xref: /openbmc/qemu/hw/ppc/spapr_nvdimm.c (revision 07b2c8e0)
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         nvdimm->readonly) {
325         return H_P2;
326     }
327 
328     switch (len) {
329     case 1:
330         if (data & 0xffffffffffffff00) {
331             return H_P2;
332         }
333         stb_p(buf, data);
334         break;
335     case 2:
336         if (data & 0xffffffffffff0000) {
337             return H_P2;
338         }
339         stw_be_p(buf, data);
340         break;
341     case 4:
342         if (data & 0xffffffff00000000) {
343             return H_P2;
344         }
345         stl_be_p(buf, data);
346         break;
347     case 8:
348         stq_be_p(buf, data);
349         break;
350     default:
351             g_assert_not_reached();
352     }
353 
354     ddc = NVDIMM_GET_CLASS(nvdimm);
355     ddc->write_label_data(nvdimm, buf, len, offset);
356 
357     return H_SUCCESS;
358 }
359 
360 static target_ulong h_scm_bind_mem(PowerPCCPU *cpu, SpaprMachineState *spapr,
361                                    target_ulong opcode, target_ulong *args)
362 {
363     uint32_t drc_index = args[0];
364     uint64_t starting_idx = args[1];
365     uint64_t no_of_scm_blocks_to_bind = args[2];
366     uint64_t target_logical_mem_addr = args[3];
367     uint64_t continue_token = args[4];
368     uint64_t size;
369     uint64_t total_no_of_scm_blocks;
370     SpaprDrc *drc = spapr_drc_by_index(drc_index);
371     hwaddr addr;
372     NVDIMMDevice *nvdimm;
373 
374     if (!drc || !drc->dev ||
375         spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
376         return H_PARAMETER;
377     }
378 
379     /*
380      * Currently continue token should be zero qemu has already bound
381      * everything and this hcall doesn't return H_BUSY.
382      */
383     if (continue_token > 0) {
384         return H_P5;
385     }
386 
387     /* Currently qemu assigns the address. */
388     if (target_logical_mem_addr != 0xffffffffffffffff) {
389         return H_OVERLAP;
390     }
391 
392     nvdimm = NVDIMM(drc->dev);
393 
394     size = object_property_get_uint(OBJECT(nvdimm),
395                                     PC_DIMM_SIZE_PROP, &error_abort);
396 
397     total_no_of_scm_blocks = size / SPAPR_MINIMUM_SCM_BLOCK_SIZE;
398 
399     if (starting_idx > total_no_of_scm_blocks) {
400         return H_P2;
401     }
402 
403     if (((starting_idx + no_of_scm_blocks_to_bind) < starting_idx) ||
404         ((starting_idx + no_of_scm_blocks_to_bind) > total_no_of_scm_blocks)) {
405         return H_P3;
406     }
407 
408     addr = object_property_get_uint(OBJECT(nvdimm),
409                                     PC_DIMM_ADDR_PROP, &error_abort);
410 
411     addr += starting_idx * SPAPR_MINIMUM_SCM_BLOCK_SIZE;
412 
413     /* Already bound, Return target logical address in R5 */
414     args[1] = addr;
415     args[2] = no_of_scm_blocks_to_bind;
416 
417     return H_SUCCESS;
418 }
419 
420 typedef struct SpaprNVDIMMDeviceFlushState {
421     uint64_t continue_token;
422     int64_t hcall_ret;
423     uint32_t drcidx;
424 
425     QLIST_ENTRY(SpaprNVDIMMDeviceFlushState) node;
426 } SpaprNVDIMMDeviceFlushState;
427 
428 typedef struct SpaprNVDIMMDevice SpaprNVDIMMDevice;
429 struct SpaprNVDIMMDevice {
430     /* private */
431     NVDIMMDevice parent_obj;
432 
433     bool hcall_flush_required;
434     uint64_t nvdimm_flush_token;
435     QLIST_HEAD(, SpaprNVDIMMDeviceFlushState) pending_nvdimm_flush_states;
436     QLIST_HEAD(, SpaprNVDIMMDeviceFlushState) completed_nvdimm_flush_states;
437 
438     /* public */
439 
440     /*
441      * The 'on' value for this property forced the qemu to enable the hcall
442      * flush for the nvdimm device even if the backend is a pmem
443      */
444     bool pmem_override;
445 };
446 
447 static int flush_worker_cb(void *opaque)
448 {
449     SpaprNVDIMMDeviceFlushState *state = opaque;
450     SpaprDrc *drc = spapr_drc_by_index(state->drcidx);
451     PCDIMMDevice *dimm;
452     HostMemoryBackend *backend;
453     int backend_fd;
454 
455     g_assert(drc != NULL);
456 
457     dimm = PC_DIMM(drc->dev);
458     backend = MEMORY_BACKEND(dimm->hostmem);
459     backend_fd = memory_region_get_fd(&backend->mr);
460 
461     if (object_property_get_bool(OBJECT(backend), "pmem", NULL)) {
462         MemoryRegion *mr = host_memory_backend_get_memory(dimm->hostmem);
463         void *ptr = memory_region_get_ram_ptr(mr);
464         size_t size = object_property_get_uint(OBJECT(dimm), PC_DIMM_SIZE_PROP,
465                                                NULL);
466 
467         /* flush pmem backend */
468         pmem_persist(ptr, size);
469     } else {
470         /* flush raw backing image */
471         if (qemu_fdatasync(backend_fd) < 0) {
472             error_report("papr_scm: Could not sync nvdimm to backend file: %s",
473                          strerror(errno));
474             return H_HARDWARE;
475         }
476     }
477 
478     return H_SUCCESS;
479 }
480 
481 static void spapr_nvdimm_flush_completion_cb(void *opaque, int hcall_ret)
482 {
483     SpaprNVDIMMDeviceFlushState *state = opaque;
484     SpaprDrc *drc = spapr_drc_by_index(state->drcidx);
485     SpaprNVDIMMDevice *s_nvdimm;
486 
487     g_assert(drc != NULL);
488 
489     s_nvdimm = SPAPR_NVDIMM(drc->dev);
490 
491     state->hcall_ret = hcall_ret;
492     QLIST_REMOVE(state, node);
493     QLIST_INSERT_HEAD(&s_nvdimm->completed_nvdimm_flush_states, state, node);
494 }
495 
496 static int spapr_nvdimm_flush_post_load(void *opaque, int version_id)
497 {
498     SpaprNVDIMMDevice *s_nvdimm = (SpaprNVDIMMDevice *)opaque;
499     SpaprNVDIMMDeviceFlushState *state;
500     HostMemoryBackend *backend = MEMORY_BACKEND(PC_DIMM(s_nvdimm)->hostmem);
501     bool is_pmem = object_property_get_bool(OBJECT(backend), "pmem", NULL);
502     bool pmem_override = object_property_get_bool(OBJECT(s_nvdimm),
503                                                   "pmem-override", NULL);
504     bool dest_hcall_flush_required = pmem_override || !is_pmem;
505 
506     if (!s_nvdimm->hcall_flush_required && dest_hcall_flush_required) {
507         error_report("The file backend for the spapr-nvdimm device %s at "
508                      "source is a pmem, use pmem=on and pmem-override=off to "
509                      "continue.", DEVICE(s_nvdimm)->id);
510         return -EINVAL;
511     }
512     if (s_nvdimm->hcall_flush_required && !dest_hcall_flush_required) {
513         error_report("The guest expects hcall-flush support for the "
514                      "spapr-nvdimm device %s, use pmem_override=on to "
515                      "continue.", DEVICE(s_nvdimm)->id);
516         return -EINVAL;
517     }
518 
519     QLIST_FOREACH(state, &s_nvdimm->pending_nvdimm_flush_states, node) {
520         thread_pool_submit_aio(flush_worker_cb, state,
521                                spapr_nvdimm_flush_completion_cb, state);
522     }
523 
524     return 0;
525 }
526 
527 static const VMStateDescription vmstate_spapr_nvdimm_flush_state = {
528      .name = "spapr_nvdimm_flush_state",
529      .version_id = 1,
530      .minimum_version_id = 1,
531      .fields = (const VMStateField[]) {
532          VMSTATE_UINT64(continue_token, SpaprNVDIMMDeviceFlushState),
533          VMSTATE_INT64(hcall_ret, SpaprNVDIMMDeviceFlushState),
534          VMSTATE_UINT32(drcidx, SpaprNVDIMMDeviceFlushState),
535          VMSTATE_END_OF_LIST()
536      },
537 };
538 
539 const VMStateDescription vmstate_spapr_nvdimm_states = {
540     .name = "spapr_nvdimm_states",
541     .version_id = 1,
542     .minimum_version_id = 1,
543     .post_load = spapr_nvdimm_flush_post_load,
544     .fields = (const VMStateField[]) {
545         VMSTATE_BOOL(hcall_flush_required, SpaprNVDIMMDevice),
546         VMSTATE_UINT64(nvdimm_flush_token, SpaprNVDIMMDevice),
547         VMSTATE_QLIST_V(completed_nvdimm_flush_states, SpaprNVDIMMDevice, 1,
548                         vmstate_spapr_nvdimm_flush_state,
549                         SpaprNVDIMMDeviceFlushState, node),
550         VMSTATE_QLIST_V(pending_nvdimm_flush_states, SpaprNVDIMMDevice, 1,
551                         vmstate_spapr_nvdimm_flush_state,
552                         SpaprNVDIMMDeviceFlushState, node),
553         VMSTATE_END_OF_LIST()
554     },
555 };
556 
557 /*
558  * Assign a token and reserve it for the new flush state.
559  */
560 static SpaprNVDIMMDeviceFlushState *spapr_nvdimm_init_new_flush_state(
561                                                 SpaprNVDIMMDevice *spapr_nvdimm)
562 {
563     SpaprNVDIMMDeviceFlushState *state;
564 
565     state = g_malloc0(sizeof(*state));
566 
567     spapr_nvdimm->nvdimm_flush_token++;
568     /* Token zero is presumed as no job pending. Assert on overflow to zero */
569     g_assert(spapr_nvdimm->nvdimm_flush_token != 0);
570 
571     state->continue_token = spapr_nvdimm->nvdimm_flush_token;
572 
573     QLIST_INSERT_HEAD(&spapr_nvdimm->pending_nvdimm_flush_states, state, node);
574 
575     return state;
576 }
577 
578 /*
579  * spapr_nvdimm_finish_flushes
580  *      Waits for all pending flush requests to complete
581  *      their execution and free the states
582  */
583 void spapr_nvdimm_finish_flushes(void)
584 {
585     SpaprNVDIMMDeviceFlushState *state, *next;
586     GSList *list, *nvdimms;
587 
588     /*
589      * Called on reset path, the main loop thread which calls
590      * the pending BHs has gotten out running in the reset path,
591      * finally reaching here. Other code path being guest
592      * h_client_architecture_support, that's early boot up.
593      */
594     nvdimms = nvdimm_get_device_list();
595     for (list = nvdimms; list; list = list->next) {
596         NVDIMMDevice *nvdimm = list->data;
597         if (object_dynamic_cast(OBJECT(nvdimm), TYPE_SPAPR_NVDIMM)) {
598             SpaprNVDIMMDevice *s_nvdimm = SPAPR_NVDIMM(nvdimm);
599             while (!QLIST_EMPTY(&s_nvdimm->pending_nvdimm_flush_states)) {
600                 aio_poll(qemu_get_aio_context(), true);
601             }
602 
603             QLIST_FOREACH_SAFE(state, &s_nvdimm->completed_nvdimm_flush_states,
604                                node, next) {
605                 QLIST_REMOVE(state, node);
606                 g_free(state);
607             }
608         }
609     }
610     g_slist_free(nvdimms);
611 }
612 
613 /*
614  * spapr_nvdimm_get_flush_status
615  *      Fetches the status of the hcall worker and returns
616  *      H_LONG_BUSY_ORDER_10_MSEC if the worker is still running.
617  */
618 static int spapr_nvdimm_get_flush_status(SpaprNVDIMMDevice *s_nvdimm,
619                                          uint64_t token)
620 {
621     SpaprNVDIMMDeviceFlushState *state, *node;
622 
623     QLIST_FOREACH(state, &s_nvdimm->pending_nvdimm_flush_states, node) {
624         if (state->continue_token == token) {
625             return H_LONG_BUSY_ORDER_10_MSEC;
626         }
627     }
628 
629     QLIST_FOREACH_SAFE(state, &s_nvdimm->completed_nvdimm_flush_states,
630                        node, node) {
631         if (state->continue_token == token) {
632             int ret = state->hcall_ret;
633             QLIST_REMOVE(state, node);
634             g_free(state);
635             return ret;
636         }
637     }
638 
639     /* If not found in complete list too, invalid token */
640     return H_P2;
641 }
642 
643 /*
644  * H_SCM_FLUSH
645  * Input: drc_index, continue-token
646  * Out: continue-token
647  * Return Value: H_SUCCESS, H_Parameter, H_P2, H_LONG_BUSY_ORDER_10_MSEC,
648  *               H_UNSUPPORTED
649  *
650  * Given a DRC Index Flush the data to backend NVDIMM device. The hcall returns
651  * H_LONG_BUSY_ORDER_10_MSEC when the flush takes longer time and the hcall
652  * needs to be issued multiple times in order to be completely serviced. The
653  * continue-token from the output to be passed in the argument list of
654  * subsequent hcalls until the hcall is completely serviced at which point
655  * H_SUCCESS or other error is returned.
656  */
657 static target_ulong h_scm_flush(PowerPCCPU *cpu, SpaprMachineState *spapr,
658                                 target_ulong opcode, target_ulong *args)
659 {
660     int ret;
661     uint32_t drc_index = args[0];
662     uint64_t continue_token = args[1];
663     SpaprDrc *drc = spapr_drc_by_index(drc_index);
664     PCDIMMDevice *dimm;
665     HostMemoryBackend *backend = NULL;
666     SpaprNVDIMMDeviceFlushState *state;
667     int fd;
668 
669     if (!drc || !drc->dev ||
670         spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
671         return H_PARAMETER;
672     }
673 
674     dimm = PC_DIMM(drc->dev);
675     if (!object_dynamic_cast(OBJECT(dimm), TYPE_SPAPR_NVDIMM)) {
676         return H_PARAMETER;
677     }
678     if (continue_token == 0) {
679         bool is_pmem = false, pmem_override = false;
680         backend = MEMORY_BACKEND(dimm->hostmem);
681         fd = memory_region_get_fd(&backend->mr);
682 
683         if (fd < 0) {
684             return H_UNSUPPORTED;
685         }
686 
687         is_pmem = object_property_get_bool(OBJECT(backend), "pmem", NULL);
688         pmem_override = object_property_get_bool(OBJECT(dimm),
689                                                 "pmem-override", NULL);
690         if (is_pmem && !pmem_override) {
691             return H_UNSUPPORTED;
692         }
693 
694         state = spapr_nvdimm_init_new_flush_state(SPAPR_NVDIMM(dimm));
695         if (!state) {
696             return H_HARDWARE;
697         }
698 
699         state->drcidx = drc_index;
700 
701         thread_pool_submit_aio(flush_worker_cb, state,
702                                spapr_nvdimm_flush_completion_cb, state);
703 
704         continue_token = state->continue_token;
705     }
706 
707     ret = spapr_nvdimm_get_flush_status(SPAPR_NVDIMM(dimm), continue_token);
708     if (H_IS_LONG_BUSY(ret)) {
709         args[0] = continue_token;
710     }
711 
712     return ret;
713 }
714 
715 static target_ulong h_scm_unbind_mem(PowerPCCPU *cpu, SpaprMachineState *spapr,
716                                      target_ulong opcode, target_ulong *args)
717 {
718     uint32_t drc_index = args[0];
719     uint64_t starting_scm_logical_addr = args[1];
720     uint64_t no_of_scm_blocks_to_unbind = args[2];
721     uint64_t continue_token = args[3];
722     uint64_t size_to_unbind;
723     Range blockrange = range_empty;
724     Range nvdimmrange = range_empty;
725     SpaprDrc *drc = spapr_drc_by_index(drc_index);
726     NVDIMMDevice *nvdimm;
727     uint64_t size, addr;
728 
729     if (!drc || !drc->dev ||
730         spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
731         return H_PARAMETER;
732     }
733 
734     /* continue_token should be zero as this hcall doesn't return H_BUSY. */
735     if (continue_token > 0) {
736         return H_P4;
737     }
738 
739     /* Check if starting_scm_logical_addr is block aligned */
740     if (!QEMU_IS_ALIGNED(starting_scm_logical_addr,
741                          SPAPR_MINIMUM_SCM_BLOCK_SIZE)) {
742         return H_P2;
743     }
744 
745     size_to_unbind = no_of_scm_blocks_to_unbind * SPAPR_MINIMUM_SCM_BLOCK_SIZE;
746     if (no_of_scm_blocks_to_unbind == 0 || no_of_scm_blocks_to_unbind !=
747                                size_to_unbind / SPAPR_MINIMUM_SCM_BLOCK_SIZE) {
748         return H_P3;
749     }
750 
751     nvdimm = NVDIMM(drc->dev);
752     size = object_property_get_int(OBJECT(nvdimm), PC_DIMM_SIZE_PROP,
753                                    &error_abort);
754     addr = object_property_get_int(OBJECT(nvdimm), PC_DIMM_ADDR_PROP,
755                                    &error_abort);
756 
757     range_init_nofail(&nvdimmrange, addr, size);
758     range_init_nofail(&blockrange, starting_scm_logical_addr, size_to_unbind);
759 
760     if (!range_contains_range(&nvdimmrange, &blockrange)) {
761         return H_P3;
762     }
763 
764     args[1] = no_of_scm_blocks_to_unbind;
765 
766     /* let unplug take care of actual unbind */
767     return H_SUCCESS;
768 }
769 
770 #define H_UNBIND_SCOPE_ALL 0x1
771 #define H_UNBIND_SCOPE_DRC 0x2
772 
773 static target_ulong h_scm_unbind_all(PowerPCCPU *cpu, SpaprMachineState *spapr,
774                                      target_ulong opcode, target_ulong *args)
775 {
776     uint64_t target_scope = args[0];
777     uint32_t drc_index = args[1];
778     uint64_t continue_token = args[2];
779     NVDIMMDevice *nvdimm;
780     uint64_t size;
781     uint64_t no_of_scm_blocks_unbound = 0;
782 
783     /* continue_token should be zero as this hcall doesn't return H_BUSY. */
784     if (continue_token > 0) {
785         return H_P4;
786     }
787 
788     if (target_scope == H_UNBIND_SCOPE_DRC) {
789         SpaprDrc *drc = spapr_drc_by_index(drc_index);
790 
791         if (!drc || !drc->dev ||
792             spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
793             return H_P2;
794         }
795 
796         nvdimm = NVDIMM(drc->dev);
797         size = object_property_get_int(OBJECT(nvdimm), PC_DIMM_SIZE_PROP,
798                                        &error_abort);
799 
800         no_of_scm_blocks_unbound = size / SPAPR_MINIMUM_SCM_BLOCK_SIZE;
801     } else if (target_scope ==  H_UNBIND_SCOPE_ALL) {
802         GSList *list, *nvdimms;
803 
804         nvdimms = nvdimm_get_device_list();
805         for (list = nvdimms; list; list = list->next) {
806             nvdimm = list->data;
807             size = object_property_get_int(OBJECT(nvdimm), PC_DIMM_SIZE_PROP,
808                                            &error_abort);
809 
810             no_of_scm_blocks_unbound += size / SPAPR_MINIMUM_SCM_BLOCK_SIZE;
811         }
812         g_slist_free(nvdimms);
813     } else {
814         return H_PARAMETER;
815     }
816 
817     args[1] = no_of_scm_blocks_unbound;
818 
819     /* let unplug take care of actual unbind */
820     return H_SUCCESS;
821 }
822 
823 static target_ulong h_scm_health(PowerPCCPU *cpu, SpaprMachineState *spapr,
824                                  target_ulong opcode, target_ulong *args)
825 {
826 
827     NVDIMMDevice *nvdimm;
828     uint64_t hbitmap = 0;
829     uint32_t drc_index = args[0];
830     SpaprDrc *drc = spapr_drc_by_index(drc_index);
831     const uint64_t hbitmap_mask = PAPR_PMEM_UNARMED;
832 
833 
834     /* Ensure that the drc is valid & is valid PMEM dimm and is plugged in */
835     if (!drc || !drc->dev ||
836         spapr_drc_type(drc) != SPAPR_DR_CONNECTOR_TYPE_PMEM) {
837         return H_PARAMETER;
838     }
839 
840     nvdimm = NVDIMM(drc->dev);
841 
842     /* Update if the nvdimm is unarmed and send its status via health bitmaps */
843     if (object_property_get_bool(OBJECT(nvdimm), NVDIMM_UNARMED_PROP, NULL)) {
844         hbitmap |= PAPR_PMEM_UNARMED;
845     }
846 
847     /* Update the out args with health bitmap/mask */
848     args[0] = hbitmap;
849     args[1] = hbitmap_mask;
850 
851     return H_SUCCESS;
852 }
853 
854 static void spapr_scm_register_types(void)
855 {
856     /* qemu/scm specific hcalls */
857     spapr_register_hypercall(H_SCM_READ_METADATA, h_scm_read_metadata);
858     spapr_register_hypercall(H_SCM_WRITE_METADATA, h_scm_write_metadata);
859     spapr_register_hypercall(H_SCM_BIND_MEM, h_scm_bind_mem);
860     spapr_register_hypercall(H_SCM_UNBIND_MEM, h_scm_unbind_mem);
861     spapr_register_hypercall(H_SCM_UNBIND_ALL, h_scm_unbind_all);
862     spapr_register_hypercall(H_SCM_HEALTH, h_scm_health);
863     spapr_register_hypercall(H_SCM_FLUSH, h_scm_flush);
864 }
865 
866 type_init(spapr_scm_register_types)
867 
868 static void spapr_nvdimm_realize(NVDIMMDevice *dimm, Error **errp)
869 {
870     SpaprNVDIMMDevice *s_nvdimm = SPAPR_NVDIMM(dimm);
871     HostMemoryBackend *backend = MEMORY_BACKEND(PC_DIMM(dimm)->hostmem);
872     bool is_pmem = object_property_get_bool(OBJECT(backend),  "pmem", NULL);
873     bool pmem_override = object_property_get_bool(OBJECT(dimm), "pmem-override",
874                                              NULL);
875     if (!is_pmem || pmem_override) {
876         s_nvdimm->hcall_flush_required = true;
877     }
878 
879     vmstate_register_any(NULL, &vmstate_spapr_nvdimm_states, dimm);
880 }
881 
882 static void spapr_nvdimm_unrealize(NVDIMMDevice *dimm)
883 {
884     vmstate_unregister(NULL, &vmstate_spapr_nvdimm_states, dimm);
885 }
886 
887 static Property spapr_nvdimm_properties[] = {
888 #ifdef CONFIG_LIBPMEM
889     DEFINE_PROP_BOOL("pmem-override", SpaprNVDIMMDevice, pmem_override, false),
890 #endif
891     DEFINE_PROP_END_OF_LIST(),
892 };
893 
894 static void spapr_nvdimm_class_init(ObjectClass *oc, void *data)
895 {
896     DeviceClass *dc = DEVICE_CLASS(oc);
897     NVDIMMClass *nvc = NVDIMM_CLASS(oc);
898 
899     nvc->realize = spapr_nvdimm_realize;
900     nvc->unrealize = spapr_nvdimm_unrealize;
901 
902     device_class_set_props(dc, spapr_nvdimm_properties);
903 }
904 
905 static void spapr_nvdimm_init(Object *obj)
906 {
907     SpaprNVDIMMDevice *s_nvdimm = SPAPR_NVDIMM(obj);
908 
909     s_nvdimm->hcall_flush_required = false;
910     QLIST_INIT(&s_nvdimm->pending_nvdimm_flush_states);
911     QLIST_INIT(&s_nvdimm->completed_nvdimm_flush_states);
912 }
913 
914 static TypeInfo spapr_nvdimm_info = {
915     .name          = TYPE_SPAPR_NVDIMM,
916     .parent        = TYPE_NVDIMM,
917     .class_init    = spapr_nvdimm_class_init,
918     .class_size    = sizeof(SPAPRNVDIMMClass),
919     .instance_size = sizeof(SpaprNVDIMMDevice),
920     .instance_init = spapr_nvdimm_init,
921 };
922 
923 static void spapr_nvdimm_register_types(void)
924 {
925     type_register_static(&spapr_nvdimm_info);
926 }
927 
928 type_init(spapr_nvdimm_register_types)
929