Skip to contentSkip to conversation

MIPS: perf: Add hardware perf events support for new Loongson-3

PATCH Superseded Huacai Chen 2 participants 6 messages Apr 29, 2020 – Apr 30, 2020

Subsystems (3)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 MIPS: perf: Add hardware perf events support for new Loongson-3

New Loongson-3 means Loongson-3A R2 (Loongson-3A2000) and newer CPUs. Loongson-3 processors have three types of PMU types (so there are three event maps): Loongson-3A1000/Loonngson-3B1000/Loongson-3B1500 is Type-1, Loongson-3A2000/Loongson-3A3000 is Type-2, Loongson-3A4000+ is Type-3.

---
 .../asm/mach-loongson64/cpu-feature-overrides.h    |   1 +
 arch/mips/kernel/perf_event_mipsxx.c               | 358 +++++++++++++++++++--
 2 files changed, 330 insertions(+), 29 deletions(-)
arch/mips/include/asm/mach-loongson64/cpu-feature-overrides.h +1 -0
index 4fab38c..b6e9c99 100644
@@ -48,5 +48,6 @@
 #define cpu_hwrena_impl_bits	0xc0000000
 #define cpu_has_mac2008_only	1
 #define cpu_has_mips_r2_exec_hazard 0
+#define cpu_has_perf_cntr_intr_bit	0
 
 #endif /* __ASM_MACH_LOONGSON64_CPU_FEATURE_OVERRIDES_H */
arch/mips/kernel/perf_event_mipsxx.c +329 -29

503 lines not on this page · raw message

Signature · 1 line
2.7.0
#1 lore raw
Quoted 9 lines · Huacai Chen

New Loongson-3 means Loongson-3A R2 (Loongson-3A2000) and newer CPUs. Loongson-3 processors have three types of PMU types (so there are three event maps): Loongson-3A1000/Loonngson-3B1000/Loongson-3B1500 is Type-1, Loongson-3A2000/Loongson-3A3000 is Type-2, Loongson-3A4000+ is Type-3.

---
 .../asm/mach-loongson64/cpu-feature-overrides.h    |   1 +
 arch/mips/kernel/perf_event_mipsxx.c               | 358 +++++++++++++++++++--

checkpatch warns about missing break/fallthrough statement and indention problems, could please look at this ? And as all the new code is only usefull for loongsoon CPUs could you try to only compile it in, if it's enabled for the image ?

Thomas.

Signature · 2 lines
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]
#2 lore raw

Hi, Thomas,

Quoted 12 lines · Thomas Bogendoerfer
Quoted 9 lines · Huacai Chen

New Loongson-3 means Loongson-3A R2 (Loongson-3A2000) and newer CPUs. Loongson-3 processors have three types of PMU types (so there are three event maps): Loongson-3A1000/Loonngson-3B1000/Loongson-3B1500 is Type-1, Loongson-3A2000/Loongson-3A3000 is Type-2, Loongson-3A4000+ is Type-3.

---
 .../asm/mach-loongson64/cpu-feature-overrides.h    |   1 +
 arch/mips/kernel/perf_event_mipsxx.c               | 358 +++++++++++++++++++--
Quoted 4 lines

checkpatch warns about missing break/fallthrough statement and indention problems, could please look at this ? And as all the new code is only usefull for loongsoon CPUs could you try to only compile it in, if it's enabled for the image ?

OK, I will solve the "fallthrough" warnings and indention warnings. But the other problem is I should use #ifdef CONFIG_CPU_LOONGSON64 to guard all new code in this file? I remember that Paul Burton suggest us to avoid use #ifdef and use runtime detection instead.

Quoted 6 lines

Thomas.

--
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea. [ RFC1925, 2.3 ]

Regards,
Huacai

#3 lore raw
Quoted 16 lines · Huacai Chen

Hi, Thomas,

Quoted 12 lines · Thomas Bogendoerfer
Quoted 9 lines · Huacai Chen

New Loongson-3 means Loongson-3A R2 (Loongson-3A2000) and newer CPUs. Loongson-3 processors have three types of PMU types (so there are three event maps): Loongson-3A1000/Loonngson-3B1000/Loongson-3B1500 is Type-1, Loongson-3A2000/Loongson-3A3000 is Type-2, Loongson-3A4000+ is Type-3.

---
 .../asm/mach-loongson64/cpu-feature-overrides.h    |   1 +
 arch/mips/kernel/perf_event_mipsxx.c               | 358 +++++++++++++++++++--
Quoted 8 lines
Quoted 4 lines

checkpatch warns about missing break/fallthrough statement and indention problems, could please look at this ? And as all the new code is only usefull for loongsoon CPUs could you try to only compile it in, if it's enabled for the image ?

OK, I will solve the "fallthrough" warnings and indention warnings. But the other problem is I should use #ifdef CONFIG_CPU_LOONGSON64 to guard all new code in this file? I remember that Paul Burton suggest us to avoid use #ifdef and use runtime detection instead.

if you do the cpu checks for current_cpu_type() == CPU_LOONGSON64 the compiler will eliminate not needed cpu code. Looks like most of the new code is done inside case CPU_LOONGSON64 blocks, so that is fine. Could you check how much size increase you get for a non loongson64 target with your patch ?

Thomas.

Signature · 2 lines
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]
#4 lore raw

Hi, Thomas,

Quoted 27 lines · Thomas Bogendoerfer
Quoted 16 lines · Huacai Chen

Hi, Thomas,

Quoted 12 lines · Thomas Bogendoerfer
Quoted 9 lines · Huacai Chen

New Loongson-3 means Loongson-3A R2 (Loongson-3A2000) and newer CPUs. Loongson-3 processors have three types of PMU types (so there are three event maps): Loongson-3A1000/Loonngson-3B1000/Loongson-3B1500 is Type-1, Loongson-3A2000/Loongson-3A3000 is Type-2, Loongson-3A4000+ is Type-3.

---
 .../asm/mach-loongson64/cpu-feature-overrides.h    |   1 +
 arch/mips/kernel/perf_event_mipsxx.c               | 358 +++++++++++++++++++--
Quoted 8 lines
Quoted 4 lines

checkpatch warns about missing break/fallthrough statement and indention problems, could please look at this ? And as all the new code is only usefull for loongsoon CPUs could you try to only compile it in, if it's enabled for the image ?

OK, I will solve the "fallthrough" warnings and indention warnings. But the other problem is I should use #ifdef CONFIG_CPU_LOONGSON64 to guard all new code in this file? I remember that Paul Burton suggest us to avoid use #ifdef and use runtime detection instead.

Quoted 5 lines

if you do the cpu checks for current_cpu_type() == CPU_LOONGSON64 the compiler will eliminate not needed cpu code. Looks like most of the new code is done inside case CPU_LOONGSON64 blocks, so that is fine. Could you check how much size increase you get for a non loongson64 target with your patch ?

I have build an IP28 kernel (ip28_defconfig+perf):

size of vmlinux with this perf patch: 10913440
size of vmlinux without this perf patch:10913296

readelf -a vmlinux with this perf patch:
Type Offset VirtAddr PhysAddr
FileSiz MemSiz Flags Align
LOAD 0x0000000000001000 0xa800000020004000 0xa800000020004000

                 0x0000000000800c0c 0x0000000000830dc0  RWE    0x1000
  NOTE           0x000000000070d020 0xa800000020710020 0xa800000020710020
                 0x000000000000003c 0x000000000000003c  R      0x4

readelf -a vmlinux without this perf patch:
Type Offset VirtAddr PhysAddr
FileSiz MemSiz Flags Align
LOAD 0x0000000000001000 0xa800000020004000 0xa800000020004000

                 0x0000000000800c0c 0x0000000000830dc0  RWE    0x1000
  NOTE           0x000000000070cff0 0xa80000002070fff0 0xa80000002070fff0
                 0x000000000000003c 0x000000000000003c  R      0x4
Quoted 7 lines

Thomas.

--
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea. [ RFC1925, 2.3 ]

#5 lore raw
Quoted 17 lines · Huacai Chen
Quoted 2 lines · Thomas Bogendoerfer

that is fine. Could you check how much size increase you get for a non loongson64 target with your patch ?

I have build an IP28 kernel (ip28_defconfig+perf):

size of vmlinux with this perf patch: 10913440
size of vmlinux without this perf patch:10913296

readelf -a vmlinux with this perf patch:
Type Offset VirtAddr PhysAddr
FileSiz MemSiz Flags Align
LOAD 0x0000000000001000 0xa800000020004000 0xa800000020004000

                 0x0000000000800c0c 0x0000000000830dc0  RWE    0x1000
  NOTE           0x000000000070d020 0xa800000020710020 0xa800000020710020
                 0x000000000000003c 0x000000000000003c  R      0x4
Quoted 7 lines

readelf -a vmlinux without this perf patch:
Type Offset VirtAddr PhysAddr
FileSiz MemSiz Flags Align
LOAD 0x0000000000001000 0xa800000020004000 0xa800000020004000

                 0x0000000000800c0c 0x0000000000830dc0  RWE    0x1000
  NOTE           0x000000000070cff0 0xa80000002070fff0 0xa80000002070fff0
                 0x000000000000003c 0x000000000000003c  R      0x4

this looks the same, are you sure you used the right images ?

Nevertheless it's not much. I've applied your new patch. Thank you for checking.

Thomas.

Signature · 2 lines
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]