Skip to contentSkip to conversation

arm_mpam: Fix MPAMCFG_MBW_PBM register setting

PATCH Applied Fenghua Yu 4 participants 5 messages Jun 7 – Jul 16 MPAM driver

lore
Share & tools
lore

Fenghua Yu

author reviewer
#0 lore raw

PATCH 1/1 arm_mpam: Fix MPAMCFG_MBW_PBM register setting

MPAMCFG_MBW_PBM is written from cfg if cfg has the MBW partition feature. It is reset when cfg does not have the MBW partition feature.

But the register handling is reversed. This may cause an incorrect register setting. For example, during an MPAM reset, reset_cfg is empty (no MBW partition feature set), and cfg->mbw_pbm is 0. Instead of resetting MPAMCFG_MBW_PBM to all 1's, the current logic will set it to cfg->mbw_pbm, which is 0.

Fix the issue by swapping the if/else branches.

Fixes a1cb6577f575 ("arm_mpam: Reset when feature configuration bit unset")
---
 drivers/resctrl/mpam_devices.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/resctrl/mpam_devices.c +2 -2
index 4b93e89c2678..d8b0383cee92 100644
@@ -1570,9 +1570,9 @@ static void mpam_reprogram_ris_partid(struct mpam_msc_ris *ris, u16 partid,
 
 	if (mpam_has_feature(mpam_feat_mbw_part, rprops)) {
 		if (mpam_has_feature(mpam_feat_mbw_part, cfg))
-			mpam_reset_msc_bitmap(msc, MPAMCFG_MBW_PBM, rprops->mbw_pbm_bits);
-		else
 			mpam_write_partsel_reg(msc, MBW_PBM, cfg->mbw_pbm);
+		else
+			mpam_reset_msc_bitmap(msc, MPAMCFG_MBW_PBM, rprops->mbw_pbm_bits);
 	}
 
 	if (mpam_has_feature(mpam_feat_mbw_min, rprops)) {
Signature · 1 line
2.43.0
Reviewed-by on v1 1/1
#1 lore raw

Hi Fenghua,

Quoted 9 lines · Fenghua Yu

MPAMCFG_MBW_PBM is written from cfg if cfg has the MBW partition feature. It is reset when cfg does not have the MBW partition feature.

But the register handling is reversed. This may cause an incorrect register setting. For example, during an MPAM reset, reset_cfg is empty (no MBW partition feature set), and cfg->mbw_pbm is 0. Instead of resetting MPAMCFG_MBW_PBM to all 1's, the current logic will set it to cfg->mbw_pbm, which is 0.

Quoted 9 lines

Fix the issue by swapping the if/else branches.

Fixes a1cb6577f575 ("arm_mpam: Reset when feature configuration bit unset")
---
  drivers/resctrl/mpam_devices.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)

The fix itself looks reasonable to me, but two questions below.

Quoted 6 lines
drivers/resctrl/mpam_devices.c +0 -0
index 4b93e89c2678..d8b0383cee92 100644
@@ -1570,9 +1570,9 @@ static void mpam_reprogram_ris_partid(struct mpam_msc_ris *ris, u16 partid,
 

drivers/resctrl/mpam_devices.c static void mpam_reprogram_ris_partid(struct mpam_msc_ris *ris, u16 partid, · line 1578

Quoted 10 lines
drivers/resctrl/mpam_devices.c +2 -2
  	if (mpam_has_feature(mpam_feat_mbw_part, rprops)) {
  		if (mpam_has_feature(mpam_feat_mbw_part, cfg))
-			mpam_reset_msc_bitmap(msc, MPAMCFG_MBW_PBM, rprops->mbw_pbm_bits);
-		else
  			mpam_write_partsel_reg(msc, MBW_PBM, cfg->mbw_pbm);
+		else
+			mpam_reset_msc_bitmap(msc, MPAMCFG_MBW_PBM, rprops->mbw_pbm_bits);
  	}
 
  	if (mpam_has_feature(mpam_feat_mbw_min, rprops)) {

Which machine or system where mpam_feat_mbw_part is set on RIS? As I can remember, this feature isn't available on grace-hopper.

Besides, I don't think this feature is well handled at present because mpam_config::mbw_pbm is only 32-bits in length, which doesn't match with the maximal length of the bit map (4096) as documented in the spec.

Thanks,
Gavin

author reviewer
#2 lore raw

Hi, Gavin,

Quoted 22 lines · Gavin Shan

Hi Fenghua,

Quoted 9 lines · Fenghua Yu

MPAMCFG_MBW_PBM is written from cfg if cfg has the MBW partition feature. It is reset when cfg does not have the MBW partition feature.

But the register handling is reversed. This may cause an incorrect register setting. For example, during an MPAM reset, reset_cfg is empty (no MBW partition feature set), and cfg->mbw_pbm is 0. Instead of resetting MPAMCFG_MBW_PBM to all 1's, the current logic will set it to cfg->mbw_pbm, which is 0.

Quoted 9 lines

Fix the issue by swapping the if/else branches.

Fixes a1cb6577f575 ("arm_mpam: Reset when feature configuration bit unset")

---
  drivers/resctrl/mpam_devices.c | 4 ++--

  1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 3 lines

The fix itself looks reasonable to me, but two questions below.

Thank you for reviewing the patch!

Quoted 15 lines
Quoted 15 lines
drivers/resctrl/mpam_devices.c +2 -2
index 4b93e89c2678..d8b0383cee92 100644
@@ -1570,9 +1570,9 @@ static void mpam_reprogram_ris_partid(struct 
 mpam_msc_ris *ris, u16 partid,
       if (mpam_has_feature(mpam_feat_mbw_part, rprops)) {
           if (mpam_has_feature(mpam_feat_mbw_part, cfg))
-            mpam_reset_msc_bitmap(msc, MPAMCFG_MBW_PBM, rprops- 
 >mbw_pbm_bits);
-        else
               mpam_write_partsel_reg(msc, MBW_PBM, cfg->mbw_pbm);
+        else
+            mpam_reset_msc_bitmap(msc, MPAMCFG_MBW_PBM, rprops- 
Quoted 6 lines
Quoted 3 lines
drivers/resctrl/mpam_devices.c +0 -0
 >mbw_pbm_bits);
       }

      if (mpam_has_feature(mpam_feat_mbw_min, rprops)) {

Which machine or system where mpam_feat_mbw_part is set on RIS? As I can remember, this feature isn't available on grace-hopper.

Neither Grace nor Vera supports this feature.

Quoted 4 lines

Besides, I don't think this feature is well handled at present because mpam_config::mbw_pbm is only 32-bits in length, which doesn't match with the maximal length of the bit map (4096) as documented in the spec.

Right. The current code only can support up to 32 bits of PBM bitmap. If PBM bitmp length is bigger than 32 bits, it's broken. I guess we will need to handle this feature properly when hardware supports more than 32 bits of PBM bitmap.

Ditto for CPBM bitmap.

This patch currently only fixes the regression issue introduced in commit a1cb6577f575 regardless size of PBM bitmap.

Thanks.

-Fenghua

maintainer Reviewed-by
#3 lore raw

Hi Fenghua,

Thanks for the fix!

Quoted 54 lines · Fenghua Yu

Hi, Gavin,

Quoted 22 lines · Gavin Shan

Hi Fenghua,

Quoted 9 lines · Fenghua Yu

MPAMCFG_MBW_PBM is written from cfg if cfg has the MBW partition feature. It is reset when cfg does not have the MBW partition feature.

But the register handling is reversed. This may cause an incorrect register setting. For example, during an MPAM reset, reset_cfg is empty (no MBW partition feature set), and cfg->mbw_pbm is 0. Instead of resetting MPAMCFG_MBW_PBM to all 1's, the current logic will set it to cfg->mbw_pbm, which is 0.

Quoted 9 lines

Fix the issue by swapping the if/else branches.

Fixes a1cb6577f575 ("arm_mpam: Reset when feature configuration bit unset")

---
  drivers/resctrl/mpam_devices.c | 4 ++--

  1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 3 lines

The fix itself looks reasonable to me, but two questions below.

Thank you for reviewing the patch!

Quoted 11 lines
Quoted 11 lines
drivers/resctrl/mpam_devices.c +0 -2
index 4b93e89c2678..d8b0383cee92 100644
@@ -1570,9 +1570,9 @@ static void mpam_reprogram_ris_partid(struct
 mpam_msc_ris *ris, u16 partid,
       if (mpam_has_feature(mpam_feat_mbw_part, rprops)) {
           if (mpam_has_feature(mpam_feat_mbw_part, cfg))
-            mpam_reset_msc_bitmap(msc, MPAMCFG_MBW_PBM, rprops- >mbw_pbm_bits);
-        else
Quoted 8 lines

drivers/resctrl/mpam_devices.c static void mpam_reprogram_ris_partid(struct mpam_msc_ris *ris, u16 partid, · line 1578

Quoted 5 lines
drivers/resctrl/mpam_devices.c +2 -0
               mpam_write_partsel_reg(msc, MBW_PBM, cfg->mbw_pbm);
+        else
+            mpam_reset_msc_bitmap(msc, MPAMCFG_MBW_PBM, rprops- >mbw_pbm_bits);
       }
       if (mpam_has_feature(mpam_feat_mbw_min, rprops)) {

Which machine or system where mpam_feat_mbw_part is set on RIS? As I can remember, this feature isn't available on grace-hopper.

Neither Grace nor Vera supports this feature.

Quoted 1 line
Quoted 6 lines
Quoted 4 lines

Besides, I don't think this feature is well handled at present because mpam_config::mbw_pbm is only 32-bits in length, which doesn't match with the maximal length of the bit map (4096) as documented in the spec. Right. The current code only can support up to 32 bits of PBM bitmap. If PBM

bitmp length is bigger than 32 bits, it's broken. I guess we will need to handle this feature properly when hardware supports more than 32 bits of PBM bitmap.

mpam_config::mbw_pbm shouldn't have made it into the driver as nothing ever sets it. Otherwise, there is no attempt to make use of the MBW_PBM other than to reset it to an appropriate value, which this patch fixes. I guess this is a lesson for me on the perils of dead code.

Quoted 2 lines

Ditto for CPBM bitmap.

If the CPBM bitmap has more than 32 bits, as checked by cache_has_usable_cpor(), then we don't attempt to configure it other than to reset. If there is hardware with more than 32 bits we'd first consider extending this to 64bits.

Quoted 3 lines

This patch currently only fixes the regression issue introduced in commit a1cb6577f575 regardless size of PBM bitmap.

Looks good to me.

Thanks,

Ben

Quoted 4 lines

Thanks.

-Fenghua

Applied
#4 lore raw
In reply to Fenghua Yu
Quoted 10 lines · Fenghua Yu

MPAMCFG_MBW_PBM is written from cfg if cfg has the MBW partition feature. It is reset when cfg does not have the MBW partition feature.

But the register handling is reversed. This may cause an incorrect register setting. For example, during an MPAM reset, reset_cfg is empty (no MBW partition feature set), and cfg->mbw_pbm is 0. Instead of resetting MPAMCFG_MBW_PBM to all 1's, the current logic will set it to cfg->mbw_pbm, which is 0.

[...]

Applied to arm64 (for-next/fixes), thanks!

[1/1] arm_mpam: Fix MPAMCFG_MBW_PBM register setting

      https://git.kernel.org/arm64/c/021118ce5ea9

Cheers,

Signature · 5 lines
Will

https://fixes.arm64.dev
https://next.arm64.dev
https://will.arm64.dev