Skip to contentSkip to conversation

Revert "powerpc/powernv/idle: Replace CPU feature check with PVR check"

PATCH Applied Pratik Rajesh Sampat 5 participants 6 messages Aug 26, 2020 – Aug 27, 2020 PowerPC

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Revert "powerpc/powernv/idle: Replace CPU feature check with PVR check"

Cpuidle stop state implementation has minor optimizations for P10 where hardware preserves more SPR registers compared to P9. The current P9 driver works for P10, although does few extra save-restores. P9 driver can provide the required power management features like SMT thread folding and core level power savings on a P10 platform.

Until the P10 stop driver is available, revert the commit which allows for only P9 systems to utilize cpuidle and blocks all idle stop states for P10.
Cpu idle states are enabled and tested on the P10 platform with this fix.

This reverts commit 8747bf36f312356f8a295a0c39ff092d65ce75ae.

Fixes 8747bf36f312 ("powerpc/powernv/idle: Replace CPU feature check with PVR check")

---
@mpe: This revert would resolve a staging issue wherein the P10 stop driver is not yet ready while cpuidle stop states need not be blocked on 5.9 for Power10 systems which could cause SMT folding related performance issues.

The P10 stop driver is in the works here:
https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-August/216773.html

 arch/powerpc/platforms/powernv/idle.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/powerpc/platforms/powernv/idle.c +1 -1
index 77513a80cef9..345ab062b21a 100644
@@ -1223,7 +1223,7 @@ static void __init pnv_probe_idle_states(void)
 		return;
 	}
 
-	if (pvr_version_is(PVR_POWER9))
+	if (cpu_has_feature(CPU_FTR_ARCH_300))
 		pnv_power9_idle_init();
 
 	for (i = 0; i < nr_pnv_idle_states; i++)
Signature · 1 line
2.25.4
#1 lore raw

Le 26/08/2020 à 10:29, Pratik Rajesh Sampat a écrit :

Quoted 34 lines

Cpuidle stop state implementation has minor optimizations for P10 where hardware preserves more SPR registers compared to P9. The current P9 driver works for P10, although does few extra save-restores. P9 driver can provide the required power management features like SMT thread folding and core level power savings on a P10 platform.

Until the P10 stop driver is available, revert the commit which allows for only P9 systems to utilize cpuidle and blocks all idle stop states for P10.
Cpu idle states are enabled and tested on the P10 platform with this fix.

This reverts commit 8747bf36f312356f8a295a0c39ff092d65ce75ae.

Fixes 8747bf36f312 ("powerpc/powernv/idle: Replace CPU feature check with PVR check")

---
@mpe: This revert would resolve a staging issue wherein the P10 stop driver is not yet ready while cpuidle stop states need not be blocked on 5.9 for Power10 systems which could cause SMT folding related performance issues.

The P10 stop driver is in the works here:
https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-August/216773.html

  arch/powerpc/platforms/powernv/idle.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
arch/powerpc/platforms/powernv/idle.c +0 -0
index 77513a80cef9..345ab062b21a 100644

arch/powerpc/platforms/powernv/idle.c static void __init pnv_probe_idle_states(void) · line 1226

Quoted 5 lines
arch/powerpc/platforms/powernv/idle.c +1 -1
@@ -1223,7 +1223,7 @@ static void __init pnv_probe_idle_states(void)
  		return;
  	}
 
-	if (pvr_version_is(PVR_POWER9))
+	if (cpu_has_feature(CPU_FTR_ARCH_300))

Why not something like:

	if (pvr_version_is(PVR_POWER9) || pvr_version_is(PVR_POWER10))
		pnv_power9_idle_init();

arch/powerpc/platforms/powernv/idle.c static void __init pnv_probe_idle_states(void) · line 1229

Quoted 4 lines
arch/powerpc/platforms/powernv/idle.c +0 -0
  		pnv_power9_idle_init();
 
  	for (i = 0; i < nr_pnv_idle_states; i++)

Christophe

#2 lore raw
Quoted 44 lines · Christophe Leroy

Le 26/08/2020 à 10:29, Pratik Rajesh Sampat a écrit :

Quoted 34 lines

Cpuidle stop state implementation has minor optimizations for P10 where hardware preserves more SPR registers compared to P9. The current P9 driver works for P10, although does few extra save-restores. P9 driver can provide the required power management features like SMT thread folding and core level power savings on a P10 platform.

Until the P10 stop driver is available, revert the commit which allows for only P9 systems to utilize cpuidle and blocks all idle stop states for P10.
Cpu idle states are enabled and tested on the P10 platform with this fix.

This reverts commit 8747bf36f312356f8a295a0c39ff092d65ce75ae.

Fixes 8747bf36f312 ("powerpc/powernv/idle: Replace CPU feature check with PVR check")

---
  @mpe: This revert would resolve a staging issue wherein the P10 stop driver is not yet ready while cpuidle stop states need not be blocked on 5.9 for Power10 systems which could cause SMT folding related performance issues.

  The P10 stop driver is in the works here:
https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-August/216773.html

  arch/powerpc/platforms/powernv/idle.c | 2 +-

  1 file changed, 1 insertion(+), 1 deletion(-)
arch/powerpc/platforms/powernv/idle.c +0 -0
index 77513a80cef9..345ab062b21a 100644

arch/powerpc/platforms/powernv/idle.c static void __init pnv_probe_idle_states(void) · line 1226

Quoted 4 lines
arch/powerpc/platforms/powernv/idle.c +1 -0
@@ -1223,7 +1223,7 @@ static void __init pnv_probe_idle_states(void)
           return;
       }
   -    if (pvr_version_is(PVR_POWER9))
+    if (cpu_has_feature(CPU_FTR_ARCH_300))

Why not something like:

Quoted 2 lines

    if (pvr_version_is(PVR_POWER9) || pvr_version_is(PVR_POWER10)) pnv_power9_idle_init();

In order to use PVR_POWER10 I would need to define it under arch/powerpc/include/asm/reg.h, which is not present in 5.9 yet.

However, if it okay with @mpe I could split out Nick's P10 stop driver
(https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-August/216773.html)
into two parts:
1. This could include minimal code to introduce the P10 PVR and the stop wrappers for it. Although this patch internally still calls into the P9 path. This should gracefully fix the issue.
2. Then later in this patch we could introduce the p10 callback methods as they are in Nick's series.

---
Thanks
Pratik

Quoted 6 lines

arch/powerpc/platforms/powernv/idle.c static void __init pnv_probe_idle_states(void)

Quoted 4 lines
arch/powerpc/platforms/powernv/idle.c +0 -0
           pnv_power9_idle_init();
         for (i = 0; i < nr_pnv_idle_states; i++)
 

Christophe

#3 lore raw
Quoted 48 lines · Pratik Sampat
Quoted 44 lines · Christophe Leroy

Le 26/08/2020 à 10:29, Pratik Rajesh Sampat a écrit :

Quoted 36 lines

Cpuidle stop state implementation has minor optimizations for P10 where hardware preserves more SPR registers compared to P9. The current P9 driver works for P10, although does few extra save-restores. P9 driver can provide the required power management features like SMT thread folding and core level power savings on a P10 platform.

Until the P10 stop driver is available, revert the commit which allows for only P9 systems to utilize cpuidle and blocks all idle stop states for P10.
Cpu idle states are enabled and tested on the P10 platform with this fix.

This reverts commit 8747bf36f312356f8a295a0c39ff092d65ce75ae.

Fixes 8747bf36f312 ("powerpc/powernv/idle: Replace CPU feature check

with PVR check")

---
  @mpe: This revert would resolve a staging issue wherein the P10 stop driver is not yet ready while cpuidle stop states need not be blocked on 5.9 for Power10 systems which could cause SMT folding related performance issues.

  The P10 stop driver is in the works here:
https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-August/216773.html

  arch/powerpc/platforms/powernv/idle.c | 2 +-

  1 file changed, 1 insertion(+), 1 deletion(-)
arch/powerpc/platforms/powernv/idle.c +0 -0
index 77513a80cef9..345ab062b21a 100644

arch/powerpc/platforms/powernv/idle.c static void __init pnv_probe_idle_states(void) · line 1226

Quoted 4 lines
arch/powerpc/platforms/powernv/idle.c +1 -0
@@ -1223,7 +1223,7 @@ static void __init pnv_probe_idle_states(void)
           return;
       }
   -    if (pvr_version_is(PVR_POWER9))
+    if (cpu_has_feature(CPU_FTR_ARCH_300))

Why not something like:

Quoted 2 lines

    if (pvr_version_is(PVR_POWER9) || pvr_version_is(PVR_POWER10)) pnv_power9_idle_init();

Quoted 6 lines

In order to use PVR_POWER10 I would need to define it under arch/powerpc/include/asm/reg.h, which is not present in 5.9 yet.

However, if it okay with @mpe I could split out Nick's P10 stop driver
(https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-August/216773.html)
into two parts:

I'll just take this for now, it's the simplest option.

cheers

Reviewed-by on v1 1/1
#4 lore raw
In reply to Pratik Rajesh Sampat
Quoted 13 lines · Pratik Rajesh Sampat

Cpuidle stop state implementation has minor optimizations for P10 where hardware preserves more SPR registers compared to P9. The current P9 driver works for P10, although does few extra save-restores. P9 driver can provide the required power management features like SMT thread folding and core level power savings on a P10 platform.

Until the P10 stop driver is available, revert the commit which allows for only P9 systems to utilize cpuidle and blocks all idle stop states for P10.
Cpu idle states are enabled and tested on the P10 platform with this fix.

Quoted 4 lines

This reverts commit 8747bf36f312356f8a295a0c39ff092d65ce75ae.

Fixes 8747bf36f312 ("powerpc/powernv/idle: Replace CPU feature check with PVR check")
Quoted 17 lines

---
@mpe: This revert would resolve a staging issue wherein the P10 stop driver is not yet ready while cpuidle stop states need not be blocked on 5.9 for Power10 systems which could cause SMT folding related performance issues.

The P10 stop driver is in the works here:
https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-August/216773.html

 arch/powerpc/platforms/powernv/idle.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/powerpc/platforms/powernv/idle.c +0 -0
index 77513a80cef9..345ab062b21a 100644

arch/powerpc/platforms/powernv/idle.c static void __init pnv_probe_idle_states(void) · line 1229

Quoted 8 lines
arch/powerpc/platforms/powernv/idle.c +1 -1
@@ -1223,7 +1223,7 @@ static void __init pnv_probe_idle_states(void)
 		return;
 	}
 
-	if (pvr_version_is(PVR_POWER9))
+	if (cpu_has_feature(CPU_FTR_ARCH_300))
 		pnv_power9_idle_init();
 
 	for (i = 0; i < nr_pnv_idle_states; i++)

This revert solves the stated problem and makes kernel v5.9 work reasonable well on P10 with stop states which are required for SMT mode changes.

Complete P10 driver has been in the works and will build on this fix and complete the required platform support and optimizations.

--Vaidy

#5 lore raw
In reply to Pratik Rajesh Sampat
Quoted 8 lines · Pratik Rajesh Sampat

Cpuidle stop state implementation has minor optimizations for P10 where hardware preserves more SPR registers compared to P9. The current P9 driver works for P10, although does few extra save-restores. P9 driver can provide the required power management features like SMT thread folding and core level power savings on a P10 platform.

[...]

Applied to powerpc/fixes.

[1/1] Revert "powerpc/powernv/idle: Replace CPU feature check with PVR check"

      https://git.kernel.org/powerpc/c/16d83a540ca4e7f1ebb2b3756869b77451d31414

cheers