[Qemu-devel] [PATCH v6 0/4] NVDIMM ACPI: introduce the framework of QEMU emulated DSM

[Qemu-devel] [PATCH v6 0/4] NVDIMM ACPI: introduce the framework of QEMU emulated DSM

[Xiao Guangrong]

This patchset is against commit f6547bcd9a1b6 (fw-cfg: support writeable
blobs) on pci branch of Michael's git tree and can be found at:
      https://github.com/xiaogr/qemu.git nvdimm-acpi-v6

Michael, two suggestions from you was not followed in this version,
one is using ssdt->buf->len to indicate the payload of SSDT and another
is dropping extra () when we build ARG3. I have replied it in the old
thread to explain why we can not.

Sorry again for i did not spot it when i reply your mail in the first
time.

Changelog in v6:
This version addresses the comments from Michael:
- move the comments belong near field in AcpiNVDIMMState
- improve the comment to explain why HDLE is set to 0 for all the devices
- fix some coding styles and improve other comments
- drop the _CRS as it is not designed for reserved memory

Changelog in v5:
Thanks to Michael's review, the changes in this version are:
- use nvdimm_debug() instead of fprintf() on the path triggered by read
  access of IO port
- introduce the data struct to represent output information to drop the
  open-code
- improve the comment to better explain the fact that currently no function
  other than function 0 is supported

Changelog in v4:
- drop the unnecessary assert() in aml_concatenate() based on Igor's
  suggestion
- introduce build_append_named_dword() and use it to simplify the code as
  Michael's suggestion

Changelog in v3:
Changes addressing Michael's comment:
- rebase the patchset against current code

Changes addressing Igor's comment:
- rename the parameters of aml_create_field() to reflect the ACPI spec
- fix the issue that the @target operand can not be optional in
  aml_concatenate() that is also cleaned up by using build_opcode_2arg_dst()

Others:
- separate the test patches to the single set and will be posted on later 
  
These changes are based on Igor's comments:
- drop ssdt.rev2 support as the memory address allocated by BIOS/OVMF
  are always 32 bits
- support to test NVDIMM tables (NFIT and NVDIMM SSDT)
- add _CRS to report its operation region
- make AML APIs change be the separated patches

This is the second part of vNVDIMM implementation which implements the
BIOS patched dsm memory and introduces the framework that allows QEMU
to emulate DSM method

Thanks to Michael's idea, we do not reserve any memory for NVDIMM ACPI,
instead we let BIOS allocate the memory and patch the address to the
offset we want

IO port is still enabled as it plays as the way to notify QEMU and pass
the patched dsm memory address, so that IO port region, 0x0a18 - 0xa20,
is reserved and it is divided into two 32 bits ports and used to pass
the low 32 bits and high 32 bits of dsm memory address to QEMU

Thanks Igor's idea, this patchset also extends DSDT/SSDT to revision 2
to apply 64 bit operations, in order to keeping compatibility, old
version (<= 2.5) still uses revision 1. Since 64 bit operations breaks
old guests (such as windows XP), we should keep the 64 bits stuff in
the private place where common ACPI operation does not touch it

Xiao Guangrong (4):
  nvdimm acpi: initialize the resource used by NVDIMM ACPI
  nvdimm acpi: introduce patched dsm memory
  nvdimm acpi: let qemu handle _DSM method
  nvdimm acpi: emulate dsm method

 hw/acpi/Makefile.objs   |   2 +-
 hw/acpi/nvdimm.c        | 231 ++++++++++++++++++++++++++++++++++++++++++++++--
 hw/i386/acpi-build.c    |  10 +--
 hw/i386/pc.c            |   6 +-
 hw/i386/pc_piix.c       |   5 ++
 hw/i386/pc_q35.c        |   8 +-
 include/hw/i386/pc.h    |   4 +-
 include/hw/mem/nvdimm.h |  32 ++++++-
 8 files changed, 274 insertions(+), 24 deletions(-)

-- 
1.8.3.1

[Qemu-devel] [PATCH v6 1/4] nvdimm acpi: initialize the resource used by NVDIMM ACPI

[Xiao Guangrong]

32 bits IO port starting from 0x0a18 in guest is reserved for NVDIMM
ACPI emulation. The table, NVDIMM_DSM_MEM_FILE, will be patched into
NVDIMM ACPI binary code

OSPM uses this port to tell QEMU the final address of the DSM memory
and notify QEMU to emulate the DSM method

Signed-off-by: Xiao Guangrong <guangrong.xiao@linux.intel.com>
---
 hw/acpi/Makefile.objs   |  2 +-
 hw/acpi/nvdimm.c        | 35 +++++++++++++++++++++++++++++++++++
 hw/i386/acpi-build.c    | 10 +---------
 hw/i386/pc.c            |  6 +++---
 hw/i386/pc_piix.c       |  5 +++++
 hw/i386/pc_q35.c        |  8 +++++++-
 include/hw/i386/pc.h    |  4 +++-
 include/hw/mem/nvdimm.h | 24 +++++++++++++++++++++++-
 8 files changed, 78 insertions(+), 16 deletions(-)

diff --git a/hw/acpi/Makefile.objs b/hw/acpi/Makefile.objs
index f3ade9a..faee86c 100644
--- a/hw/acpi/Makefile.objs
+++ b/hw/acpi/Makefile.objs
@@ -2,7 +2,7 @@ common-obj-$(CONFIG_ACPI_X86) += core.o piix4.o pcihp.o
 common-obj-$(CONFIG_ACPI_X86_ICH) += ich9.o tco.o
 common-obj-$(CONFIG_ACPI_CPU_HOTPLUG) += cpu_hotplug.o cpu_hotplug_acpi_table.o
 common-obj-$(CONFIG_ACPI_MEMORY_HOTPLUG) += memory_hotplug.o memory_hotplug_acpi_table.o
-common-obj-$(CONFIG_ACPI_NVDIMM) += nvdimm.o
+obj-$(CONFIG_ACPI_NVDIMM) += nvdimm.o
 common-obj-$(CONFIG_ACPI) += acpi_interface.o
 common-obj-$(CONFIG_ACPI) += bios-linker-loader.o
 common-obj-$(CONFIG_ACPI) += aml-build.o
diff --git a/hw/acpi/nvdimm.c b/hw/acpi/nvdimm.c
index 49ee68e..8568b20 100644
--- a/hw/acpi/nvdimm.c
+++ b/hw/acpi/nvdimm.c
@@ -29,6 +29,7 @@
 #include "qemu/osdep.h"
 #include "hw/acpi/acpi.h"
 #include "hw/acpi/aml-build.h"
+#include "hw/nvram/fw_cfg.h"
 #include "hw/mem/nvdimm.h"
 
 static int nvdimm_plugged_device_list(Object *obj, void *opaque)
@@ -370,6 +371,40 @@ static void nvdimm_build_nfit(GSList *device_list, GArray *table_offsets,
     g_array_free(structures, true);
 }
 
+static uint64_t
+nvdimm_dsm_read(void *opaque, hwaddr addr, unsigned size)
+{
+    return 0;
+}
+
+static void
+nvdimm_dsm_write(void *opaque, hwaddr addr, uint64_t val, unsigned size)
+{
+}
+
+static const MemoryRegionOps nvdimm_dsm_ops = {
+    .read = nvdimm_dsm_read,
+    .write = nvdimm_dsm_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+    },
+};
+
+void nvdimm_init_acpi_state(AcpiNVDIMMState *state, MemoryRegion *io,
+                            FWCfgState *fw_cfg, Object *owner)
+{
+    memory_region_init_io(&state->io_mr, owner, &nvdimm_dsm_ops, state,
+                          "nvdimm-acpi-io", NVDIMM_ACPI_IO_LEN);
+    memory_region_add_subregion(io, NVDIMM_ACPI_IO_BASE, &state->io_mr);
+
+    state->dsm_mem = g_array_new(false, true /* clear */, 1);
+    acpi_data_push(state->dsm_mem, TARGET_PAGE_SIZE);
+    fw_cfg_add_file(fw_cfg, NVDIMM_DSM_MEM_FILE, state->dsm_mem->data,
+                    state->dsm_mem->len);
+}
+
 #define NVDIMM_COMMON_DSM      "NCAL"
 
 static void nvdimm_build_common_dsm(Aml *dev)
diff --git a/hw/i386/acpi-build.c b/hw/i386/acpi-build.c
index f26ac9b..738985a 100644
--- a/hw/i386/acpi-build.c
+++ b/hw/i386/acpi-build.c
@@ -39,7 +39,6 @@
 #include "hw/isa/isa.h"
 #include "hw/block/fdc.h"
 #include "hw/acpi/memory_hotplug.h"
-#include "hw/mem/nvdimm.h"
 #include "sysemu/tpm.h"
 #include "hw/acpi/tpm.h"
 #include "sysemu/tpm_backend.h"
@@ -2624,13 +2623,6 @@ static bool acpi_has_iommu(void)
     return intel_iommu && !ambiguous;
 }
 
-static bool acpi_has_nvdimm(void)
-{
-    PCMachineState *pcms = PC_MACHINE(qdev_get_machine());
-
-    return pcms->nvdimm;
-}
-
 static
 void acpi_build(AcpiBuildTables *tables)
 {
@@ -2715,7 +2707,7 @@ void acpi_build(AcpiBuildTables *tables)
         build_dmar_q35(tables_blob, tables->linker);
     }
 
-    if (acpi_has_nvdimm()) {
+    if (pcms->acpi_nvdimm_state.is_enabled) {
         nvdimm_build_acpi(table_offsets, tables_blob, tables->linker);
     }
 
diff --git a/hw/i386/pc.c b/hw/i386/pc.c
index d7fea61..d44490a 100644
--- a/hw/i386/pc.c
+++ b/hw/i386/pc.c
@@ -1854,14 +1854,14 @@ static bool pc_machine_get_nvdimm(Object *obj, Error **errp)
 {
     PCMachineState *pcms = PC_MACHINE(obj);
 
-    return pcms->nvdimm;
+    return pcms->acpi_nvdimm_state.is_enabled;
 }
 
 static void pc_machine_set_nvdimm(Object *obj, bool value, Error **errp)
 {
     PCMachineState *pcms = PC_MACHINE(obj);
 
-    pcms->nvdimm = value;
+    pcms->acpi_nvdimm_state.is_enabled = value;
 }
 
 static void pc_machine_initfn(Object *obj)
@@ -1900,7 +1900,7 @@ static void pc_machine_initfn(Object *obj)
                                     &error_abort);
 
     /* nvdimm is disabled on default. */
-    pcms->nvdimm = false;
+    pcms->acpi_nvdimm_state.is_enabled = false;
     object_property_add_bool(obj, PC_MACHINE_NVDIMM, pc_machine_get_nvdimm,
                              pc_machine_set_nvdimm, &error_abort);
 }
diff --git a/hw/i386/pc_piix.c b/hw/i386/pc_piix.c
index 6f8c2cd..6a69b23 100644
--- a/hw/i386/pc_piix.c
+++ b/hw/i386/pc_piix.c
@@ -274,6 +274,11 @@ static void pc_init1(MachineState *machine,
     if (pcmc->pci_enabled) {
         pc_pci_device_init(pci_bus);
     }
+
+    if (pcms->acpi_nvdimm_state.is_enabled) {
+        nvdimm_init_acpi_state(&pcms->acpi_nvdimm_state, system_io,
+                               pcms->fw_cfg, OBJECT(pcms));
+    }
 }
 
 /* Looking for a pc_compat_2_4() function? It doesn't exist.
diff --git a/hw/i386/pc_q35.c b/hw/i386/pc_q35.c
index 46522c9..17915b0 100644
--- a/hw/i386/pc_q35.c
+++ b/hw/i386/pc_q35.c
@@ -61,6 +61,7 @@ static void pc_q35_init(MachineState *machine)
     PCIDevice *lpc;
     BusState *idebus[MAX_SATA_PORTS];
     ISADevice *rtc_state;
+    MemoryRegion *system_io = get_system_io();
     MemoryRegion *pci_memory;
     MemoryRegion *rom_memory;
     MemoryRegion *ram_memory;
@@ -160,7 +161,7 @@ static void pc_q35_init(MachineState *machine)
     q35_host->mch.ram_memory = ram_memory;
     q35_host->mch.pci_address_space = pci_memory;
     q35_host->mch.system_memory = get_system_memory();
-    q35_host->mch.address_space_io = get_system_io();
+    q35_host->mch.address_space_io = system_io;
     q35_host->mch.below_4g_mem_size = pcms->below_4g_mem_size;
     q35_host->mch.above_4g_mem_size = pcms->above_4g_mem_size;
     /* pci */
@@ -251,6 +252,11 @@ static void pc_q35_init(MachineState *machine)
     if (pcmc->pci_enabled) {
         pc_pci_device_init(host_bus);
     }
+
+    if (pcms->acpi_nvdimm_state.is_enabled) {
+        nvdimm_init_acpi_state(&pcms->acpi_nvdimm_state, system_io,
+                               pcms->fw_cfg, OBJECT(pcms));
+    }
 }
 
 #define DEFINE_Q35_MACHINE(suffix, name, compatfn, optionfn) \
diff --git a/include/hw/i386/pc.h b/include/hw/i386/pc.h
index 472754c..a05feb7 100644
--- a/include/hw/i386/pc.h
+++ b/include/hw/i386/pc.h
@@ -17,6 +17,7 @@
 #include "hw/boards.h"
 #include "hw/compat.h"
 #include "hw/mem/pc-dimm.h"
+#include "hw/mem/nvdimm.h"
 
 #define HPET_INTCAP "hpet-intcap"
 
@@ -57,7 +58,8 @@ struct PCMachineState {
     uint64_t max_ram_below_4g;
     OnOffAuto vmport;
     OnOffAuto smm;
-    bool nvdimm;
+
+    AcpiNVDIMMState acpi_nvdimm_state;
 
     /* RAM information (sizes, addresses, configuration): */
     ram_addr_t below_4g_mem_size, above_4g_mem_size;
diff --git a/include/hw/mem/nvdimm.h b/include/hw/mem/nvdimm.h
index 49183c1..26d5b78 100644
--- a/include/hw/mem/nvdimm.h
+++ b/include/hw/mem/nvdimm.h
@@ -25,8 +25,30 @@
 
 #include "hw/mem/pc-dimm.h"
 
-#define TYPE_NVDIMM      "nvdimm"
+#define TYPE_NVDIMM             "nvdimm"
 
+#define NVDIMM_DSM_MEM_FILE     "etc/acpi/nvdimm-mem"
+
+/*
+ * 32 bits IO port starting from 0x0a18 in guest is reserved for
+ * NVDIMM ACPI emulation.
+ */
+#define NVDIMM_ACPI_IO_BASE     0x0a18
+#define NVDIMM_ACPI_IO_LEN      4
+
+struct AcpiNVDIMMState {
+    /* detect if NVDIMM support is enabled. */
+    bool is_enabled;
+
+    /* the data of the fw_cfg file NVDIMM_DSM_MEM_FILE. */
+    GArray *dsm_mem;
+    /* the IO region used by OSPM to transfer control to QEMU. */
+    MemoryRegion io_mr;
+};
+typedef struct AcpiNVDIMMState AcpiNVDIMMState;
+
+void nvdimm_init_acpi_state(AcpiNVDIMMState *state, MemoryRegion *io,
+                            FWCfgState *fw_cfg, Object *owner);
 void nvdimm_build_acpi(GArray *table_offsets, GArray *table_data,
                        GArray *linker);
 #endif
-- 
1.8.3.1

[Qemu-devel] [PATCH v6 2/4] nvdimm acpi: introduce patched dsm memory

[Xiao Guangrong]

The dsm memory is used to save the input parameters and store
the dsm result which is filled by QEMU.

The address of dsm memory is decided by bios and patched into
int32 object named "MEMA"

Signed-off-by: Xiao Guangrong <guangrong.xiao@linux.intel.com>
---
 hw/acpi/nvdimm.c | 20 +++++++++++++++++---
 1 file changed, 17 insertions(+), 3 deletions(-)

diff --git a/hw/acpi/nvdimm.c b/hw/acpi/nvdimm.c
index 8568b20..90032e5 100644
--- a/hw/acpi/nvdimm.c
+++ b/hw/acpi/nvdimm.c
@@ -29,6 +29,7 @@
 #include "qemu/osdep.h"
 #include "hw/acpi/acpi.h"
 #include "hw/acpi/aml-build.h"
+#include "hw/acpi/bios-linker-loader.h"
 #include "hw/nvram/fw_cfg.h"
 #include "hw/mem/nvdimm.h"
 
@@ -406,6 +407,7 @@ void nvdimm_init_acpi_state(AcpiNVDIMMState *state, MemoryRegion *io,
 }
 
 #define NVDIMM_COMMON_DSM      "NCAL"
+#define NVDIMM_ACPI_MEM_ADDR   "MEMA"
 
 static void nvdimm_build_common_dsm(Aml *dev)
 {
@@ -471,6 +473,7 @@ static void nvdimm_build_ssdt(GSList *device_list, GArray *table_offsets,
                               GArray *table_data, GArray *linker)
 {
     Aml *ssdt, *sb_scope, *dev;
+    int mem_addr_offset, nvdimm_ssdt;
 
     acpi_add_table(table_offsets, table_data);
 
@@ -500,13 +503,24 @@ static void nvdimm_build_ssdt(GSList *device_list, GArray *table_offsets,
     nvdimm_build_nvdimm_devices(device_list, dev);
 
     aml_append(sb_scope, dev);
-
     aml_append(ssdt, sb_scope);
+
+    nvdimm_ssdt = table_data->len;
+
     /* copy AML table into ACPI tables blob and patch header there */
     g_array_append_vals(table_data, ssdt->buf->data, ssdt->buf->len);
+    mem_addr_offset = build_append_named_dword(table_data,
+                                               NVDIMM_ACPI_MEM_ADDR);
+
+    bios_linker_loader_alloc(linker, NVDIMM_DSM_MEM_FILE, TARGET_PAGE_SIZE,
+                             false /* high memory */);
+    bios_linker_loader_add_pointer(linker, ACPI_BUILD_TABLE_FILE,
+                                   NVDIMM_DSM_MEM_FILE, table_data,
+                                   table_data->data + mem_addr_offset,
+                                   sizeof(uint32_t));
     build_header(linker, table_data,
-        (void *)(table_data->data + table_data->len - ssdt->buf->len),
-        "SSDT", ssdt->buf->len, 1, NULL, "NVDIMM");
+        (void *)(table_data->data + nvdimm_ssdt),
+        "SSDT", table_data->len - nvdimm_ssdt, 1, NULL, "NVDIMM");
     free_aml_allocator();
 }
 
-- 
1.8.3.1

[Qemu-devel] [PATCH v6 3/4] nvdimm acpi: let qemu handle _DSM method

[Xiao Guangrong]

If dsm memory is successfully patched, we let qemu fully emulate
the dsm method

This patch saves _DSM input parameters into dsm memory, tell dsm
memory address to QEMU, then fetch the result from the dsm memory

Signed-off-by: Xiao Guangrong <guangrong.xiao@linux.intel.com>
---
 hw/acpi/nvdimm.c | 120 ++++++++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 115 insertions(+), 5 deletions(-)

diff --git a/hw/acpi/nvdimm.c b/hw/acpi/nvdimm.c
index 90032e5..19c2642 100644
--- a/hw/acpi/nvdimm.c
+++ b/hw/acpi/nvdimm.c
@@ -372,6 +372,24 @@ static void nvdimm_build_nfit(GSList *device_list, GArray *table_offsets,
     g_array_free(structures, true);
 }
 
+struct NvdimmDsmIn {
+    uint32_t handle;
+    uint32_t revision;
+    uint32_t function;
+    /* the remaining size in the page is used by arg3. */
+    union {
+        uint8_t arg3[0];
+    };
+} QEMU_PACKED;
+typedef struct NvdimmDsmIn NvdimmDsmIn;
+
+struct NvdimmDsmOut {
+    /* the size of buffer filled by QEMU. */
+    uint32_t len;
+    uint8_t data[0];
+} QEMU_PACKED;
+typedef struct NvdimmDsmOut NvdimmDsmOut;
+
 static uint64_t
 nvdimm_dsm_read(void *opaque, hwaddr addr, unsigned size)
 {
@@ -411,11 +429,18 @@ void nvdimm_init_acpi_state(AcpiNVDIMMState *state, MemoryRegion *io,
 
 static void nvdimm_build_common_dsm(Aml *dev)
 {
-    Aml *method, *ifctx, *function;
+    Aml *method, *ifctx, *function, *dsm_mem, *unpatched, *result_size;
     uint8_t byte_list[1];
 
-    method = aml_method(NVDIMM_COMMON_DSM, 4, AML_NOTSERIALIZED);
+    method = aml_method(NVDIMM_COMMON_DSM, 4, AML_SERIALIZED);
     function = aml_arg(2);
+    dsm_mem = aml_name(NVDIMM_ACPI_MEM_ADDR);
+
+    /*
+     * do not support any method if DSM memory address has not been
+     * patched.
+     */
+    unpatched = aml_if(aml_equal(dsm_mem, aml_int(0x0)));
 
     /*
      * function 0 is called to inquire what functions are supported by
@@ -424,12 +449,38 @@ static void nvdimm_build_common_dsm(Aml *dev)
     ifctx = aml_if(aml_equal(function, aml_int(0)));
     byte_list[0] = 0 /* No function Supported */;
     aml_append(ifctx, aml_return(aml_buffer(1, byte_list)));
-    aml_append(method, ifctx);
+    aml_append(unpatched, ifctx);
 
     /* No function is supported yet. */
     byte_list[0] = 1 /* Not Supported */;
-    aml_append(method, aml_return(aml_buffer(1, byte_list)));
+    aml_append(unpatched, aml_return(aml_buffer(1, byte_list)));
+    aml_append(method, unpatched);
 
+    /*
+     * The HDLE indicates the DSM function is issued from which device,
+     * it is not used at this time as no function is supported yet.
+     * Currently we make it always be 0 for all the devices and will set
+     * the appropriate value once real function is implemented.
+     */
+    aml_append(method, aml_store(aml_int(0x0), aml_name("HDLE")));
+    aml_append(method, aml_store(aml_arg(1), aml_name("REVS")));
+    aml_append(method, aml_store(aml_arg(2), aml_name("FUNC")));
+
+    /*
+     * tell QEMU about the real address of DSM memory, then QEMU
+     * gets the control and fills the result in DSM memory.
+     */
+    aml_append(method, aml_store(dsm_mem, aml_name("NTFI")));
+
+    result_size = aml_local(1);
+    aml_append(method, aml_store(aml_name("RLEN"), result_size));
+    aml_append(method, aml_store(aml_shiftleft(result_size, aml_int(3)),
+                                 result_size));
+    aml_append(method, aml_create_field(aml_name("ODAT"), aml_int(0),
+                                        result_size, "OBUF"));
+    aml_append(method, aml_concatenate(aml_buffer(0, NULL), aml_name("OBUF"),
+                                       aml_arg(6)));
+    aml_append(method, aml_return(aml_arg(6)));
     aml_append(dev, method);
 }
 
@@ -472,7 +523,7 @@ static void nvdimm_build_nvdimm_devices(GSList *device_list, Aml *root_dev)
 static void nvdimm_build_ssdt(GSList *device_list, GArray *table_offsets,
                               GArray *table_data, GArray *linker)
 {
-    Aml *ssdt, *sb_scope, *dev;
+    Aml *ssdt, *sb_scope, *dev, *field;
     int mem_addr_offset, nvdimm_ssdt;
 
     acpi_add_table(table_offsets, table_data);
@@ -497,6 +548,65 @@ static void nvdimm_build_ssdt(GSList *device_list, GArray *table_offsets,
      */
     aml_append(dev, aml_name_decl("_HID", aml_string("ACPI0012")));
 
+    /* map DSM memory and IO into ACPI namespace. */
+    aml_append(dev, aml_operation_region("NPIO", AML_SYSTEM_IO,
+               aml_int(NVDIMM_ACPI_IO_BASE), NVDIMM_ACPI_IO_LEN));
+    aml_append(dev, aml_operation_region("NRAM", AML_SYSTEM_MEMORY,
+               aml_name(NVDIMM_ACPI_MEM_ADDR), TARGET_PAGE_SIZE));
+
+    /*
+     * DSM notifier:
+     * NTFI: write the address of DSM memory and notify QEMU to emulate
+     *       the access.
+     *
+     * It is the IO port so that accessing them will cause VM-exit, the
+     * control will be transferred to QEMU.
+     */
+    field = aml_field("NPIO", AML_DWORD_ACC, AML_NOLOCK, AML_PRESERVE);
+    aml_append(field, aml_named_field("NTFI",
+               sizeof(uint32_t) * BITS_PER_BYTE));
+    aml_append(dev, field);
+
+    /*
+     * DSM input:
+     * HDLE: store device's handle, it's zero if the _DSM call happens
+     *       on NVDIMM Root Device.
+     * REVS: store the Arg1 of _DSM call.
+     * FUNC: store the Arg2 of _DSM call.
+     * ARG3: store the Arg3 of _DSM call.
+     *
+     * They are RAM mapping on host so that these accesses never cause
+     * VM-EXIT.
+     */
+    field = aml_field("NRAM", AML_DWORD_ACC, AML_NOLOCK, AML_PRESERVE);
+    aml_append(field, aml_named_field("HDLE",
+               sizeof(typeof_field(NvdimmDsmIn, handle)) * BITS_PER_BYTE));
+    aml_append(field, aml_named_field("REVS",
+               sizeof(typeof_field(NvdimmDsmIn, revision)) * BITS_PER_BYTE));
+    aml_append(field, aml_named_field("FUNC",
+               sizeof(typeof_field(NvdimmDsmIn, function)) * BITS_PER_BYTE));
+    aml_append(field, aml_named_field("ARG3",
+               (TARGET_PAGE_SIZE - offsetof(NvdimmDsmIn, arg3)) *
+                BITS_PER_BYTE));
+    aml_append(dev, field);
+
+    /*
+     * DSM output:
+     * RLEN: the size of the buffer filled by QEMU.
+     * ODAT: the buffer QEMU uses to store the result.
+     *
+     * Since the page is reused by both input and out, the input data
+     * will be lost after storing new result into ODAT so we should fetch
+     * all the input data before writing the result.
+     */
+    field = aml_field("NRAM", AML_DWORD_ACC, AML_NOLOCK, AML_PRESERVE);
+    aml_append(field, aml_named_field("RLEN",
+               sizeof(typeof_field(NvdimmDsmOut, len)) * BITS_PER_BYTE));
+    aml_append(field, aml_named_field("ODAT",
+               (TARGET_PAGE_SIZE - offsetof(NvdimmDsmOut, data)) *
+                     BITS_PER_BYTE));
+    aml_append(dev, field);
+
     nvdimm_build_common_dsm(dev);
     nvdimm_build_device_dsm(dev);
 
-- 
1.8.3.1

[Qemu-devel] [PATCH v6 4/4] nvdimm acpi: emulate dsm method

[Xiao Guangrong]

Emulate dsm method after IO VM-exit

Currently, we only introduce the framework and no function is actually
supported

Signed-off-by: Xiao Guangrong <guangrong.xiao@linux.intel.com>
---
 hw/acpi/nvdimm.c        | 56 +++++++++++++++++++++++++++++++++++++++++++++++++
 include/hw/mem/nvdimm.h |  8 +++++++
 2 files changed, 64 insertions(+)

diff --git a/hw/acpi/nvdimm.c b/hw/acpi/nvdimm.c
index 19c2642..d9048c0 100644
--- a/hw/acpi/nvdimm.c
+++ b/hw/acpi/nvdimm.c
@@ -390,15 +390,71 @@ struct NvdimmDsmOut {
 } QEMU_PACKED;
 typedef struct NvdimmDsmOut NvdimmDsmOut;
 
+struct NvdimmDsmFunc0Out {
+    /* the size of buffer filled by QEMU. */
+     uint32_t len;
+     uint32_t supported_func;
+} QEMU_PACKED;
+typedef struct NvdimmDsmFunc0Out NvdimmDsmFunc0Out;
+
+struct NvdimmDsmFuncNoPayloadOut {
+    /* the size of buffer filled by QEMU. */
+     uint32_t len;
+     uint32_t func_ret_status;
+} QEMU_PACKED;
+typedef struct NvdimmDsmFuncNoPayloadOut NvdimmDsmFuncNoPayloadOut;
+
 static uint64_t
 nvdimm_dsm_read(void *opaque, hwaddr addr, unsigned size)
 {
+    nvdimm_debug("BUG: we never read _DSM IO Port.\n");
     return 0;
 }
 
 static void
 nvdimm_dsm_write(void *opaque, hwaddr addr, uint64_t val, unsigned size)
 {
+    NvdimmDsmIn *in;
+    hwaddr dsm_mem_addr = val;
+
+    nvdimm_debug("dsm memory address %#lx.\n", dsm_mem_addr);
+
+    /*
+     * The DSM memory is mapped to guest address space so an evil guest
+     * can change its content while we are doing DSM emulation. Avoid
+     * this by copying DSM memory to QEMU local memory.
+     */
+    in = g_malloc(TARGET_PAGE_SIZE);
+    cpu_physical_memory_read(dsm_mem_addr, in, TARGET_PAGE_SIZE);
+
+    le32_to_cpus(&in->revision);
+    le32_to_cpus(&in->function);
+    le32_to_cpus(&in->handle);
+
+    nvdimm_debug("Revision %#x Handler %#x Function %#x.\n", in->revision,
+                 in->handle, in->function);
+
+    /*
+     * function 0 is called to inquire which functions are supported by
+     * OSPM
+     */
+    if (in->function == 0) {
+        NvdimmDsmFunc0Out func0 = {
+            .len = cpu_to_le32(sizeof(func0)),
+             /* No function supported other than function 0 */
+            .supported_func = cpu_to_le32(0),
+        };
+        cpu_physical_memory_write(dsm_mem_addr, &func0, sizeof func0);
+    } else {
+        /* No function except function 0 is supported yet. */
+        NvdimmDsmFuncNoPayloadOut out = {
+            .len = cpu_to_le32(sizeof(out)),
+            .func_ret_status = cpu_to_le32(1)  /* Not Supported */,
+        };
+        cpu_physical_memory_write(dsm_mem_addr, &out, sizeof(out));
+    }
+
+    g_free(in);
 }
 
 static const MemoryRegionOps nvdimm_dsm_ops = {
diff --git a/include/hw/mem/nvdimm.h b/include/hw/mem/nvdimm.h
index 26d5b78..517de9c 100644
--- a/include/hw/mem/nvdimm.h
+++ b/include/hw/mem/nvdimm.h
@@ -25,6 +25,14 @@
 
 #include "hw/mem/pc-dimm.h"
 
+#define NVDIMM_DEBUG 0
+#define nvdimm_debug(fmt, ...)                                \
+    do {                                                      \
+        if (NVDIMM_DEBUG) {                                   \
+            fprintf(stderr, "nvdimm: " fmt, ## __VA_ARGS__);  \
+        }                                                     \
+    } while (0)
+
 #define TYPE_NVDIMM             "nvdimm"
 
 #define NVDIMM_DSM_MEM_FILE     "etc/acpi/nvdimm-mem"
-- 
1.8.3.1