init,
913 };
914 
915 static void spapr_nvdimm_register_types(void)
916 {
917     type_register_static(&spapr_nvdimm_info);
918 }
919 
920 type_init(spapr_nvdimm_register_types)
921