Skip to contentSkip to conversation

hppa CPU reset and speedup

PATCH Superseded Philippe Mathieu-Daudé 2 participants 15 messages Dec 30, 2024 – Dec 31, 2024

v3 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

v3 cover hppa CPU reset and speedup

Respin of:
https://lore.kernel.org/qemu-devel/[email protected]/
"Add CPU reset function and speed up runtime and translation."

- Remove hppa_cpu_init()
- Reset PSW using M bit (untested)

Helge, could we add a functional test booting Linux?

Helge Deller (4):
target/hppa: Convert hppa_cpu_init() to ResetHold handler hw/hppa: Reset vCPUs calling resettable_reset() target/hppa: Set PC on vCPU reset
target/hppa: Speed up hppa_is_pa20()

Philippe Mathieu-Daudé (1):
target/hppa: Only set PSW 'M' bit on reset

 target/hppa/cpu.h | 11 +++++++++--
 hw/hppa/machine.c |  6 +++---
 target/hppa/cpu.c | 20 +++++++++++++++++---
 3 files changed, 29 insertions(+), 8 deletions(-)
Signature · 1 line
2.47.1
#1 lore raw

PATCH v3 1/5 target/hppa: Convert hppa_cpu_init() to ResetHold handler

From: Helge Deller <[email protected]>

hppa_cpu_initfn() is called once when a HPPA CPU instance is initialized, but it sets fields which should be set each time a CPU resets. Rename it as a reset handler, having it matching the ResettablePhases::hold() signature, and register it as ResettableClass handler.

Since on reset the CPU registers and TLB entries are expected to be zero, add a memset() call clearing CPUHPPAState up to the &end_reset_fields marker.

---
 target/hppa/cpu.h |  5 +++++
 target/hppa/cpu.c | 14 ++++++++++++--
 2 files changed, 17 insertions(+), 2 deletions(-)
target/hppa/cpu.h +5 -0
index 22a6510e087..c1d69c1a835 100644
@@ -263,6 +263,9 @@ typedef struct CPUArchState {
     IntervalTreeRoot tlb_root;
 
     HPPATLBEntry tlb[HPPA_TLB_ENTRIES];
+
+    /* Fields up to this point are cleared by a CPU reset */
+    struct {} end_reset_fields;
 } CPUHPPAState;
 
 /**
@@ -281,6 +284,7 @@ struct ArchCPU {
 /**
  * HPPACPUClass:
  * @parent_realize: The parent class' realize handler.
+ * @parent_phases: The parent class' reset phase handlers.
  *
  * An HPPA CPU model.
  */
@@ -288,6 +292,7 @@ struct HPPACPUClass {
     CPUClass parent_class;
 
     DeviceRealize parent_realize;
+    ResettablePhases parent_phases;
 };
 
 #include "exec/cpu-all.h"
target/hppa/cpu.c +12 -2
index 47d01609557..d784bcdd602 100644
@@ -193,13 +193,20 @@ static void hppa_cpu_realizefn(DeviceState *dev, Error **errp)
     tcg_cflags_set(cs, CF_PCREL);
 }
 
-static void hppa_cpu_initfn(Object *obj)
+static void hppa_cpu_reset_hold(Object *obj, ResetType type)
 {
+    HPPACPUClass *scc = HPPA_CPU_GET_CLASS(obj);
     CPUState *cs = CPU(obj);
     HPPACPU *cpu = HPPA_CPU(obj);
     CPUHPPAState *env = &cpu->env;
 
+    if (scc->parent_phases.hold) {
+        scc->parent_phases.hold(obj, type);
+    }
     cs->exception_index = -1;
+
+    memset(env, 0, offsetof(CPUHPPAState, end_reset_fields));
+
     cpu_hppa_loaded_fr0(env);
     cpu_hppa_put_psw(env, PSW_W);
 }
@@ -242,10 +249,14 @@ static void hppa_cpu_class_init(ObjectClass *oc, void *data)
     DeviceClass *dc = DEVICE_CLASS(oc);
     CPUClass *cc = CPU_CLASS(oc);
     HPPACPUClass *acc = HPPA_CPU_CLASS(oc);
+    ResettableClass *rc = RESETTABLE_CLASS(oc);
 
     device_class_set_parent_realize(dc, hppa_cpu_realizefn,
                                     &acc->parent_realize);
 
+    resettable_class_set_parent_phases(rc, NULL, hppa_cpu_reset_hold, NULL,
+                                       &acc->parent_phases);
+
     cc->class_by_name = hppa_cpu_class_by_name;
     cc->has_work = hppa_cpu_has_work;
     cc->mmu_index = hppa_cpu_mmu_index;
@@ -269,7 +280,6 @@ static const TypeInfo hppa_cpu_type_infos[] = {
         .parent = TYPE_CPU,
         .instance_size = sizeof(HPPACPU),
         .instance_align = __alignof(HPPACPU),
-        .instance_init = hppa_cpu_initfn,
         .abstract = false,
         .class_size = sizeof(HPPACPUClass),
         .class_init = hppa_cpu_class_init,
Signature · 1 line
2.47.1
#2 lore raw

PATCH v3 2/5 hw/hppa: Reset vCPUs calling resettable_reset()

In reply to Philippe Mathieu-Daudé

From: Helge Deller <[email protected]>

Rather than manually (and incompletely) resetting vCPUs, call resettable_reset() which will fully reset the vCPUs. Remove redundant assignations.

---
 hw/hppa/machine.c | 6 +++---
 target/hppa/cpu.c | 1 +
 2 files changed, 4 insertions(+), 3 deletions(-)
hw/hppa/machine.c +3 -3
index 65259308e2e..8230f43e41c 100644
@@ -655,12 +655,12 @@ static void hppa_machine_reset(MachineState *ms, ResetType type)
     for (i = 0; i < smp_cpus; i++) {
         CPUState *cs = CPU(cpu[i]);
 
+        /* reset CPU */
+        resettable_reset(OBJECT(cs), RESET_TYPE_COLD);
+
         cpu_set_pc(cs, firmware_entry);
         cpu[i]->env.psw = PSW_Q;
         cpu[i]->env.gr[5] = CPU_HPA + i * 0x1000;
-
-        cs->exception_index = -1;
-        cs->halted = 0;
     }
 
     /* already initialized by machine_hppa_init()? */
target/hppa/cpu.c +1 -0
index d784bcdd602..41538d39d62 100644
@@ -204,6 +204,7 @@ static void hppa_cpu_reset_hold(Object *obj, ResetType type)
         scc->parent_phases.hold(obj, type);
     }
     cs->exception_index = -1;
+    cs->halted = 0;
 
     memset(env, 0, offsetof(CPUHPPAState, end_reset_fields));
 
Signature · 1 line
2.47.1
#3 lore raw

PATCH v3 3/5 target/hppa: Only set PSW 'M' bit on reset

In reply to Philippe Mathieu-Daudé

On reset:

"All PSW bits except the M bit is reset. The M bit is set."

Commit 1a19da0da44 ("target/hppa: Fill in hppa_cpu_do_interrupt / hppa_cpu_exec_interrupt") inadvertently set the W bit at RESET, remove it and set the M bit.

---
Only tested using QEMU mainstream test suite, IOW not much.

---
 target/hppa/cpu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
target/hppa/cpu.c +1 -1
index 41538d39d62..dbd46842841 100644
@@ -209,7 +209,7 @@ static void hppa_cpu_reset_hold(Object *obj, ResetType type)
     memset(env, 0, offsetof(CPUHPPAState, end_reset_fields));
 
     cpu_hppa_loaded_fr0(env);
-    cpu_hppa_put_psw(env, PSW_W);
+    cpu_hppa_put_psw(env, PSW_M);
 }
 
 static ObjectClass *hppa_cpu_class_by_name(const char *cpu_model)
Signature · 1 line
2.47.1
#4 lore raw

PATCH v3 4/5 target/hppa: Set PC on vCPU reset

In reply to Philippe Mathieu-Daudé

From: Helge Deller <[email protected]>

On reset:

"The CPU begins fetching instructions from address 0xf0000004.
This address is in PDC space."

Switch vCPUs to 32-bit mode (PSW_W bit is not set) and start execution at address 0xf0000004.

---
 target/hppa/cpu.c | 1 +
 1 file changed, 1 insertion(+)
target/hppa/cpu.c +1 -0
index dbd46842841..7278b7ca6b5 100644
@@ -205,6 +205,7 @@ static void hppa_cpu_reset_hold(Object *obj, ResetType type)
     }
     cs->exception_index = -1;
     cs->halted = 0;
+    cpu_set_pc(cs, 0xf0000004);
 
     memset(env, 0, offsetof(CPUHPPAState, end_reset_fields));
 
Signature · 1 line
2.47.1
#5 lore raw

PATCH v3 5/5 target/hppa: Speed up hppa_is_pa20()

In reply to Philippe Mathieu-Daudé

From: Helge Deller <[email protected]>

Although the hppa_is_pa20() helper is costly due to string comparisons in object_dynamic_cast(), it is called quite often during memory lookups and at each start of a block of instruction translations. Speed hppa_is_pa20() up by calling object_dynamic_cast() only once at CPU creation and store the result in the is_pa20 of struct CPUArchState.

---
 target/hppa/cpu.h | 6 ++++--
 target/hppa/cpu.c | 2 ++
 2 files changed, 6 insertions(+), 2 deletions(-)
target/hppa/cpu.h +4 -2
index c1d69c1a835..083d4f5a56a 100644
@@ -266,6 +266,8 @@ typedef struct CPUArchState {
 
     /* Fields up to this point are cleared by a CPU reset */
     struct {} end_reset_fields;
+
+    bool is_pa20;
 } CPUHPPAState;
 
 /**
@@ -297,9 +299,9 @@ struct HPPACPUClass {
 
 #include "exec/cpu-all.h"
 
-static inline bool hppa_is_pa20(CPUHPPAState *env)
+static inline bool hppa_is_pa20(const CPUHPPAState *env)
 {
-    return object_dynamic_cast(OBJECT(env_cpu(env)), TYPE_HPPA64_CPU) != NULL;
+    return env->is_pa20;
 }
 
 static inline int HPPA_BTLB_ENTRIES(CPUHPPAState *env)
target/hppa/cpu.c +2 -0
index 7278b7ca6b5..6e5434a8e99 100644
@@ -209,6 +209,8 @@ static void hppa_cpu_reset_hold(Object *obj, ResetType type)
 
     memset(env, 0, offsetof(CPUHPPAState, end_reset_fields));
 
+    env->is_pa20 = !!object_dynamic_cast(obj, TYPE_HPPA64_CPU);
+
     cpu_hppa_loaded_fr0(env);
     cpu_hppa_put_psw(env, PSW_M);
 }
Signature · 1 line
2.47.1
#6 lore raw
In reply to Philippe Mathieu-Daudé

Hi Philippe,

Quoted 5 lines · Philippe Mathieu-Daudé

Respin of:
https://lore.kernel.org/qemu-devel/[email protected]/
"Add CPU reset function and speed up runtime and translation."

- Remove hppa_cpu_init()

Thanks for picking up my patches and integrating them properly with the reset code.
But sadly your changes somehow break hppa 64-bit CPU support. I think it's to when the reset code is called.

Easy reproducer (no other options/disc/network needed):
./qemu-system-hppa -smp cpus=4 -nographic -machine C3700

Quoted 1 line

- Reset PSW using M bit (untested)

We haven't implemented PSW-M support and the first thing the firmware does is to reprogram PSW. So, basically it's not needed.

Quoted 1 line

Helge, could we add a functional test booting Linux?

What exactly are you looking for?
Some trivial preinstalled disc image with kernels? Any examples?

Helge

Quoted 14 lines

Helge Deller (4):
target/hppa: Convert hppa_cpu_init() to ResetHold handler hw/hppa: Reset vCPUs calling resettable_reset() target/hppa: Set PC on vCPU reset
target/hppa: Speed up hppa_is_pa20()

Philippe Mathieu-Daudé (1):
target/hppa: Only set PSW 'M' bit on reset

  target/hppa/cpu.h | 11 +++++++++--
  hw/hppa/machine.c |  6 +++---
  target/hppa/cpu.c | 20 +++++++++++++++++---
  3 files changed, 29 insertions(+), 8 deletions(-)
#7 lore raw
Quoted 9 lines · Helge Deller

Hi Philippe,

Quoted 5 lines · Philippe Mathieu-Daudé

Respin of:
https://lore.kernel.org/qemu-devel/20241229234154.32250-1- [email protected]/ "Add CPU reset function and speed up runtime and translation."

- Remove hppa_cpu_init()

Quoted 7 lines

Thanks for picking up my patches and integrating them properly with the reset code.
But sadly your changes somehow break hppa 64-bit CPU support. I think it's to when the reset code is called.

Easy reproducer (no other options/disc/network needed):
./qemu-system-hppa -smp cpus=4 -nographic -machine C3700

[*]

Quoted 7 lines
Quoted 2 lines

- Reset PSW using M bit (untested)

We haven't implemented PSW-M support and the first thing the firmware does is to reprogram PSW. So, basically it's not needed.

Quoted 5 lines
Quoted 1 line

Helge, could we add a functional test booting Linux?

What exactly are you looking for?
Some trivial preinstalled disc image with kernels? Any examples?

Your [*] example running SeaBIOS is perfect.

See tests/functional/test_riscv_opensbi.py, but using:

wait_for_console_pattern('SeaBIOS PA-RISC 64-bit Firmware')

Quoted 18 lines

Helge

Quoted 14 lines

Helge Deller (4):
   target/hppa: Convert hppa_cpu_init() to ResetHold handler hw/hppa: Reset vCPUs calling resettable_reset() target/hppa: Set PC on vCPU reset
   target/hppa: Speed up hppa_is_pa20()

Philippe Mathieu-Daudé (1):
   target/hppa: Only set PSW 'M' bit on reset

  target/hppa/cpu.h | 11 +++++++++--
  hw/hppa/machine.c |  6 +++---
  target/hppa/cpu.c | 20 +++++++++++++++++---
  3 files changed, 29 insertions(+), 8 deletions(-)
#8 lore raw
Quoted 33 lines · Philippe Mathieu-Daudé
Quoted 9 lines · Helge Deller

Hi Philippe,

Quoted 5 lines · Philippe Mathieu-Daudé

Respin of:
https://lore.kernel.org/qemu-devel/20241229234154.32250-1- [email protected]/ "Add CPU reset function and speed up runtime and translation."

- Remove hppa_cpu_init()

Quoted 7 lines

Thanks for picking up my patches and integrating them properly with the reset code.
But sadly your changes somehow break hppa 64-bit CPU support. I think it's to when the reset code is called.

Easy reproducer (no other options/disc/network needed):
./qemu-system-hppa -smp cpus=4 -nographic -machine C3700

[*]

Quoted 7 lines
Quoted 2 lines

- Reset PSW using M bit (untested)

We haven't implemented PSW-M support and the first thing the firmware does is to reprogram PSW. So, basically it's not needed.

Quoted 5 lines
Quoted 1 line

Helge, could we add a functional test booting Linux?

What exactly are you looking for?
Some trivial preinstalled disc image with kernels? Any examples?

Quoted 5 lines

Your [*] example running SeaBIOS is perfect.

See tests/functional/test_riscv_opensbi.py, but using:

  wait_for_console_pattern('SeaBIOS PA-RISC 64-bit Firmware')

Then I suggest 2 testcases (32-bit and 64-bit CPU, dropped SMP flag):
./qemu-system-hppa -nographic -machine B160L ./qemu-system-hppa -nographic -machine C3700

Helge

Quoted 20 lines
Quoted 19 lines

Helge

Quoted 14 lines

Helge Deller (4):
   target/hppa: Convert hppa_cpu_init() to ResetHold handler hw/hppa: Reset vCPUs calling resettable_reset() target/hppa: Set PC on vCPU reset
   target/hppa: Speed up hppa_is_pa20()

Philippe Mathieu-Daudé (1):
   target/hppa: Only set PSW 'M' bit on reset

  target/hppa/cpu.h | 11 +++++++++--
  hw/hppa/machine.c |  6 +++---
  target/hppa/cpu.c | 20 +++++++++++++++++---
  3 files changed, 29 insertions(+), 8 deletions(-)
#9 lore raw
In reply to Helge Deller
Quoted 7 lines · Helge Deller

Hi Philippe,

Quoted 4 lines · Philippe Mathieu-Daudé

Respin of:
https://lore.kernel.org/qemu-devel/20241229234154.32250-1- [email protected]/ "Add CPU reset function and speed up runtime and translation."

Quoted 6 lines
Quoted 1 line

- Remove hppa_cpu_init()

Thanks for picking up my patches and integrating them properly with the reset code.
But sadly your changes somehow break hppa 64-bit CPU support. I think it's to when the reset code is called.

Likely hppa_is_pa20() called *before* CPU reset...?

Quoted 3 lines

Easy reproducer (no other options/disc/network needed):
./qemu-system-hppa -smp cpus=4 -nographic -machine C3700

#10 lore raw
Quoted 8 lines · Philippe Mathieu-Daudé
Quoted 7 lines · Helge Deller

Hi Philippe,

Quoted 4 lines · Philippe Mathieu-Daudé

Respin of:
https://lore.kernel.org/qemu-devel/20241229234154.32250-1- [email protected]/ "Add CPU reset function and speed up runtime and translation."

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

- Remove hppa_cpu_init()

Thanks for picking up my patches and integrating them properly with the reset code.
But sadly your changes somehow break hppa 64-bit CPU support. I think it's to when the reset code is called.

Likely hppa_is_pa20() called *before* CPU reset...?

Indeed:

(lldb) bt
* thread #1, queue = 'com.apple.main-thread', stop reason = breakpoint 1.10
* frame #0: 0x000000010024df30 qemu-system-hppa`hppa_ptlbe [inlined]
hppa_is_pa20(env=0x000000011f822230) at cpu.h:304:17

    frame #1: 0x000000010024df30 qemu-system-hppa`hppa_ptlbe [inlined] 
HPPA_BTLB_ENTRIES(env=0x000000011f822230) at cpu.h:309:12
    frame #2: 0x000000010024df30 
qemu-system-hppa`hppa_ptlbe(env=0x000000011f822230) at mem_helper.c:642:29
    frame #3: 0x0000000100250564 
qemu-system-hppa`hppa_cpu_realizefn(dev=0x000000011f81fa00, errp=<unavailable>) at cpu.c:188:9
    frame #4: 0x00000001002f812c 

qemu-system-hppa`device_set_realized(obj=<unavailable>, value=<unavailable>, errp=0x000000016fdfebc0) at qdev.c:495:13

Quoted 4 lines
Quoted 4 lines

Easy reproducer (no other options/disc/network needed):
./qemu-system-hppa -smp cpus=4 -nographic -machine C3700

#11 lore raw
Quoted 20 lines · Philippe Mathieu-Daudé
Quoted 8 lines · Philippe Mathieu-Daudé
Quoted 7 lines · Helge Deller

Hi Philippe,

Quoted 4 lines · Philippe Mathieu-Daudé

Respin of:
https://lore.kernel.org/qemu-devel/20241229234154.32250-1- [email protected]/ "Add CPU reset function and speed up runtime and translation."

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

- Remove hppa_cpu_init()

Thanks for picking up my patches and integrating them properly with the reset code.
But sadly your changes somehow break hppa 64-bit CPU support. I think it's to when the reset code is called.

Likely hppa_is_pa20() called *before* CPU reset...?

Indeed:

Quoted 12 lines

(lldb) bt
* thread #1, queue = 'com.apple.main-thread', stop reason = breakpoint 1.10
  * frame #0: 0x000000010024df30 qemu-system-hppa`hppa_ptlbe [inlined]
hppa_is_pa20(env=0x000000011f822230) at cpu.h:304:17
    frame #1: 0x000000010024df30 qemu-system-hppa`hppa_ptlbe [inlined]
HPPA_BTLB_ENTRIES(env=0x000000011f822230) at cpu.h:309:12
    frame #2: 0x000000010024df30 qemu-system-
hppa`hppa_ptlbe(env=0x000000011f822230) at mem_helper.c:642:29
    frame #3: 0x0000000100250564 qemu-system-
hppa`hppa_cpu_realizefn(dev=0x000000011f81fa00, errp=<unavailable>) at cpu.c:188:9
    frame #4: 0x00000001002f812c qemu-system-
hppa`device_set_realized(obj=<unavailable>, value=<unavailable>, errp=0x000000016fdfebc0) at qdev.c:495:13

Assigning is_pa20 in instance_init() fixes it:

-- >8 --

target/hppa/cpu.c +8 -2
index 6e5434a8e99..b0bc9d35e4c 100644
@@ -193,6 +193,13 @@ static void hppa_cpu_realizefn(DeviceState *dev, Error **errp)
     tcg_cflags_set(cs, CF_PCREL);
 }
 
+static void hppa_cpu_initfn(Object *obj)
+{
+    CPUHPPAState *env = cpu_env(CPU(obj));
+
+    env->is_pa20 = !!object_dynamic_cast(obj, TYPE_HPPA64_CPU);
+}
+
 static void hppa_cpu_reset_hold(Object *obj, ResetType type)
 {
     HPPACPUClass *scc = HPPA_CPU_GET_CLASS(obj);
@@ -209,8 +216,6 @@ static void hppa_cpu_reset_hold(Object *obj, ResetType type)
 
     memset(env, 0, offsetof(CPUHPPAState, end_reset_fields));
 
-    env->is_pa20 = !!object_dynamic_cast(obj, TYPE_HPPA64_CPU);
-
     cpu_hppa_loaded_fr0(env);
     cpu_hppa_put_psw(env, PSW_M);
 }
@@ -284,6 +289,7 @@ static const TypeInfo hppa_cpu_type_infos[] = {
         .parent = TYPE_CPU,
         .instance_size = sizeof(HPPACPU),
         .instance_align = __alignof(HPPACPU),
+        .instance_init = hppa_cpu_initfn,
         .abstract = false,
         .class_size = sizeof(HPPACPUClass),
         .class_init = hppa_cpu_class_init,

---

#12 lore raw
In reply to Helge Deller
Quoted 13 lines · Helge Deller

Hi Philippe,

Quoted 5 lines · Philippe Mathieu-Daudé

Respin of:
https://lore.kernel.org/qemu-devel/20241229234154.32250-1- [email protected]/ "Add CPU reset function and speed up runtime and translation."

- Remove hppa_cpu_init()

Thanks for picking up my patches and integrating them properly with the reset code.

Quoted 5 lines
Quoted 1 line

- Reset PSW using M bit (untested)

We haven't implemented PSW-M support and the first thing the firmware does is to reprogram PSW. So, basically it's not needed.

Good to know it isn't needed yet (I grepped and noticed very few uses). Are you OK with the patch as is (as it matches the spec)?

#13 lore raw
Quoted 16 lines · Philippe Mathieu-Daudé
Quoted 15 lines · Helge Deller

Hi Philippe,

Quoted 5 lines · Philippe Mathieu-Daudé

Respin of:
https://lore.kernel.org/qemu-devel/20241229234154.32250-1- [email protected]/ "Add CPU reset function and speed up runtime and translation."

- Remove hppa_cpu_init()

Thanks for picking up my patches and integrating them properly with the reset code.

Quoted 1 line

- Reset PSW using M bit (untested)

Quoted 7 lines
Quoted 3 lines

We haven't implemented PSW-M support and the first thing the firmware does is to reprogram PSW. So, basically it's not needed.

Good to know it isn't needed yet (I grepped and noticed very few uses). Are you OK with the patch as is (as it matches the spec)?

At least I don't have any objections as long as 32- and 64-bit CPUs still boot.

Will you respin the series, then I can test and review?

Helge

#14 lore raw
Quoted 21 lines · Helge Deller
Quoted 20 lines · Philippe Mathieu-Daudé
Quoted 18 lines · Helge Deller

Hi Philippe,

Quoted 5 lines · Philippe Mathieu-Daudé

Respin of:
https://lore.kernel.org/qemu-devel/20241229234154.32250-1- [email protected]/ "Add CPU reset function and speed up runtime and translation."

- Remove hppa_cpu_init()

Thanks for picking up my patches and integrating them properly with the reset code.

Quoted 1 line

- Reset PSW using M bit (untested)

We haven't implemented PSW-M support and the first thing the firmware does is to reprogram PSW. So, basically it's not needed.

Quoted 8 lines
Quoted 3 lines

Good to know it isn't needed yet (I grepped and noticed very few uses). Are you OK with the patch as is (as it matches the spec)?

At least I don't have any objections as long as 32- and 64-bit CPUs still boot.

Will you respin the series, then I can test and review?

OK!