Skip to contentSkip to conversation

generate dynamic _CRS for motherboard resources

RFC In review Igor Mammedov 4 participants 30 messages Feb 7, 2014 – Feb 19, 2014

lore
Share & tools
lore
#0 lore raw

cover generate dynamic _CRS for motherboard resources

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

So replace static hole punching with dynamically consumed resources in a child device on PCI0 bus. i.e generate PNP0C02 device as a child of PCI0 bus at runtime and consume GPE0, PCI/CPU hotplug IO resources in it instead of punching holes in static PCI0._CRS.

Tested with Windows XPsp3, Vista, Windows Server 2003, 2008, 2012r2.

PS:
Series adds several ASL like macros to simplify code for dynamic generation of AML structures.

Igor Mammedov (9):
Revert "pc: Q35 DSDT: exclude CPU hotplug IO range from PCI bus
resources"
Revert "pc: PIIX DSDT: exclude CPU/PCI hotplug & GPE0 IO range from
PCI bus resources"
Partial revert "pc: ACPI: expose PRST IO range via _CRS"
acpi: replace opencoded opcodes with defines
acpi: add PNP0C02 to PCI0 bus
acpi: consume GPE0 IO resources in PNP0C02 device
acpi: consume CPU hotplug IO resource in PNP0C02 device
pcihp: expose PCI hotplug MMIO base/length as properties of piix4pm
acpi: consume PCIHP IO resource in PNP0C02 device

 hw/acpi/pcihp.c                   |   28 ++++++
 hw/acpi/piix4.c                   |    1 +
 hw/i386/acpi-build.c              |  177 ++++++++++++++++++++++++++++++++++--
 hw/i386/acpi-dsdt-cpu-hotplug.dsl |   11 ---
 hw/i386/acpi-dsdt-pci-crs.dsl     |   15 +++-
 hw/i386/acpi-dsdt.dsl             |   39 --------
 hw/i386/q35-acpi-dsdt.dsl         |   16 ----
 include/hw/acpi/pcihp.h           |    4 +
 8 files changed, 214 insertions(+), 77 deletions(-)
#1 lore raw

RFC 1/9 Revert "pc: Q35 DSDT: exclude CPU hotplug IO range from PCI bus resources"

This reverts commit 562e56a9f8e627b2a4ef60037507361ce3cb4e6d.

---
 hw/i386/acpi-dsdt-pci-crs.dsl |    7 +++++++
 hw/i386/acpi-dsdt.dsl         |    7 -------
 hw/i386/q35-acpi-dsdt.dsl     |    8 --------
 3 files changed, 7 insertions(+), 15 deletions(-)
hw/i386/acpi-dsdt-pci-crs.dsl +7 -0
index 4648e90..8b631d1 100644
@@ -30,6 +30,13 @@ Scope(\_SB.PCI0) {
             0x01,               // Address Alignment
             0x08,               // Address Length
             )
+        WordIO(ResourceProducer, MinFixed, MaxFixed, PosDecode, EntireRange,
+            0x0000,             // Address Space Granularity
+            0x0000,             // Address Range Minimum
+            0x0CF7,             // Address Range Maximum
+            0x0000,             // Address Translation Offset
+            0x0CF8,             // Address Length
+            ,, , TypeStatic)
         BOARD_SPECIFIC_PCI_RESOURSES
         DWordMemory(ResourceProducer, PosDecode, MinFixed, MaxFixed, Cacheable, ReadWrite,
             0x00000000,         // Address Space Granularity
hw/i386/acpi-dsdt.dsl +0 -7
index b23d5e0..5546b4f 100644
@@ -38,13 +38,6 @@ DefinitionBlock (
 #define BOARD_SPECIFIC_PCI_RESOURSES \
      WordIO(ResourceProducer, MinFixed, MaxFixed, PosDecode, EntireRange, \
          0x0000, \
-         0x0000, \
-         0x0CF7, \
-         0x0000, \
-         0x0CF8, \
-         ,, , TypeStatic) \
-     WordIO(ResourceProducer, MinFixed, MaxFixed, PosDecode, EntireRange, \
-         0x0000, \
          0x0D00, \
          0xADFF, \
          0x0000, \
hw/i386/q35-acpi-dsdt.dsl +0 -8
index d618e9e..cf6c03d 100644
@@ -51,14 +51,6 @@ DefinitionBlock (
 #define BOARD_SPECIFIC_PCI_RESOURSES \
      WordIO(ResourceProducer, MinFixed, MaxFixed, PosDecode, EntireRange, \
          0x0000, \
-         0x0000, \
-         0x0CD7, \
-         0x0000, \
-         0x0CD8, \
-         ,, , TypeStatic) \
-     /* 0xcd8-0xcf7 hole for CPU hotplug, hw/acpi/ich9.c:ICH9_PROC_BASE */ \
-     WordIO(ResourceProducer, MinFixed, MaxFixed, PosDecode, EntireRange, \
-         0x0000, \
          0x0D00, \
          0xFFFF, \
          0x0000, \
Signature · 1 line
1.7.1
#2 lore raw

RFC 2/9 Revert "pc: PIIX DSDT: exclude CPU/PCI hotplug & GPE0 IO range from PCI bus resources"

In reply to Igor Mammedov

This reverts commit 1aa149b479a479323121251f1e8e676765cb354d.

---
 hw/i386/acpi-dsdt-pci-crs.dsl |    8 +++++++-
 hw/i386/acpi-dsdt.dsl         |   32 --------------------------------
 hw/i386/q35-acpi-dsdt.dsl     |    8 --------
 3 files changed, 7 insertions(+), 41 deletions(-)
hw/i386/acpi-dsdt-pci-crs.dsl +7 -1
index 8b631d1..b375a19 100644
@@ -37,7 +37,13 @@ Scope(\_SB.PCI0) {
             0x0000,             // Address Translation Offset
             0x0CF8,             // Address Length
             ,, , TypeStatic)
-        BOARD_SPECIFIC_PCI_RESOURSES
+        WordIO(ResourceProducer, MinFixed, MaxFixed, PosDecode, EntireRange,
+            0x0000,             // Address Space Granularity
+            0x0D00,             // Address Range Minimum
+            0xFFFF,             // Address Range Maximum
+            0x0000,             // Address Translation Offset
+            0xF300,             // Address Length
+            ,, , TypeStatic)
         DWordMemory(ResourceProducer, PosDecode, MinFixed, MaxFixed, Cacheable, ReadWrite,
             0x00000000,         // Address Space Granularity
             0x000A0000,         // Address Range Minimum
hw/i386/acpi-dsdt.dsl +0 -32
index 5546b4f..28ce1c6 100644
@@ -35,38 +35,6 @@ DefinitionBlock (
 /****************************************************************
  * PCI Bus definition
  ****************************************************************/
-#define BOARD_SPECIFIC_PCI_RESOURSES \
-     WordIO(ResourceProducer, MinFixed, MaxFixed, PosDecode, EntireRange, \
-         0x0000, \
-         0x0D00, \
-         0xADFF, \
-         0x0000, \
-         0xA100, \
-         ,, , TypeStatic) \
-     /* 0xae00-0xae0e hole for PCI hotplug, hw/acpi/piix4.c:PCI_HOTPLUG_ADDR */ \
-     WordIO(ResourceProducer, MinFixed, MaxFixed, PosDecode, EntireRange, \
-         0x0000, \
-         0xAE0F, \
-         0xAEFF, \
-         0x0000, \
-         0x00F1, \
-         ,, , TypeStatic) \
-     /* 0xaf00-0xaf1f hole for CPU hotplug, hw/acpi/piix4.c:PIIX4_PROC_BASE */ \
-     WordIO(ResourceProducer, MinFixed, MaxFixed, PosDecode, EntireRange, \
-         0x0000, \
-         0xAF20, \
-         0xAFDF, \
-         0x0000, \
-         0x00C0, \
-         ,, , TypeStatic) \
-     /* 0xafe0-0xafe3 hole for ACPI.GPE0, hw/acpi/piix4.c:GPE_BASE */ \
-     WordIO(ResourceProducer, MinFixed, MaxFixed, PosDecode, EntireRange, \
-         0x0000, \
-         0xAFE4, \
-         0xFFFF, \
-         0x0000, \
-         0x501C, \
-         ,, , TypeStatic)
 
     Scope(\_SB) {
         Device(PCI0) {
hw/i386/q35-acpi-dsdt.dsl +0 -8
index cf6c03d..596d752 100644
@@ -48,14 +48,6 @@ DefinitionBlock (
 /****************************************************************
  * PCI Bus definition
  ****************************************************************/
-#define BOARD_SPECIFIC_PCI_RESOURSES \
-     WordIO(ResourceProducer, MinFixed, MaxFixed, PosDecode, EntireRange, \
-         0x0000, \
-         0x0D00, \
-         0xFFFF, \
-         0x0000, \
-         0xF300, \
-         ,, , TypeStatic)
 
     Scope(\_SB) {
         Device(PCI0) {
Signature · 1 line
1.7.1
#3 lore raw

RFC 3/9 Partial revert "pc: ACPI: expose PRST IO range via _CRS"

In reply to Igor Mammedov

This reverts part of commit 61a3f63560ccd2b5e8c9134e9213a1cff36f26bf that adds Device(CPU_HOTPLUG_RESOURCE_DEVICE) with CPU hotplug IO resource in its _CRS.

Conflicts:

	hw/i386/acpi-dsdt-cpu-hotplug.dsl
---
 hw/i386/acpi-dsdt-cpu-hotplug.dsl |   11 -----------
 1 files changed, 0 insertions(+), 11 deletions(-)
hw/i386/acpi-dsdt-cpu-hotplug.dsl +0 -11
index dee4843..3acfe1b 100644
@@ -16,7 +16,6 @@
 /****************************************************************
  * CPU hotplug
  ****************************************************************/
-#define CPU_HOTPLUG_RESOURCE_DEVICE PRES
 
 Scope(\_SB) {
     /* Objects filled in by run-time generated SSDT */
@@ -91,14 +90,4 @@ Scope(\_SB) {
             Increment(Local0)
         }
     }
-
-    Device(CPU_HOTPLUG_RESOURCE_DEVICE) {
-        Name(_HID, "ACPI0004")
-
-        Name(_CRS, ResourceTemplate() {
-            IO(Decode16, CPU_STATUS_BASE, CPU_STATUS_BASE, 0, CPU_STATUS_LEN)
-        })
-
-        Name(_STA, 0xB) /* present, functioning, decoding, not shown in UI */
-    }
 }
Signature · 1 line
1.7.1
#4 lore raw

RFC 4/9 acpi: replace opencoded opcodes with defines

In reply to Igor Mammedov
---
 hw/i386/acpi-build.c |   28 ++++++++++++++++++----------
 1 files changed, 18 insertions(+), 10 deletions(-)
hw/i386/acpi-build.c +18 -10
index 6a43a7d..1dbe5ce 100644
@@ -224,6 +224,14 @@ static void acpi_get_pci_info(PcPciInfo *info)
 #define ACPI_BUILD_TABLE_FILE "etc/acpi/tables"
 #define ACPI_BUILD_RSDP_FILE "etc/acpi/rsdp"
 
+#define BytePrefix         0x0A
+#define WordPrefix         0x0B
+#define DWordPrefix        0x0C
+
+#define NameOp             0x08
+#define ScopeOp            0x10
+#define DeviceOp           0x82
+
 static void
 build_header(GArray *linker, GArray *table_data,
              AcpiTableHeader *h, uint32_t sig, int len, uint8_t rev)
@@ -364,13 +372,13 @@ static void build_append_value(GArray *table, uint32_t value, int size)
 
     switch (size) {
     case 1:
-        prefix = 0x0A; /* BytePrefix */
+        prefix = BytePrefix;
         break;
     case 2:
-        prefix = 0x0B; /* WordPrefix */
+        prefix = WordPrefix;
         break;
     case 4:
-        prefix = 0x0C; /* DWordPrefix */
+        prefix = DWordPrefix;
         break;
     default:
         assert(0);
@@ -762,24 +770,24 @@ static void build_pci_bus_end(PCIBus *bus, void *bus_state)
     bool bus_hotplug_support = false;
 
     if (bus->parent_dev) {
-        op = 0x82; /* DeviceOp */
+        op = DeviceOp;
         build_append_nameseg(bus_table, "S%.02X_",
                              bus->parent_dev->devfn);
-        build_append_byte(bus_table, 0x08); /* NameOp */
+        build_append_byte(bus_table, NameOp);
         build_append_nameseg(bus_table, "_SUN");
         build_append_value(bus_table, PCI_SLOT(bus->parent_dev->devfn), 1);
-        build_append_byte(bus_table, 0x08); /* NameOp */
+        build_append_byte(bus_table, NameOp);
         build_append_nameseg(bus_table, "_ADR");
         build_append_value(bus_table, (PCI_SLOT(bus->parent_dev->devfn) << 16) |
                            PCI_FUNC(bus->parent_dev->devfn), 4);
     } else {
-        op = 0x10; /* ScopeOp */;
+        op = ScopeOp;
         build_append_nameseg(bus_table, "PCI0");
     }
 
     bsel = object_property_get_qobject(OBJECT(bus), ACPI_PCIHP_PROP_BSEL, NULL);
     if (bsel) {
-        build_append_byte(bus_table, 0x08); /* NameOp */
+        build_append_byte(bus_table, NameOp);
         build_append_nameseg(bus_table, "BSEL");
         build_append_int(bus_table, qint_get_int(qobject_to_qint(bsel)));
     }
@@ -962,7 +970,7 @@ build_ssdt(GArray *table_data, GArray *linker,
 
     {
         GArray *sb_scope = build_alloc_array();
-        uint8_t op = 0x10; /* ScopeOp */
+        uint8_t op = ScopeOp;
 
         build_append_nameseg(sb_scope, "_SB_");
 
@@ -983,7 +991,7 @@ build_ssdt(GArray *table_data, GArray *linker,
         build_append_notify_method(sb_scope, "NTFY", "CP%0.02X", acpi_cpus);
 
         /* build "Name(CPON, Package() { One, One, ..., Zero, Zero, ... })" */
-        build_append_byte(sb_scope, 0x08); /* NameOp */
+        build_append_byte(sb_scope, NameOp);
         build_append_nameseg(sb_scope, "CPON");
 
         {
Signature · 1 line
1.7.1
#5 lore raw

Re: [Qemu-devel] [RFC 4/9] acpi: replace opencoded opcodes with defines

Quoted 1 line · Igor Mammedov

The reason I avoided doing this is that this conflicts with qemu coding style which only uses camel case for types.

So as a minimum this needs a comment explaining that we are using the names from ACPI spec as-is, that's why we deviate from the coding style, to simplify matching against that.

Something like below:

hw/i386/acpi-build.c static void acpi_get_pci_info(PcPciInfo *info) · line 225

Quoted 11 lines
---
 hw/i386/acpi-build.c |   28 ++++++++++++++++++----------
 1 files changed, 18 insertions(+), 10 deletions(-)
hw/i386/acpi-build.c +0 -0
index 6a43a7d..1dbe5ce 100644
@@ -224,6 +224,14 @@ static void acpi_get_pci_info(PcPciInfo *info)
 #define ACPI_BUILD_TABLE_FILE "etc/acpi/tables"
 #define ACPI_BUILD_RSDP_FILE "etc/acpi/rsdp"

/* Constants from ACPI spec 5.0a:
* ACPI Machine Language (AML) Specification */

We probably should add in spec link as well.

hw/i386/acpi-build.c static void acpi_get_pci_info(PcPciInfo *info) · line 229

Quoted 3 lines
hw/i386/acpi-build.c +3 -0
+#define BytePrefix         0x0A
+#define WordPrefix         0x0B
+#define DWordPrefix        0x0C

Not sure about these ones.
There's a single user, and naming is different from rest of operators which makes it
a bit confusing.
Maybe define near the user?

hw/i386/acpi-build.c static void acpi_get_pci_info(PcPciInfo *info) · line 233

Quoted 4 lines
hw/i386/acpi-build.c +4 -0
+
+#define NameOp             0x08
+#define ScopeOp            0x10
+#define DeviceOp           0x82

Hmm if we are doing this let's do this for all Ops.

Quoted 71 lines
hw/i386/acpi-build.c +11 -10
+
 static void
 build_header(GArray *linker, GArray *table_data,
              AcpiTableHeader *h, uint32_t sig, int len, uint8_t rev)
@@ -364,13 +372,13 @@ static void build_append_value(GArray *table, uint32_t value, int size)
 
     switch (size) {
     case 1:
-        prefix = 0x0A; /* BytePrefix */
+        prefix = BytePrefix;
         break;
     case 2:
-        prefix = 0x0B; /* WordPrefix */
+        prefix = WordPrefix;
         break;
     case 4:
-        prefix = 0x0C; /* DWordPrefix */
+        prefix = DWordPrefix;
         break;
     default:
         assert(0);
@@ -762,24 +770,24 @@ static void build_pci_bus_end(PCIBus *bus, void *bus_state)
     bool bus_hotplug_support = false;
 
     if (bus->parent_dev) {
-        op = 0x82; /* DeviceOp */
+        op = DeviceOp;
         build_append_nameseg(bus_table, "S%.02X_",
                              bus->parent_dev->devfn);
-        build_append_byte(bus_table, 0x08); /* NameOp */
+        build_append_byte(bus_table, NameOp);
         build_append_nameseg(bus_table, "_SUN");
         build_append_value(bus_table, PCI_SLOT(bus->parent_dev->devfn), 1);
-        build_append_byte(bus_table, 0x08); /* NameOp */
+        build_append_byte(bus_table, NameOp);
         build_append_nameseg(bus_table, "_ADR");
         build_append_value(bus_table, (PCI_SLOT(bus->parent_dev->devfn) << 16) |
                            PCI_FUNC(bus->parent_dev->devfn), 4);
     } else {
-        op = 0x10; /* ScopeOp */;
+        op = ScopeOp;
         build_append_nameseg(bus_table, "PCI0");
     }
 
     bsel = object_property_get_qobject(OBJECT(bus), ACPI_PCIHP_PROP_BSEL, NULL);
     if (bsel) {
-        build_append_byte(bus_table, 0x08); /* NameOp */
+        build_append_byte(bus_table, NameOp);
         build_append_nameseg(bus_table, "BSEL");
         build_append_int(bus_table, qint_get_int(qobject_to_qint(bsel)));
     }
@@ -962,7 +970,7 @@ build_ssdt(GArray *table_data, GArray *linker,
 
     {
         GArray *sb_scope = build_alloc_array();
-        uint8_t op = 0x10; /* ScopeOp */
+        uint8_t op = ScopeOp;
 
         build_append_nameseg(sb_scope, "_SB_");
 
@@ -983,7 +991,7 @@ build_ssdt(GArray *table_data, GArray *linker,
         build_append_notify_method(sb_scope, "NTFY", "CP%0.02X", acpi_cpus);
 
         /* build "Name(CPON, Package() { One, One, ..., Zero, Zero, ... })" */
-        build_append_byte(sb_scope, 0x08); /* NameOp */
+        build_append_byte(sb_scope, NameOp);
         build_append_nameseg(sb_scope, "CPON");
 
         {
Signature · 1 line
1.7.1
#6 lore raw

RFC 5/9 acpi: add PNP0C02 to PCI0 bus

In reply to Igor Mammedov
---
 hw/i386/acpi-build.c |   55 ++++++++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 55 insertions(+), 0 deletions(-)
hw/i386/acpi-build.c +55 -0
index 1dbe5ce..f0bedbd 100644
@@ -492,6 +492,55 @@ static inline void acpi_add_table(GArray *table_offsets, GArray *table_data)
     g_array_append_val(table_offsets, offset);
 }
 
+static uint8_t Hex2Digit(char c)
+{
+    if (c >= 'A') {
+        return c - 'A' + 10;
+    }
+    return c - '0';
+}
+
+static uint32_t encodeEisaId(const char *str)
+{
+    uint32_t ret;
+    g_assert(strlen(str) == 7);
+    ret = (str[0] - 0x40) << 26 |
+          (str[1] - 0x40) << 21 |
+          (str[2] - 0x40) << 16 |
+          Hex2Digit(str[3]) << 12 |
+          Hex2Digit(str[4]) << 8 |
+          Hex2Digit(str[5]) << 4 |
+          Hex2Digit(str[6]);
+    return bswap32(ret);
+}
+
+#define ACPI_SCOPE(ctx, name, ...) {\
+    GArray *name = build_alloc_array(); \
+    build_append_nameseg(name, stringify(name)); \
+    __VA_ARGS__; \
+    build_package(name, ScopeOp, 0); \
+    build_append_array(ctx, name); \
+    build_free_array(name); \
+}
+
+#define ACPI_NAME(ctx, name) { \
+    build_append_byte(ctx, NameOp); \
+    build_append_nameseg(ctx, name); \
+}
+
+#define ACPI_EISAID(ctx, val) { \
+    build_append_value(ctx, encodeEisaId(val), sizeof(uint32_t)); \
+}
+
+#define ACPI_DEVICE(ctx, name, ...) {\
+    GArray *name = build_alloc_array(); \
+    build_append_nameseg(name, stringify(name)); \
+    __VA_ARGS__; \
+    build_extop_package(name, DeviceOp); \
+    build_append_array(ctx, name); \
+    build_free_array(name); \
+}
+
 /* FACS */
 static void
 build_facs(GArray *table_data, GArray *linker, PcGuestInfo *guest_info)
@@ -1032,6 +1081,12 @@ build_ssdt(GArray *table_data, GArray *linker,
             build_pci_bus_state_cleanup(&hotplug_state);
         }
 
+        ACPI_SCOPE(sb_scope, PCI0,
+            ACPI_DEVICE(PCI0, MRES,
+                ACPI_NAME(MRES, "_HID"); ACPI_EISAID(MRES, "PNP0C02");
+            );
+        );
+
         build_package(sb_scope, op, 3);
         build_append_array(table_data, sb_scope);
         build_free_array(sb_scope);
Signature · 1 line
1.7.1
#7 lore raw

Re: [Qemu-devel] [RFC 5/9] acpi: add PNP0C02 to PCI0 bus

Quoted 25 lines · Igor Mammedov
---
 hw/i386/acpi-build.c |   55 ++++++++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 55 insertions(+), 0 deletions(-)
hw/i386/acpi-build.c +12 -0
index 1dbe5ce..f0bedbd 100644
@@ -492,6 +492,55 @@ static inline void acpi_add_table(GArray *table_offsets, GArray *table_data)
     g_array_append_val(table_offsets, offset);
 }
 
+static uint8_t Hex2Digit(char c)
+{
+    if (c >= 'A') {
+        return c - 'A' + 10;
+    }
+    return c - '0';
+}
+
+static uint32_t encodeEisaId(const char *str)
+{
+    uint32_t ret;
+    g_assert(strlen(str) == 7);

hw/i386/acpi-build.c static inline void acpi_add_table(GArray *table_offsets, GArray *table_data) · line 516

Quoted 10 lines
hw/i386/acpi-build.c +10 -0
+    ret = (str[0] - 0x40) << 26 |
+          (str[1] - 0x40) << 21 |
+          (str[2] - 0x40) << 16 |
+          Hex2Digit(str[3]) << 12 |
+          Hex2Digit(str[4]) << 8 |
+          Hex2Digit(str[5]) << 4 |
+          Hex2Digit(str[6]);
+    return bswap32(ret);
+}
+

Why the camel case here?
Seems uncalled for ...

Quoted 31 lines
hw/i386/acpi-build.c +27 -0
+#define ACPI_SCOPE(ctx, name, ...) {\
+    GArray *name = build_alloc_array(); \
+    build_append_nameseg(name, stringify(name)); \
+    __VA_ARGS__; \
+    build_package(name, ScopeOp, 0); \
+    build_append_array(ctx, name); \
+    build_free_array(name); \
+}
+
+#define ACPI_NAME(ctx, name) { \
+    build_append_byte(ctx, NameOp); \
+    build_append_nameseg(ctx, name); \
+}
+
+#define ACPI_EISAID(ctx, val) { \
+    build_append_value(ctx, encodeEisaId(val), sizeof(uint32_t)); \
+}
+
+#define ACPI_DEVICE(ctx, name, ...) {\
+    GArray *name = build_alloc_array(); \
+    build_append_nameseg(name, stringify(name)); \
+    __VA_ARGS__; \
+    build_extop_package(name, DeviceOp); \
+    build_append_array(ctx, name); \
+    build_free_array(name); \
+}
+
 /* FACS */
 static void
 build_facs(GArray *table_data, GArray *linker, PcGuestInfo *guest_info)

hw/i386/acpi-build.c build_ssdt(GArray *table_data, GArray *linker, · line 1089

Quoted 9 lines
hw/i386/acpi-build.c +6 -0
@@ -1032,6 +1081,12 @@ build_ssdt(GArray *table_data, GArray *linker,
             build_pci_bus_state_cleanup(&hotplug_state);
         }
 
+        ACPI_SCOPE(sb_scope, PCI0,
+            ACPI_DEVICE(PCI0, MRES,
+                ACPI_NAME(MRES, "_HID"); ACPI_EISAID(MRES, "PNP0C02");
+            );
+        );
+

Wow you managed to make C look like ACPI, including 4-character identifier limitation. This needs some thought.

Quoted 5 lines
hw/i386/acpi-build.c +0 -0
         build_package(sb_scope, op, 3);
         build_append_array(table_data, sb_scope);
         build_free_array(sb_scope);
Signature · 1 line
1.7.1
#8 lore raw

RFC 6/9 acpi: consume GPE0 IO resources in PNP0C02 device

In reply to Igor Mammedov
---
 hw/i386/acpi-build.c |   62 ++++++++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 62 insertions(+), 0 deletions(-)
hw/i386/acpi-build.c +62 -0
index f0bedbd..ce5f715 100644
@@ -230,8 +230,13 @@ static void acpi_get_pci_info(PcPciInfo *info)
 
 #define NameOp             0x08
 #define ScopeOp            0x10
+#define BufferOp           0x11
 #define DeviceOp           0x82
 
+#define EndTag             0x79
+#define Decode16           0x1
+#define Decode10           0x0
+
 static void
 build_header(GArray *linker, GArray *table_data,
              AcpiTableHeader *h, uint32_t sig, int len, uint8_t rev)
@@ -406,6 +411,25 @@ static void build_append_int(GArray *table, uint32_t value)
     }
 }
 
+static void build_prepend_int(GArray *array, uint32_t value)
+{
+    GArray *data = build_alloc_array();
+
+    build_append_int(data, value);
+    g_array_prepend_vals(array, data->data, data->len);
+    build_free_array(data);
+}
+
+static void build_buffer(GArray *package, unsigned BufferSize)
+{
+    uint32_t len = package->len > BufferSize ? package->len : BufferSize;
+
+    /* TODO: buffer padding if BufferSize > actual buffer length */
+    build_prepend_int(package, len);
+    build_prepend_package_length(package, 0);
+    build_prepend_byte(package, BufferOp);
+}
+
 static GArray *build_alloc_method(const char *name, uint8_t arg_count)
 {
     GArray *method = build_alloc_array();
@@ -523,6 +547,14 @@ static uint32_t encodeEisaId(const char *str)
     build_free_array(name); \
 }
 
+#define ACPI_BUFFER(ctx, name, min_size, ...) { \
+    GArray *name = build_alloc_array(); \
+    __VA_ARGS__; \
+    build_buffer(name, min_size); \
+    build_append_array(ctx, name); \
+    build_free_array(name); \
+}
+
 #define ACPI_NAME(ctx, name) { \
     build_append_byte(ctx, NameOp); \
     build_append_nameseg(ctx, name); \
@@ -541,6 +573,29 @@ static uint32_t encodeEisaId(const char *str)
     build_free_array(name); \
 }
 
+#define ACPI_ENDTAG(ctx) { \
+    build_append_byte(ctx, EndTag); \
+    build_append_byte(ctx, 0); \
+}
+
+#define ACPI_RESOURCE_TEMPLATE(ctx, name, ...) { \
+    ACPI_BUFFER(ctx, name, 0, \
+        __VA_ARGS__; \
+        ACPI_ENDTAG(name); \
+    ) \
+}
+
+#define ACPI_IO(ctx, _DEC, _MIN_BASE, _MAX_BASE, _ALN, _LEN) { \
+    build_append_byte(ctx, 0x47 /* IO port descriptor */); \
+    build_append_byte(ctx, _DEC); \
+    build_append_byte(ctx, _MIN_BASE  & 0xff); \
+    build_append_byte(ctx, (_MIN_BASE >> 8) & 0xff); \
+    build_append_byte(ctx, _MAX_BASE  & 0xff); \
+    build_append_byte(ctx, (_MAX_BASE >> 8) & 0xff); \
+    build_append_byte(ctx, _ALN); \
+    build_append_byte(ctx, _LEN); \
+}
+
 /* FACS */
 static void
 build_facs(GArray *table_data, GArray *linker, PcGuestInfo *guest_info)
@@ -1084,6 +1139,13 @@ build_ssdt(GArray *table_data, GArray *linker,
         ACPI_SCOPE(sb_scope, PCI0,
             ACPI_DEVICE(PCI0, MRES,
                 ACPI_NAME(MRES, "_HID"); ACPI_EISAID(MRES, "PNP0C02");
+                ACPI_NAME(MRES, "_CRS"); ACPI_RESOURCE_TEMPLATE(MRES, RESBUF,
+                    ACPI_IO(RESBUF, Decode16,
+                            pm->gpe0_blk,      /* _MIN */
+                            pm->gpe0_blk,      /* _MAX */
+                            0x0,               /* _ALN */
+                            pm->gpe0_blk_len); /* _LEN */
+                );
             );
         );
 
Signature · 1 line
1.7.1
#9 lore raw

Re: [Qemu-devel] [RFC 6/9] acpi: consume GPE0 IO resources in PNP0C02 device

Quoted 10 lines · Igor Mammedov
---
 hw/i386/acpi-build.c |   62 ++++++++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 62 insertions(+), 0 deletions(-)
hw/i386/acpi-build.c +0 -0
index f0bedbd..ce5f715 100644

hw/i386/acpi-build.c static void acpi_get_pci_info(PcPciInfo *info) · line 236

Quoted 7 lines
hw/i386/acpi-build.c +2 -0
@@ -230,8 +230,13 @@ static void acpi_get_pci_info(PcPciInfo *info)
 
 #define NameOp             0x08
 #define ScopeOp            0x10
+#define BufferOp           0x11
 #define DeviceOp           0x82
 
+#define EndTag             0x79

I would say we should use the values from Table 6-162 Small Resource Items.
Wrap them in a function to get the full resource.

hw/i386/acpi-build.c static void acpi_get_pci_info(PcPciInfo *info) · line 239

Quoted 3 lines
hw/i386/acpi-build.c +3 -0
+#define Decode16           0x1
+#define Decode10           0x0
+

This is the name from ASL, it's really _DEC value.

Quoted 11 lines
hw/i386/acpi-build.c +4 -0
 static void
 build_header(GArray *linker, GArray *table_data,
              AcpiTableHeader *h, uint32_t sig, int len, uint8_t rev)
@@ -406,6 +411,25 @@ static void build_append_int(GArray *table, uint32_t value)
     }
 }
 
+static void build_prepend_int(GArray *array, uint32_t value)
+{
+    GArray *data = build_alloc_array();
+

hw/i386/acpi-build.c static void build_append_int(GArray *table, uint32_t value) · line 427

Quoted 10 lines
hw/i386/acpi-build.c +10 -0
+    build_append_int(data, value);
+    g_array_prepend_vals(array, data->data, data->len);
+    build_free_array(data);
+}
+
+static void build_buffer(GArray *package, unsigned BufferSize)
+{
+    uint32_t len = package->len > BufferSize ? package->len : BufferSize;
+
+    /* TODO: buffer padding if BufferSize > actual buffer length */

Not sure what this means.
So assert here?
Or just make it work ...

hw/i386/acpi-build.c static void build_append_int(GArray *table, uint32_t value) · line 430

Quoted 3 lines
hw/i386/acpi-build.c +3 -0
+    build_prepend_int(package, len);
+    build_prepend_package_length(package, 0);
+    build_prepend_byte(package, BufferOp);

prepend is confusing.
Just do it like we do for methods:

        build_append_and_cleanup_buffer(template, buffer);

hw/i386/acpi-build.c static uint32_t encodeEisaId(const char *str) · line 550

Quoted 10 lines
hw/i386/acpi-build.c +3 -0
+}
+
 static GArray *build_alloc_method(const char *name, uint8_t arg_count)
 {
     GArray *method = build_alloc_array();
@@ -523,6 +547,14 @@ static uint32_t encodeEisaId(const char *str)
     build_free_array(name); \
 }
 
+#define ACPI_BUFFER(ctx, name, min_size, ...) { \

Why pass in min_size?
the only reason we have it in existing code was I wanted ACPI to be bit for bit compatible with what seabios generated.

We can drop minsize everywhere ...

Quoted 11 lines
hw/i386/acpi-build.c +7 -0
+    GArray *name = build_alloc_array(); \
+    __VA_ARGS__; \
+    build_buffer(name, min_size); \
+    build_append_array(ctx, name); \
+    build_free_array(name); \
+}
+
 #define ACPI_NAME(ctx, name) { \
     build_append_byte(ctx, NameOp); \
     build_append_nameseg(ctx, name); \

hw/i386/acpi-build.c static uint32_t encodeEisaId(const char *str) · line 578

Quoted 6 lines
hw/i386/acpi-build.c +3 -0
@@ -541,6 +573,29 @@ static uint32_t encodeEisaId(const char *str)
     build_free_array(name); \
 }
 
+#define ACPI_ENDTAG(ctx) { \
+    build_append_byte(ctx, EndTag); \
+    build_append_byte(ctx, 0); \

Confused.
what's going on with the checksum here? What fills it in?
why don't we add in the correct byte straight away?

hw/i386/acpi-build.c static uint32_t encodeEisaId(const char *str) · line 584

Quoted 6 lines
hw/i386/acpi-build.c +6 -0
+}
+
+#define ACPI_RESOURCE_TEMPLATE(ctx, name, ...) { \
+    ACPI_BUFFER(ctx, name, 0, \
+        __VA_ARGS__; \
+        ACPI_ENDTAG(name); \

Ugh.
Not worth the ugliness in my opinion. Just add end tag explicitly.

hw/i386/acpi-build.c static uint32_t encodeEisaId(const char *str) · line 588

Quoted 4 lines
hw/i386/acpi-build.c +4 -0
+    ) \
+}
+
+#define ACPI_IO(ctx, _DEC, _MIN_BASE, _MAX_BASE, _ALN, _LEN) { \

C spec says
— All identifiers that begin with an underscore and either an uppercase letter or another
underscore are always reserved for any use.
— All identifiers that begin with an underscore are always reserved for use as identifiers
so we try to avoid these.

Quoted 17 lines
hw/i386/acpi-build.c +10 -0
+    build_append_byte(ctx, 0x47 /* IO port descriptor */); \
+    build_append_byte(ctx, _DEC); \
+    build_append_byte(ctx, _MIN_BASE  & 0xff); \
+    build_append_byte(ctx, (_MIN_BASE >> 8) & 0xff); \
+    build_append_byte(ctx, _MAX_BASE  & 0xff); \
+    build_append_byte(ctx, (_MAX_BASE >> 8) & 0xff); \
+    build_append_byte(ctx, _ALN); \
+    build_append_byte(ctx, _LEN); \
+}
+
 /* FACS */
 static void
 build_facs(GArray *table_data, GArray *linker, PcGuestInfo *guest_info)
@@ -1084,6 +1139,13 @@ build_ssdt(GArray *table_data, GArray *linker,
         ACPI_SCOPE(sb_scope, PCI0,
             ACPI_DEVICE(PCI0, MRES,
                 ACPI_NAME(MRES, "_HID"); ACPI_EISAID(MRES, "PNP0C02");

hw/i386/acpi-build.c build_ssdt(GArray *table_data, GArray *linker, · line 1151

Quoted 10 lines
hw/i386/acpi-build.c +7 -0
+                ACPI_NAME(MRES, "_CRS"); ACPI_RESOURCE_TEMPLATE(MRES, RESBUF,
+                    ACPI_IO(RESBUF, Decode16,
 
+                            pm->gpe0_blk,      /* _MIN */
+                            pm->gpe0_blk,      /* _MAX */
+                            0x0,               /* _ALN */
+                            pm->gpe0_blk_len); /* _LEN */
+                );
             );
         );

Ugh, that's too tricky I'm afraid.

how about:

        crs = build_alloc_array();
        buf = build_alloc_buffer();
        build_append_io(buf, .... );
        build_append_and_cleanup_buffer(crs, buf);

make everything use static functions, not macros.

Quoted 3 lines
Signature · 2 lines
-- 
1.7.1
#10 lore raw

RFC 7/9 acpi: consume CPU hotplug IO resource in PNP0C02 device

In reply to Igor Mammedov
---
 hw/i386/acpi-build.c |   10 ++++++++++
 1 files changed, 10 insertions(+), 0 deletions(-)
hw/i386/acpi-build.c +10 -0
index ce5f715..5cd0c80 100644
@@ -70,6 +70,8 @@ typedef struct AcpiPmInfo {
     uint32_t gpe0_blk;
     uint32_t gpe0_blk_len;
     uint32_t io_base;
+    uint16_t cpuhp_io_base;
+    uint16_t cpuhp_io_len;
 } AcpiPmInfo;
 
 typedef struct AcpiMiscInfo {
@@ -141,11 +143,14 @@ static void acpi_get_pm_info(AcpiPmInfo *pm)
     Object *obj = NULL;
     QObject *o;
 
+    pm->cpuhp_io_len = ACPI_GPE_PROC_LEN;
     if (piix) {
         obj = piix;
+        pm->cpuhp_io_base = PIIX4_CPU_HOTPLUG_IO_BASE;
     }
     if (lpc) {
         obj = lpc;
+        pm->cpuhp_io_base = ICH9_CPU_HOTPLUG_IO_BASE;
     }
     assert(obj);
 
@@ -1145,6 +1150,11 @@ build_ssdt(GArray *table_data, GArray *linker,
                             pm->gpe0_blk,      /* _MAX */
                             0x0,               /* _ALN */
                             pm->gpe0_blk_len); /* _LEN */
+                    ACPI_IO(RESBUF, Decode16,
+                            pm->cpuhp_io_base, /* _MIN */
+                            pm->cpuhp_io_base, /* _MAX */
+                            0x0,               /* _ALN */
+                            pm->cpuhp_io_len); /* _LEN */
                 );
             );
         );
Signature · 1 line
1.7.1
#11 lore raw

Re: [Qemu-devel] [RFC 7/9] acpi: consume CPU hotplug IO resource in PNP0C02 device

Quoted 23 lines · Igor Mammedov
---
 hw/i386/acpi-build.c |   10 ++++++++++
 1 files changed, 10 insertions(+), 0 deletions(-)
hw/i386/acpi-build.c +3 -0
index ce5f715..5cd0c80 100644
@@ -70,6 +70,8 @@ typedef struct AcpiPmInfo {
     uint32_t gpe0_blk;
     uint32_t gpe0_blk_len;
     uint32_t io_base;
+    uint16_t cpuhp_io_base;
+    uint16_t cpuhp_io_len;
 } AcpiPmInfo;
 
 typedef struct AcpiMiscInfo {
@@ -141,11 +143,14 @@ static void acpi_get_pm_info(AcpiPmInfo *pm)
     Object *obj = NULL;
     QObject *o;
 
+    pm->cpuhp_io_len = ACPI_GPE_PROC_LEN;

hw/i386/acpi-build.c static void acpi_get_pm_info(AcpiPmInfo *pm) · line 156

Quoted 10 lines
hw/i386/acpi-build.c +2 -0
     if (piix) {
         obj = piix;
+        pm->cpuhp_io_base = PIIX4_CPU_HOTPLUG_IO_BASE;
     }
     if (lpc) {
         obj = lpc;
+        pm->cpuhp_io_base = ICH9_CPU_HOTPLUG_IO_BASE;
     }
     assert(obj);
 

Not a must but would be nicer to get these as
device properties.

If we change this, cpuhp_io_len can be used directly ...

Quoted 14 lines
hw/i386/acpi-build.c +5 -0
@@ -1145,6 +1150,11 @@ build_ssdt(GArray *table_data, GArray *linker,
                             pm->gpe0_blk,      /* _MAX */
                             0x0,               /* _ALN */
                             pm->gpe0_blk_len); /* _LEN */
+                    ACPI_IO(RESBUF, Decode16,
+                            pm->cpuhp_io_base, /* _MIN */
+                            pm->cpuhp_io_base, /* _MAX */
+                            0x0,               /* _ALN */
+                            pm->cpuhp_io_len); /* _LEN */
                 );
             );
         );
Signature · 1 line
1.7.1
#12 lore raw

RFC 8/9 pcihp: expose PCI hotplug MMIO base/length as properties of piix4pm

In reply to Igor Mammedov

with introduction of PCIHP, MMIO range becomes changable at runtime so it's not possible to statically punch hole PCI bus _CRS.

Making IO base/length available as readonly properties allow acpi builder to get values and reserve PCI hotplug IO range at runtime later.

---
 hw/acpi/pcihp.c         |   28 ++++++++++++++++++++++++++++
 hw/acpi/piix4.c         |    1 +
 include/hw/acpi/pcihp.h |    4 ++++
 3 files changed, 33 insertions(+), 0 deletions(-)
hw/acpi/pcihp.c +28 -0
index 1ce6fc2..629d364 100644
@@ -37,6 +37,7 @@
 #include "hw/pci/pci_bus.h"
 #include "qom/qom-qobject.h"
 #include "qapi/qmp/qint.h"
+#include "qapi/visitor.h"
 
 //#define DEBUG
 
@@ -309,3 +310,30 @@ const VMStateDescription vmstate_acpi_pcihp_pci_status = {
         VMSTATE_END_OF_LIST()
     }
 };
+
+static void acpi_pcihp_get_io_addr(Object *obj, Visitor *v, void *opaque,
+                                     const char *name, Error **errp)
+{
+    AcpiPciHpState *s = opaque;
+    MemoryRegionSection mr_info = memory_region_find(&s->io, 0, 1);
+    uint16_t value = mr_info.offset_within_address_space;
+
+    visit_type_uint16(v, &value, name, errp);
+}
+
+static void acpi_pcihp_get_io_len(Object *obj, Visitor *v, void *opaque,
+                                     const char *name, Error **errp)
+{
+    AcpiPciHpState *s = opaque;
+    uint16_t value = memory_region_size(&s->io);
+
+    visit_type_uint16(v, &value, name, errp);
+}
+
+void acpi_pcihp_add_mmio_properties(Object *obj, AcpiPciHpState *s)
+{
+    object_property_add(obj, ACPI_PCIHP_IO_ADDR, "uint16",
+                        acpi_pcihp_get_io_addr, NULL, NULL, s, NULL);
+    object_property_add(obj, ACPI_PCIHP_IO_LEN, "uint16",
+                        acpi_pcihp_get_io_len, NULL, NULL, s, NULL);
+}
hw/acpi/piix4.c +1 -0
index 7a0efcb..034c5cd 100644
@@ -558,6 +558,7 @@ static void piix4_acpi_system_hot_add_init(MemoryRegion *parent,
                           "acpi-gpe0", GPE_LEN);
     memory_region_add_subregion(parent, GPE_BASE, &s->io_gpe);
 
+    acpi_pcihp_add_mmio_properties(OBJECT(s), &s->acpi_pci_hotplug);
     acpi_pcihp_init(&s->acpi_pci_hotplug, bus, parent,
                     s->use_acpi_pci_hotplug);
 
include/hw/acpi/pcihp.h +4 -0
index 0a90e4a..0eb4e1c 100644
@@ -70,4 +70,8 @@ extern const VMStateDescription vmstate_acpi_pcihp_pci_status;
                                   vmstate_acpi_pcihp_pci_status, \
                                   AcpiPciHpPciStatus)
 
+#define ACPI_PCIHP_IO_ADDR "pcihp-io-addr"
+#define ACPI_PCIHP_IO_LEN  "pcihp-io-len"
+
+void acpi_pcihp_add_mmio_properties(Object *obj, AcpiPciHpState *s);
 #endif
Signature · 1 line
1.7.1
#13 lore raw

Re: [Qemu-devel] [RFC 8/9] pcihp: expose PCI hotplug MMIO base/length as properties of piix4pm

Quoted 9 lines · Igor Mammedov

with introduction of PCIHP, MMIO range becomes changable at runtime so it's not possible to statically punch hole PCI bus _CRS.

Making IO base/length available as readonly properties allow acpi builder to get values and reserve PCI hotplug IO range at runtime later.

No objection but I wonder: isn't it already possible to enumerate memory regions?

Quoted 76 lines
---
 hw/acpi/pcihp.c         |   28 ++++++++++++++++++++++++++++
 hw/acpi/piix4.c         |    1 +
 include/hw/acpi/pcihp.h |    4 ++++
 3 files changed, 33 insertions(+), 0 deletions(-)
hw/acpi/pcihp.c +28 -0
index 1ce6fc2..629d364 100644
@@ -37,6 +37,7 @@
 #include "hw/pci/pci_bus.h"
 #include "qom/qom-qobject.h"
 #include "qapi/qmp/qint.h"
+#include "qapi/visitor.h"
 
 //#define DEBUG
 
@@ -309,3 +310,30 @@ const VMStateDescription vmstate_acpi_pcihp_pci_status = {
         VMSTATE_END_OF_LIST()
     }
 };
+
+static void acpi_pcihp_get_io_addr(Object *obj, Visitor *v, void *opaque,
+                                     const char *name, Error **errp)
+{
+    AcpiPciHpState *s = opaque;
+    MemoryRegionSection mr_info = memory_region_find(&s->io, 0, 1);
+    uint16_t value = mr_info.offset_within_address_space;
+
+    visit_type_uint16(v, &value, name, errp);
+}
+
+static void acpi_pcihp_get_io_len(Object *obj, Visitor *v, void *opaque,
+                                     const char *name, Error **errp)
+{
+    AcpiPciHpState *s = opaque;
+    uint16_t value = memory_region_size(&s->io);
+
+    visit_type_uint16(v, &value, name, errp);
+}
+
+void acpi_pcihp_add_mmio_properties(Object *obj, AcpiPciHpState *s)
+{
+    object_property_add(obj, ACPI_PCIHP_IO_ADDR, "uint16",
+                        acpi_pcihp_get_io_addr, NULL, NULL, s, NULL);
+    object_property_add(obj, ACPI_PCIHP_IO_LEN, "uint16",
+                        acpi_pcihp_get_io_len, NULL, NULL, s, NULL);
+}
hw/acpi/piix4.c +1 -0
index 7a0efcb..034c5cd 100644
@@ -558,6 +558,7 @@ static void piix4_acpi_system_hot_add_init(MemoryRegion *parent,
                           "acpi-gpe0", GPE_LEN);
     memory_region_add_subregion(parent, GPE_BASE, &s->io_gpe);
 
+    acpi_pcihp_add_mmio_properties(OBJECT(s), &s->acpi_pci_hotplug);
     acpi_pcihp_init(&s->acpi_pci_hotplug, bus, parent,
                     s->use_acpi_pci_hotplug);
 
include/hw/acpi/pcihp.h +4 -0
index 0a90e4a..0eb4e1c 100644
@@ -70,4 +70,8 @@ extern const VMStateDescription vmstate_acpi_pcihp_pci_status;
                                   vmstate_acpi_pcihp_pci_status, \
                                   AcpiPciHpPciStatus)
 
+#define ACPI_PCIHP_IO_ADDR "pcihp-io-addr"
+#define ACPI_PCIHP_IO_LEN  "pcihp-io-len"
+
+void acpi_pcihp_add_mmio_properties(Object *obj, AcpiPciHpState *s);
 #endif
Signature · 1 line
1.7.1
#14 lore raw

RFC 9/9 acpi: consume PCIHP IO resource in PNP0C02 device

In reply to Igor Mammedov
---
 hw/i386/acpi-build.c |   22 ++++++++++++++++++++++
 1 files changed, 22 insertions(+), 0 deletions(-)
hw/i386/acpi-build.c +22 -0
index 5cd0c80..e3d145b 100644
@@ -72,6 +72,8 @@ typedef struct AcpiPmInfo {
     uint32_t io_base;
     uint16_t cpuhp_io_base;
     uint16_t cpuhp_io_len;
+    uint16_t pcihp_io_base;
+    uint16_t pcihp_io_len;
 } AcpiPmInfo;
 
 typedef struct AcpiMiscInfo {
@@ -189,6 +191,19 @@ static void acpi_get_pm_info(AcpiPmInfo *pm)
                                            NULL);
     pm->gpe0_blk_len = object_property_get_int(obj, ACPI_PM_PROP_GPE0_BLK_LEN,
                                                NULL);
+    {
+        Error  *local_error = NULL;
+        uint16_t val = object_property_get_int(obj, ACPI_PCIHP_IO_ADDR,
+                                               &local_error);
+        if (!local_error) {
+            pm->pcihp_io_base = val;
+        }
+
+        val = object_property_get_int(obj, ACPI_PCIHP_IO_LEN, &local_error);
+        if (!local_error) {
+            pm->pcihp_io_len = val;
+        }
+    }
 }
 
 static void acpi_get_misc_info(AcpiMiscInfo *info)
@@ -1155,6 +1170,13 @@ build_ssdt(GArray *table_data, GArray *linker,
                             pm->cpuhp_io_base, /* _MAX */
                             0x0,               /* _ALN */
                             pm->cpuhp_io_len); /* _LEN */
+                    if (pm->pcihp_io_base) {
+                        ACPI_IO(RESBUF, Decode16,
+                                pm->pcihp_io_base, /* _MIN */
+                                pm->pcihp_io_base, /* _MAX */
+                                0x0,               /* _ALN */
+                                pm->pcihp_io_len); /* _LEN */
+                    }
                 );
             );
         );
Signature · 1 line
1.7.1
#15 lore raw
In reply to Igor Mammedov
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

What makes it runtime changeable?

Quoted 5 lines

So replace static hole punching with dynamically consumed resources in a child device on PCI0 bus. i.e generate PNP0C02 device as a child of PCI0 bus at runtime and consume GPE0, PCI/CPU hotplug IO resources in it instead of punching holes in static PCI0._CRS.

It seems that we are being too exact with
IO resources here.
Can't we roughly reserve 0xae00 to 0xafff
and be done with it?

Quoted 28 lines

Tested with Windows XPsp3, Vista, Windows Server 2003, 2008, 2012r2.

PS:
Series adds several ASL like macros to simplify code for dynamic generation of AML structures.

Igor Mammedov (9):
Revert "pc: Q35 DSDT: exclude CPU hotplug IO range from PCI bus
resources"
Revert "pc: PIIX DSDT: exclude CPU/PCI hotplug & GPE0 IO range from
PCI bus resources"
Partial revert "pc: ACPI: expose PRST IO range via _CRS"
acpi: replace opencoded opcodes with defines
acpi: add PNP0C02 to PCI0 bus
acpi: consume GPE0 IO resources in PNP0C02 device
acpi: consume CPU hotplug IO resource in PNP0C02 device
pcihp: expose PCI hotplug MMIO base/length as properties of piix4pm
acpi: consume PCIHP IO resource in PNP0C02 device

 hw/acpi/pcihp.c                   |   28 ++++++
 hw/acpi/piix4.c                   |    1 +
 hw/i386/acpi-build.c              |  177 ++++++++++++++++++++++++++++++++++--
 hw/i386/acpi-dsdt-cpu-hotplug.dsl |   11 ---
 hw/i386/acpi-dsdt-pci-crs.dsl     |   15 +++-
 hw/i386/acpi-dsdt.dsl             |   39 --------
 hw/i386/q35-acpi-dsdt.dsl         |   16 ----
 include/hw/acpi/pcihp.h           |    4 +
 8 files changed, 214 insertions(+), 77 deletions(-)
#16 lore raw
Quoted 6 lines · Michael S. Tsirkin
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

What makes it runtime changeable?

machine type. q35 / piix map them at different locations.

Also we might want to this also for devices which are runtime-configurable (isa-debugcon, pvpanic, ...).

cheers,
Gerd

#17 lore raw
Quoted 9 lines · Gerd Hoffmann
Quoted 6 lines · Michael S. Tsirkin
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

What makes it runtime changeable?

machine type. q35 / piix map them at different locations.

That's not dynamic. We can load the correct ones per DSDT.

Quoted 2 lines

Also we might want to this also for devices which are runtime-configurable (isa-debugcon, pvpanic, ...).

That's more convincing, but I don't want
knowledge of all these devices in acpi-build.
Also we need to make seabios avoid these ranges
when enumerating devices.
How does it know to avoid them ATM?

Quoted 4 lines

cheers,
Gerd

#18 lore raw
Quoted 13 lines · Michael S. Tsirkin
Quoted 9 lines · Gerd Hoffmann
Quoted 6 lines · Michael S. Tsirkin
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

What makes it runtime changeable?

machine type. q35 / piix map them at different locations.

That's not dynamic. We can load the correct ones per DSDT.

Quoted 8 lines
Quoted 2 lines

Also we might want to this also for devices which are runtime-configurable (isa-debugcon, pvpanic, ...).

That's more convincing, but I don't want
knowledge of all these devices in acpi-build.
Also we need to make seabios avoid these ranges
when enumerating devices.
How does it know to avoid them ATM?

seabios maps io ports @ 0xc000 up.
recently it has changed to use 0x1000 -> 0xa000 region in case the hole above 0xc000 is too small.

In other words: It doesn't map anything below 0x1000 and it avoids 0xa000 -> 0xbfff. Hardcoded. I want lift the later restriction on q35, by moving pmbase (0xb000 atm) out of the way, so seabios can use the whole 0x1000 -> 0xffff range, but that is still wip.

cheers,
Gerd

#19 lore raw
Quoted 27 lines · Gerd Hoffmann
Quoted 13 lines · Michael S. Tsirkin
Quoted 9 lines · Gerd Hoffmann
Quoted 6 lines · Michael S. Tsirkin
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

What makes it runtime changeable?

machine type. q35 / piix map them at different locations.

That's not dynamic. We can load the correct ones per DSDT.

Quoted 8 lines
Quoted 2 lines

Also we might want to this also for devices which are runtime-configurable (isa-debugcon, pvpanic, ...).

That's more convincing, but I don't want
knowledge of all these devices in acpi-build.
Also we need to make seabios avoid these ranges
when enumerating devices.
How does it know to avoid them ATM?

seabios maps io ports @ 0xc000 up.
recently it has changed to use 0x1000 -> 0xa000 region in case the hole above 0xc000 is too small.

Quoted 8 lines

In other words: It doesn't map anything below 0x1000 and it avoids 0xa000 -> 0xbfff. Hardcoded. I want lift the later restriction on q35, by moving pmbase (0xb000 atm) out of the way, so seabios can use the whole 0x1000 -> 0xffff range, but that is still wip.

cheers,
Gerd

okay so we'll want some fwcfg for that correct?
whatever fills that, can share logic with acpi
generation :)

That might or might not make it dynamic enough to make it worth bothering - alternative is just two version of acpi depending on machine type.

Signature · 1 line
MST
#20 lore raw
In reply to Gerd Hoffmann
Quoted 6 lines · On Mon, 17 Feb 2014 09:32:35 +0100 Gerd Hoffmann <[email protected]> wrote:
Quoted 5 lines · Michael S. Tsirkin
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

Quoted 6 lines
Quoted 1 line

What makes it runtime changeable?

machine type. q35 / piix map them at different locations.

Also we might want to this also for devices which are runtime-configurable (isa-debugcon, pvpanic, ...).

I'd convert simple devices that conditionally enabled at startup time, from static definition + patching into completely dynamically generated when device present. For example pvpanic falls in to this category.

That would result in smaller ACPI tables guest has to deal with.

Quoted 5 lines

cheers,
Gerd

#21 lore raw
Quoted 13 lines · Igor Mammedov
Quoted 10 lines · On Mon, 17 Feb 2014 09:32:35 +0100 Gerd Hoffmann <[email protected]> wrote:
Quoted 6 lines · Michael S. Tsirkin
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

What makes it runtime changeable?

machine type. q35 / piix map them at different locations.

Quoted 8 lines
Quoted 2 lines

Also we might want to this also for devices which are runtime-configurable (isa-debugcon, pvpanic, ...).

I'd convert simple devices that conditionally enabled at startup time, from static definition + patching into completely dynamically generated when device present. For example pvpanic falls in to this category.

That would result in smaller ACPI tables guest has to deal with.

I could be mistaken, but AFAIR this caused the windows device manager to pop up in windows? Ie. if you have a windows guest and cold-boot it twice, once with the device present (generated into ACPI) and once with the device absent (not generated into ACPI), then you get hardware changes. Whereas, if the device is always present and you only patch _STA, then windows doesn't perceive it as a hw change.

Do I recall it right?...

You could argue that "a new device indeed warrants a device manager popup", but esp. for isa-debugcon and pvpanic, you might want to enable those opportunistically, without triggering a new hw dialog. Pvpanic triggering the device manager was exactly what drew frowns, for its original implementation. IIRC.

Anyway pls. feel free to ignore this comment, it just crossed my mind. (And of course it's not related to your series.)

Thanks
Laszlo

#22 lore raw
Quoted 21 lines · On Tue, 18 Feb 2014 23:04:13 +0100 Laszlo Ersek <[email protected]> wrote:
Quoted 20 lines · Igor Mammedov
Quoted 6 lines · On Mon, 17 Feb 2014 09:32:35 +0100 Gerd Hoffmann <[email protected]> wrote:
Quoted 5 lines · Michael S. Tsirkin
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

Quoted 6 lines
Quoted 1 line

What makes it runtime changeable?

machine type. q35 / piix map them at different locations.

Also we might want to this also for devices which are runtime-configurable (isa-debugcon, pvpanic, ...).

I'd convert simple devices that conditionally enabled at startup time, from static definition + patching into completely dynamically generated when device present. For example pvpanic falls in to this category.

Quoted 8 lines
Quoted 1 line

That would result in smaller ACPI tables guest has to deal with.

I could be mistaken, but AFAIR this caused the windows device manager to pop up in windows? Ie. if you have a windows guest and cold-boot it twice, once with the device present (generated into ACPI) and once with the device absent (not generated into ACPI), then you get hardware changes. Whereas, if the device is always present and you only patch _STA, then windows doesn't perceive it as a hw change.

Is is irrelevant whether device is statically or dynamically created, user will face the same issue and has a choice to install driver or tell windows to ignore device forever.

Quoted 1 line

Do I recall it right?...

Device manager pop-ups only once when new device is added to install driver. If later, the device in the same location with the same id appears/disappears, Device manager handles it silently.

Quoted 6 lines

You could argue that "a new device indeed warrants a device manager popup", but esp. for isa-debugcon and pvpanic, you might want to enable those opportunistically, without triggering a new hw dialog. Pvpanic triggering the device manager was exactly what drew frowns, for its original implementation. IIRC.

the above applies to these cases as well, i.e. if you install driver for it then there won't be any pop-ups later. If there is no driver (which is the case) one needs to tell to Device manager to ignore this device, and then there shouldn't be an additional pop-ups later.

Quoted 6 lines

Anyway pls. feel free to ignore this comment, it just crossed my mind. (And of course it's not related to your series.)

Thanks
Laszlo

#23 lore raw
In reply to Michael S. Tsirkin
Quoted 6 lines · On Sun, 16 Feb 2014 17:53:45 +0200 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

What makes it runtime changeable?

piix machine "acpi-pci-hotplug-with-bridge-support=on" effectively changes size of pcihp MMIO region

Quoted 11 lines
Quoted 6 lines

So replace static hole punching with dynamically consumed resources in a child device on PCI0 bus. i.e generate PNP0C02 device as a child of PCI0 bus at runtime and consume GPE0, PCI/CPU hotplug IO resources in it instead of punching holes in static PCI0._CRS.

It seems that we are being too exact with
IO resources here.
Can't we roughly reserve 0xae00 to 0xafff
and be done with it?

that would be easiest way for this specific case if we could agree for ranges on PIIX/Q35 machines.

But I also use it as excuse to introduce ASL like macros so that building dynamic SSDT would be easier i.e. replace template patching with single place where dynamic device is defined and its values are set in simple and straightforward manner.

Quoted 28 lines
Quoted 28 lines

Tested with Windows XPsp3, Vista, Windows Server 2003, 2008, 2012r2.

PS:
Series adds several ASL like macros to simplify code for dynamic generation of AML structures.

Igor Mammedov (9):
Revert "pc: Q35 DSDT: exclude CPU hotplug IO range from PCI bus
resources"
Revert "pc: PIIX DSDT: exclude CPU/PCI hotplug & GPE0 IO range from
PCI bus resources"
Partial revert "pc: ACPI: expose PRST IO range via _CRS"
acpi: replace opencoded opcodes with defines
acpi: add PNP0C02 to PCI0 bus
acpi: consume GPE0 IO resources in PNP0C02 device
acpi: consume CPU hotplug IO resource in PNP0C02 device
pcihp: expose PCI hotplug MMIO base/length as properties of piix4pm
acpi: consume PCIHP IO resource in PNP0C02 device

 hw/acpi/pcihp.c                   |   28 ++++++
 hw/acpi/piix4.c                   |    1 +
 hw/i386/acpi-build.c              |  177 ++++++++++++++++++++++++++++++++++--
 hw/i386/acpi-dsdt-cpu-hotplug.dsl |   11 ---
 hw/i386/acpi-dsdt-pci-crs.dsl     |   15 +++-
 hw/i386/acpi-dsdt.dsl             |   39 --------
 hw/i386/q35-acpi-dsdt.dsl         |   16 ----
 include/hw/acpi/pcihp.h           |    4 +
 8 files changed, 214 insertions(+), 77 deletions(-)
Signature · 2 lines
Regards,
  Igor
#24 lore raw
Quoted 11 lines · Igor Mammedov
Quoted 6 lines · On Sun, 16 Feb 2014 17:53:45 +0200 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

What makes it runtime changeable?

piix machine "acpi-pci-hotplug-with-bridge-support=on" effectively changes size of pcihp MMIO region

It adds 4 bytes. Let's just reserve a reasonably sized region, like 512 bytes.

Quoted 11 lines
Quoted 11 lines
Quoted 6 lines

So replace static hole punching with dynamically consumed resources in a child device on PCI0 bus. i.e generate PNP0C02 device as a child of PCI0 bus at runtime and consume GPE0, PCI/CPU hotplug IO resources in it instead of punching holes in static PCI0._CRS.

It seems that we are being too exact with
IO resources here.
Can't we roughly reserve 0xae00 to 0xafff
and be done with it?

Quoted 7 lines

that would be easiest way for this specific case if we could agree for ranges on PIIX/Q35 machines.

But I also use it as excuse to introduce ASL like macros so that building dynamic SSDT would be easier i.e. replace template patching with single place where dynamic device is defined and its values are set in simple and straightforward manner.

We don't need excuses really :)
But the approach itself needs some work IMHO, in particular ASL like syntax by using macros and varargs is not something to strive for IMHO :)

Why don't we just pass GArrays around?

crs = build_alloc_crs();
build_append_resource(crs, ....)

Or if you really want this, using array of GArray:

build_append_crs(ssdt,

		{
			build_alloc_resource(...),
			build_alloc_resource(...),
			build_alloc_resource(...),
			build_alloc_resource(...),
			NULL
		})
Quoted 33 lines
Quoted 28 lines
Quoted 28 lines

Tested with Windows XPsp3, Vista, Windows Server 2003, 2008, 2012r2.

PS:
Series adds several ASL like macros to simplify code for dynamic generation of AML structures.

Igor Mammedov (9):
Revert "pc: Q35 DSDT: exclude CPU hotplug IO range from PCI bus
resources"
Revert "pc: PIIX DSDT: exclude CPU/PCI hotplug & GPE0 IO range from
PCI bus resources"
Partial revert "pc: ACPI: expose PRST IO range via _CRS"
acpi: replace opencoded opcodes with defines
acpi: add PNP0C02 to PCI0 bus
acpi: consume GPE0 IO resources in PNP0C02 device
acpi: consume CPU hotplug IO resource in PNP0C02 device
pcihp: expose PCI hotplug MMIO base/length as properties of piix4pm
acpi: consume PCIHP IO resource in PNP0C02 device

 hw/acpi/pcihp.c                   |   28 ++++++
 hw/acpi/piix4.c                   |    1 +
 hw/i386/acpi-build.c              |  177 ++++++++++++++++++++++++++++++++++--
 hw/i386/acpi-dsdt-cpu-hotplug.dsl |   11 ---
 hw/i386/acpi-dsdt-pci-crs.dsl     |   15 +++-
 hw/i386/acpi-dsdt.dsl             |   39 --------
 hw/i386/q35-acpi-dsdt.dsl         |   16 ----
 include/hw/acpi/pcihp.h           |    4 +
 8 files changed, 214 insertions(+), 77 deletions(-)
Signature · 2 lines
Regards,
  Igor
#25 lore raw
Quoted 9 lines · On Mon, 17 Feb 2014 13:02:18 +0200 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 8 lines · Igor Mammedov
Quoted 5 lines · On Sun, 16 Feb 2014 17:53:45 +0200 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

What makes it runtime changeable?

piix machine "acpi-pci-hotplug-with-bridge-support=on" effectively changes size of pcihp MMIO region

It adds 4 bytes. Let's just reserve a reasonably sized region, like 512 bytes.

Reserve where and for how log it will be enough? Eventually we will continue punching holes or increasing size of this one big hole.

Q35 adds more to the problem with guest programmable pm_io_base, we can't statically punch hole for it.

Quoted 25 lines
Quoted 12 lines
Quoted 8 lines
Quoted 7 lines

So replace static hole punching with dynamically consumed resources in a child device on PCI0 bus. i.e generate PNP0C02 device as a child of PCI0 bus at runtime and consume GPE0, PCI/CPU hotplug IO resources in it instead of punching holes in static PCI0._CRS.

Quoted 4 lines

It seems that we are being too exact with
IO resources here.
Can't we roughly reserve 0xae00 to 0xafff
and be done with it?

Quoted 7 lines

that would be easiest way for this specific case if we could agree for ranges on PIIX/Q35 machines.

But I also use it as excuse to introduce ASL like macros so that building dynamic SSDT would be easier i.e. replace template patching with single place where dynamic device is defined and its values are set in simple and straightforward manner.

We don't need excuses really :)
But the approach itself needs some work IMHO, in particular ASL like syntax by using macros and varargs is not something to strive for IMHO :)

Quoted 4 lines

Why don't we just pass GArrays around?

crs = build_alloc_crs();
build_append_resource(crs, ....)

I don't like much macros myself, but compared to alternative approach ^^^ (I've tried it) it's still harder to read/write ASL constructs right.

I think that creating AML using ASL like flow and primitives is better than building it using custom APIs, since it looks like it was documented in spec.
Macros+vararg provide a necessary degree of flexibility to create AML using ASL like flow and constructs.

Quoted 4 lines

Or if you really want this, using array of GArray:

Quoted 8 lines

build_append_crs(ssdt,

		{
			build_alloc_resource(...),
			build_alloc_resource(...),
			build_alloc_resource(...),
			build_alloc_resource(...),
			NULL
		})

that would work if there was a way to specify arbitrary statements inside embedded array (like in the last hunk of [9/9]).

Quoted 35 lines
Quoted 35 lines
Quoted 30 lines
Quoted 30 lines

Tested with Windows XPsp3, Vista, Windows Server 2003, 2008, 2012r2.

PS:
Series adds several ASL like macros to simplify code for dynamic generation of AML structures.

Igor Mammedov (9):
Revert "pc: Q35 DSDT: exclude CPU hotplug IO range from PCI bus
resources"
Revert "pc: PIIX DSDT: exclude CPU/PCI hotplug & GPE0 IO range from
PCI bus resources"
Partial revert "pc: ACPI: expose PRST IO range via _CRS"
acpi: replace opencoded opcodes with defines
acpi: add PNP0C02 to PCI0 bus
acpi: consume GPE0 IO resources in PNP0C02 device
acpi: consume CPU hotplug IO resource in PNP0C02 device
pcihp: expose PCI hotplug MMIO base/length as properties of piix4pm
acpi: consume PCIHP IO resource in PNP0C02 device

 hw/acpi/pcihp.c                   |   28 ++++++
 hw/acpi/piix4.c                   |    1 +
 hw/i386/acpi-build.c              |  177 ++++++++++++++++++++++++++++++++++--
 hw/i386/acpi-dsdt-cpu-hotplug.dsl |   11 ---
 hw/i386/acpi-dsdt-pci-crs.dsl     |   15 +++-
 hw/i386/acpi-dsdt.dsl             |   39 --------
 hw/i386/q35-acpi-dsdt.dsl         |   16 ----
 include/hw/acpi/pcihp.h           |    4 +
 8 files changed, 214 insertions(+), 77 deletions(-)
Signature · 2 lines
Regards,
  Igor
#26 lore raw
Quoted 13 lines · Igor Mammedov
Quoted 10 lines · On Mon, 17 Feb 2014 13:02:18 +0200 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 9 lines · Igor Mammedov
Quoted 6 lines · On Sun, 16 Feb 2014 17:53:45 +0200 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

What makes it runtime changeable?

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

piix machine "acpi-pci-hotplug-with-bridge-support=on" effectively changes size of pcihp MMIO region

It adds 4 bytes. Let's just reserve a reasonably sized region, like 512 bytes.

Reserve where and for how log it will be enough? Eventually we will continue punching holes or increasing size of this one big hole.

I agree this might become necessary. But why not keep it simple while we can? The advantage of ACPI in qemu is we can fix it without touching guest code. Let's use it.

So if there's just pm_io_base, we can just write
it in ASL and patch in that value.

Quoted 3 lines

Q35 adds more to the problem with guest programmable pm_io_base, we can't statically punch hole for it.

Aha. That's a good point.
But I still worry how we'll handle e.g. pvpanic.
Paolo had an idea to scan the memory range,
and punch holes for each section, making an
exception for regions owned by PCI devices.
I thought it would work originally too, but it
seems that e.g. VGA range is also owned by pci
devices sometimes.

Quoted 34 lines
Quoted 25 lines
Quoted 12 lines
Quoted 8 lines
Quoted 7 lines

So replace static hole punching with dynamically consumed resources in a child device on PCI0 bus. i.e generate PNP0C02 device as a child of PCI0 bus at runtime and consume GPE0, PCI/CPU hotplug IO resources in it instead of punching holes in static PCI0._CRS.

Quoted 4 lines

It seems that we are being too exact with
IO resources here.
Can't we roughly reserve 0xae00 to 0xafff
and be done with it?

Quoted 7 lines

that would be easiest way for this specific case if we could agree for ranges on PIIX/Q35 machines.

But I also use it as excuse to introduce ASL like macros so that building dynamic SSDT would be easier i.e. replace template patching with single place where dynamic device is defined and its values are set in simple and straightforward manner.

We don't need excuses really :)
But the approach itself needs some work IMHO, in particular ASL like syntax by using macros and varargs is not something to strive for IMHO :)

Quoted 4 lines

Why don't we just pass GArrays around?

crs = build_alloc_crs();
build_append_resource(crs, ....)

I don't like much macros myself, but compared to alternative approach ^^^ (I've tried it) it's still harder to read/write ASL constructs right.

Quoted 5 lines

I think that creating AML using ASL like flow and primitives is better than building it using custom APIs, since it looks like it was documented in spec.
Macros+vararg provide a necessary degree of flexibility to create AML using ASL like flow and constructs.

Only you lose all type safety. When compiling ASL with IASL at least it does some type checking.

Quoted 12 lines
Quoted 4 lines

Or if you really want this, using array of GArray:

Quoted 8 lines

build_append_crs(ssdt,

		{
			build_alloc_resource(...),
			build_alloc_resource(...),
			build_alloc_resource(...),
			build_alloc_resource(...),
			NULL
		})
Quoted 2 lines

that would work if there was a way to specify arbitrary statements inside embedded array (like in the last hunk of [9/9]).

In C we have functions for this :)
Last chunk:

- if (pm->pcihp_io_base) {
- ACPI_IO(RESBUF, Decode16,

-                                pm->pcihp_io_base, /* _MIN */
-                                pm->pcihp_io_base, /* _MAX */
-                                0x0,               /* _ALN */
-                                pm->pcihp_io_len); /* _LEN */

- }

+ build_io_resource_16(pm->pcihp_io_base, pm->pcihp_io_base,

			     0x0, pm->pcihp_io_len);

Which actually makes one think. Isn't _MAX wrong here? I'd expect pm->pcihp_io_base + pm->pcihp_io_len - 1

I would also key this off pcihp_io_len.

Quoted 35 lines
Quoted 35 lines
Quoted 35 lines
Quoted 30 lines
Quoted 30 lines

Tested with Windows XPsp3, Vista, Windows Server 2003, 2008, 2012r2.

PS:
Series adds several ASL like macros to simplify code for dynamic generation of AML structures.

Igor Mammedov (9):
Revert "pc: Q35 DSDT: exclude CPU hotplug IO range from PCI bus
resources"
Revert "pc: PIIX DSDT: exclude CPU/PCI hotplug & GPE0 IO range from
PCI bus resources"
Partial revert "pc: ACPI: expose PRST IO range via _CRS"
acpi: replace opencoded opcodes with defines
acpi: add PNP0C02 to PCI0 bus
acpi: consume GPE0 IO resources in PNP0C02 device
acpi: consume CPU hotplug IO resource in PNP0C02 device
pcihp: expose PCI hotplug MMIO base/length as properties of piix4pm
acpi: consume PCIHP IO resource in PNP0C02 device

 hw/acpi/pcihp.c                   |   28 ++++++
 hw/acpi/piix4.c                   |    1 +
 hw/i386/acpi-build.c              |  177 ++++++++++++++++++++++++++++++++++--
 hw/i386/acpi-dsdt-cpu-hotplug.dsl |   11 ---
 hw/i386/acpi-dsdt-pci-crs.dsl     |   15 +++-
 hw/i386/acpi-dsdt.dsl             |   39 --------
 hw/i386/q35-acpi-dsdt.dsl         |   16 ----
 include/hw/acpi/pcihp.h           |    4 +
 8 files changed, 214 insertions(+), 77 deletions(-)
Signature · 2 lines
Regards,
  Igor
#27 lore raw
Quoted 34 lines · On Tue, 18 Feb 2014 13:33:17 +0200 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 13 lines · Igor Mammedov
Quoted 10 lines · On Mon, 17 Feb 2014 13:02:18 +0200 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 9 lines · Igor Mammedov
Quoted 6 lines · On Sun, 16 Feb 2014 17:53:45 +0200 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 3 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

What makes it runtime changeable?

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

piix machine "acpi-pci-hotplug-with-bridge-support=on" effectively changes size of pcihp MMIO region

It adds 4 bytes. Let's just reserve a reasonably sized region, like 512 bytes.

Reserve where and for how log it will be enough? Eventually we will continue punching holes or increasing size of this one big hole.

I agree this might become necessary. But why not keep it simple while we can? The advantage of ACPI in qemu is we can fix it without touching guest code. Let's use it.

So if there's just pm_io_base, we can just write
it in ASL and patch in that value.

Quoted 3 lines

Q35 adds more to the problem with guest programmable pm_io_base, we can't statically punch hole for it.

Quoted 8 lines

Aha. That's a good point.
But I still worry how we'll handle e.g. pvpanic.
Paolo had an idea to scan the memory range,
and punch holes for each section, making an
exception for regions owned by PCI devices.
I thought it would work originally too, but it
seems that e.g. VGA range is also owned by pci
devices sometimes.

from ACPI pov, I think we need to punch holes or define _CRS in respective consuming device explicitly, doing it automatically looping through MemoryRegions going to be tricky.

but for SeaBIOS to find out what ports are reserved looping might be a right way, if it needs this information.

Quoted 35 lines
Quoted 35 lines
Quoted 26 lines
Quoted 13 lines
Quoted 9 lines
Quoted 8 lines

So replace static hole punching with dynamically consumed resources in a child device on PCI0 bus. i.e generate PNP0C02 device as a child of PCI0 bus at runtime and consume GPE0, PCI/CPU hotplug IO resources in it instead of punching holes in static PCI0._CRS.

Quoted 4 lines

It seems that we are being too exact with
IO resources here.
Can't we roughly reserve 0xae00 to 0xafff
and be done with it?

Quoted 7 lines

that would be easiest way for this specific case if we could agree for ranges on PIIX/Q35 machines.

But I also use it as excuse to introduce ASL like macros so that building dynamic SSDT would be easier i.e. replace template patching with single place where dynamic device is defined and its values are set in simple and straightforward manner.

We don't need excuses really :)
But the approach itself needs some work IMHO, in particular ASL like syntax by using macros and varargs is not something to strive for IMHO :)

Quoted 4 lines

Why don't we just pass GArrays around?

crs = build_alloc_crs();
build_append_resource(crs, ....)

I don't like much macros myself, but compared to alternative approach ^^^ (I've tried it) it's still harder to read/write ASL constructs right.

Quoted 8 lines
Quoted 5 lines

I think that creating AML using ASL like flow and primitives is better than building it using custom APIs, since it looks like it was documented in spec.
Macros+vararg provide a necessary degree of flexibility to create AML using ASL like flow and constructs.

Only you lose all type safety. When compiling ASL with IASL at least it does some type checking.

short of embedded IASL solution, it's the same type safety as we could provide with C, with extra checks inside primitives. So that API user won't have to worry about it.

Quoted 13 lines
Quoted 13 lines
Quoted 13 lines

Or if you really want this, using array of GArray:

build_append_crs(ssdt,

		{
			build_alloc_resource(...),
			build_alloc_resource(...),
			build_alloc_resource(...),
			build_alloc_resource(...),
			NULL
		})
Quoted 4 lines
Quoted 2 lines

that would work if there was a way to specify arbitrary statements inside embedded array (like in the last hunk of [9/9]).

In C we have functions for this :)

I don't say it's impossible to write using C functions to express the same logic, but it would be harder to read/write code that way. For example below snippet would look like:

GArray *return_array_if(bool, garray){
if (bool == true) return garray;
return empty_garray;
}

build_append_crs(ssdt,
{
build_alloc_resource(...),

                        ...,
                        return_array_if(have_pcihp,
                                        build_alloc_resource(...)
                        ),

and that will turn out in explosive growth of "return_array_if" like functions to accommodate every tweak we would to do.
On contrary, macros with inline code make it simpler/easier to write/read code, almost like embedded ASL.

Quoted 12 lines

Last chunk:

(quoted diff) +1 -7
-                    if (pm->pcihp_io_base) {
-                        ACPI_IO(RESBUF, Decode16,
-                                pm->pcihp_io_base, /* _MIN */
-                                pm->pcihp_io_base, /* _MAX */
-                                0x0,               /* _ALN */
-                                pm->pcihp_io_len); /* _LEN */
-                    }
 
 
+	build_io_resource_16(pm->pcihp_io_base, pm->pcihp_io_base,
Quoted 1 line
			     0x0, pm->pcihp_io_len);

That's not the same, Do you mean:

build_io_resource_16_IF(have_pcihp, ....);

and returning empty GArray if have_pcihp == false?

If it's the case than that's what I'm trying to avoid, in favor of pretty well described in spec ASL like primitives with some C snippets to tweak generated AML object content.

If we use C++11 lambda feature we could achieve almost the same simple to use API as macro based, but I guess it's not an option yet.

Quoted 4 lines

Which actually makes one think. Isn't _MAX wrong here? I'd expect pm->pcihp_io_base + pm->pcihp_io_len - 1

nope,
"
19.5.62 IO (IO Resource Descriptor Macro)
...
AddressMax evaluates to a 16-bit integer that specifies the maximum acceptable STARTING address for the I/O range.
"

Quoted 38 lines

I would also key this off pcihp_io_len.

Quoted 35 lines
Quoted 35 lines
Quoted 35 lines
Quoted 30 lines
Quoted 30 lines

Tested with Windows XPsp3, Vista, Windows Server 2003, 2008, 2012r2.

PS:
Series adds several ASL like macros to simplify code for dynamic generation of AML structures.

Igor Mammedov (9):
Revert "pc: Q35 DSDT: exclude CPU hotplug IO range from PCI bus
resources"
Revert "pc: PIIX DSDT: exclude CPU/PCI hotplug & GPE0 IO range from
PCI bus resources"
Partial revert "pc: ACPI: expose PRST IO range via _CRS"
acpi: replace opencoded opcodes with defines
acpi: add PNP0C02 to PCI0 bus
acpi: consume GPE0 IO resources in PNP0C02 device
acpi: consume CPU hotplug IO resource in PNP0C02 device
pcihp: expose PCI hotplug MMIO base/length as properties of piix4pm
acpi: consume PCIHP IO resource in PNP0C02 device

 hw/acpi/pcihp.c                   |   28 ++++++
 hw/acpi/piix4.c                   |    1 +
 hw/i386/acpi-build.c              |  177 ++++++++++++++++++++++++++++++++++--
 hw/i386/acpi-dsdt-cpu-hotplug.dsl |   11 ---
 hw/i386/acpi-dsdt-pci-crs.dsl     |   15 +++-
 hw/i386/acpi-dsdt.dsl             |   39 --------
 hw/i386/q35-acpi-dsdt.dsl         |   16 ----
 include/hw/acpi/pcihp.h           |    4 +
 8 files changed, 214 insertions(+), 77 deletions(-)
Signature · 2 lines
Regards,
  Igor
#28 lore raw
In reply to Igor Mammedov
Quoted 9 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

So replace static hole punching with dynamically consumed resources in a child device on PCI0 bus. i.e generate PNP0C02 device as a child of PCI0 bus at runtime and consume GPE0, PCI/CPU hotplug IO resources in it instead of punching holes in static PCI0._CRS.

Nice. Can you try to do that for the mmconf xbar in memory space too while being at it?

cheers,
Gerd

#29 lore raw
Quoted 5 lines · On Mon, 17 Feb 2014 09:29:20 +0100 Gerd Hoffmann <[email protected]> wrote:
Quoted 4 lines · Igor Mammedov

Since introduction of PCIHP, it became problematic to punch hole in PCI0._CRS statically since PCI hotplug region size became runtime changeable.

Quoted 8 lines
Quoted 5 lines

So replace static hole punching with dynamically consumed resources in a child device on PCI0 bus. i.e generate PNP0C02 device as a child of PCI0 bus at runtime and consume GPE0, PCI/CPU hotplug IO resources in it instead of punching holes in static PCI0._CRS.

Nice. Can you try to do that for the mmconf xbar in memory space too while being at it?

If I manage to convince mst in using macro approach.

PS:
Although his build_append_foo() functions have tremendously improved dynamic AML generation (made it much less error-prone), the current acpi-build.c becomes harder to read (just look at new pcihp parts). So I'm arguing in favor of more simple/intuitive API approach even if it uses macros (I'm really not fun of it, but result is much clear/maintainable code).

Quoted 5 lines

cheers,
Gerd