Skip to contentSkip to conversation

powerpc/pseries: group lmb operation and memblock's

PATCH Superseded Pingfan Liu 3 participants 10 messages Jul 22, 2020 – Jul 28, 2020

Subsystems (3)

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/2 powerpc/pseries: group lmb operation and memblock's

This patch prepares for the incoming patch which swaps the order of KOBJ_ uevent and dt's updating.

It has no functional effect, just groups lmb operation and memblock's in order to insert dt updating operation easily, and makes it easier to review.

Cc Michael Ellerman <[email protected]>
Cc Hari Bathini <[email protected]>
Cc Nathan Lynch <[email protected]>

To: [email protected]
Cc: [email protected]
---
v2 -> v3: rebase onto ppc next-test branch

---
 arch/powerpc/platforms/pseries/hotplug-memory.c | 26 ++++++++++++++++---------
 1 file changed, 17 insertions(+), 9 deletions(-)
arch/powerpc/platforms/pseries/hotplug-memory.c +17 -9
index 5d545b7..1a3ac3b 100644
@@ -355,7 +355,8 @@ static int dlpar_add_lmb(struct drmem_lmb *);
 static int dlpar_remove_lmb(struct drmem_lmb *lmb)
 {
 	unsigned long block_sz;
-	int rc;
+	phys_addr_t base_addr;
+	int rc, nid;
 
 	if (!lmb_is_removable(lmb))
 		return -EINVAL;
@@ -364,17 +365,19 @@ static int dlpar_remove_lmb(struct drmem_lmb *lmb)
 	if (rc)
 		return rc;
 
+	base_addr = lmb->base_addr;
+	nid = lmb->nid;
 	block_sz = pseries_memory_block_size();
 
-	__remove_memory(lmb->nid, lmb->base_addr, block_sz);
-
-	/* Update memory regions for memory remove */
-	memblock_remove(lmb->base_addr, block_sz);
-
 	invalidate_lmb_associativity_index(lmb);
 	lmb_clear_nid(lmb);
 	lmb->flags &= ~DRCONF_MEM_ASSIGNED;
 
+	__remove_memory(nid, base_addr, block_sz);
+
+	/* Update memory regions for memory remove */
+	memblock_remove(base_addr, block_sz);
+
 	return 0;
 }
 
@@ -603,6 +606,8 @@ static int dlpar_add_lmb(struct drmem_lmb *lmb)
 	}
 
 	lmb_set_nid(lmb);
+	lmb->flags |= DRCONF_MEM_ASSIGNED;
+
 	block_sz = memory_block_size_bytes();
 
 	/* Add the memory */
@@ -614,11 +619,14 @@ static int dlpar_add_lmb(struct drmem_lmb *lmb)
 
 	rc = dlpar_online_lmb(lmb);
 	if (rc) {
-		__remove_memory(lmb->nid, lmb->base_addr, block_sz);
+		int nid = lmb->nid;
+		phys_addr_t base_addr = lmb->base_addr;
+
 		invalidate_lmb_associativity_index(lmb);
 		lmb_clear_nid(lmb);
-	} else {
-		lmb->flags |= DRCONF_MEM_ASSIGNED;
+		lmb->flags &= ~DRCONF_MEM_ASSIGNED;
+
+		__remove_memory(nid, base_addr, block_sz);
 	}
 
 	return rc;
Signature · 7 lines
2.7.5


_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec
#1 lore raw

PATCH v3 2/2 powerpc/pseries: update device tree before ejecting hotplug uevents

A bug is observed on pseries by taking the following steps on rhel:
-1. drmgr -c mem -r -q 5
-2. echo c > /proc/sysrq-trigger

And then, the failure looks like:
kdump: saving to /sysroot//var/crash/127.0.0.1-2020-01-16-02:06:14/
kdump: saving vmcore-dmesg.txt
kdump: saving vmcore-dmesg.txt complete
kdump: saving vmcore
Checking for memory holes : [ 0.0 %] / Checking for memory holes : [100.0 %] | Excluding unnecessary pages : [100.0 %] \ Copying data : [ 0.3 %] - eta: 38s[ 44.337636] hash-mmu: mm: Hashing failure ! EA=0x7fffba400000 access=0x8000000000000004 current=makedumpfile

[   44.337663] hash-mmu:     trap=0x300 vsid=0x13a109c ssize=1 base psize=2 psize 2 pte=0xc000000050000504
[   44.337677] hash-mmu: mm: Hashing failure ! EA=0x7fffba400000 access=0x8000000000000004 current=makedumpfile
[   44.337692] hash-mmu:     trap=0x300 vsid=0x13a109c ssize=1 base psize=2 psize 2 pte=0xc000000050000504
[   44.337708] makedumpfile[469]: unhandled signal 7 at 00007fffba400000 nip 00007fffbbc4d7fc lr 000000011356ca3c code 2
[   44.338548] Core dump to |/bin/false pipe failed

/lib/kdump-lib-initramfs.sh: line 98: 469 Bus error $CORE_COLLECTOR /proc/vmcore $_mp/$KDUMP_PATH/$HOST_IP-$DATEDIR/vmcore-incomplete
kdump: saving vmcore failed

* Root cause *
After analyzing, it turns out that in the current implementation,
when hot-removing lmb, the KOBJ_REMOVE event ejects before the dt updating as the code __remove_memory() comes before drmem_update_dt().
So in kdump kernel, when read_from_oldmem() resorts to pSeries_lpar_hpte_insert() to install hpte, but fails with -2 due to non-exist pfn. And finally, low_hash_fault() raise SIGBUS to process, as it can be observed "Bus error"

From a viewpoint of listener and publisher, the publisher notifies the listener before data is ready. This introduces a problem where udev launches kexec-tools (due to KOBJ_REMOVE) and loads a stale dt before updating. And in capture kernel, makedumpfile will access the memory based on the stale dt info, and hit a SIGBUS error due to an un-existed lmb.

* Fix *
In order to fix this issue, update dt before __remove_memory(), and
accordingly the same rule in hot-add path.

This will introduce extra dt updating payload for each involved lmb when hotplug. But it should be fine since drmem_update_dt() is memory based operation and hotplug is not a hot path.

Cc Michael Ellerman <[email protected]>
Cc Hari Bathini <[email protected]>
Cc Nathan Lynch <[email protected]>

To: [email protected]
Cc: [email protected]
---
v2 -> v3: rebase onto ppc next-test branch

---
 arch/powerpc/platforms/pseries/hotplug-memory.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
arch/powerpc/platforms/pseries/hotplug-memory.c +3 -3
index 1a3ac3b..def8cb3f 100644
@@ -372,6 +372,7 @@ static int dlpar_remove_lmb(struct drmem_lmb *lmb)
 	invalidate_lmb_associativity_index(lmb);
 	lmb_clear_nid(lmb);
 	lmb->flags &= ~DRCONF_MEM_ASSIGNED;
+	drmem_update_dt();
 
 	__remove_memory(nid, base_addr, block_sz);
 
@@ -607,6 +608,7 @@ static int dlpar_add_lmb(struct drmem_lmb *lmb)
 
 	lmb_set_nid(lmb);
 	lmb->flags |= DRCONF_MEM_ASSIGNED;
+	drmem_update_dt();
 
 	block_sz = memory_block_size_bytes();
 
@@ -625,6 +627,7 @@ static int dlpar_add_lmb(struct drmem_lmb *lmb)
 		invalidate_lmb_associativity_index(lmb);
 		lmb_clear_nid(lmb);
 		lmb->flags &= ~DRCONF_MEM_ASSIGNED;
+		drmem_update_dt();
 
 		__remove_memory(nid, base_addr, block_sz);
 	}
@@ -877,9 +880,6 @@ int dlpar_memory(struct pseries_hp_errorlog *hp_elog)
 		break;
 	}
 
-	if (!rc)
-		rc = drmem_update_dt();
-
 	unlock_device_hotplug();
 	return rc;
 }
Signature · 7 lines
2.7.5


_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec
#2 lore raw

Re: [PATCHv3 2/2] powerpc/pseries: update device tree before ejecting hotplug uevents

Quoted 1 line · Pingfan Liu

A bug is observed on pseries by taking the following steps on rhel:

                                                                ^
                                                                RHEL

I assume it happens on mainline too?

Quoted 57 lines

-1. drmgr -c mem -r -q 5
-2. echo c > /proc/sysrq-trigger

And then, the failure looks like:
kdump: saving to /sysroot//var/crash/127.0.0.1-2020-01-16-02:06:14/
kdump: saving vmcore-dmesg.txt
kdump: saving vmcore-dmesg.txt complete
kdump: saving vmcore
Checking for memory holes : [ 0.0 %] / Checking for memory holes : [100.0 %] | Excluding unnecessary pages : [100.0 %] \ Copying data : [ 0.3 %] - eta: 38s[ 44.337636] hash-mmu: mm: Hashing failure ! EA=0x7fffba400000 access=0x8000000000000004 current=makedumpfile

[   44.337663] hash-mmu:     trap=0x300 vsid=0x13a109c ssize=1 base psize=2 psize 2 pte=0xc000000050000504
[   44.337677] hash-mmu: mm: Hashing failure ! EA=0x7fffba400000 access=0x8000000000000004 current=makedumpfile
[   44.337692] hash-mmu:     trap=0x300 vsid=0x13a109c ssize=1 base psize=2 psize 2 pte=0xc000000050000504
[   44.337708] makedumpfile[469]: unhandled signal 7 at 00007fffba400000 nip 00007fffbbc4d7fc lr 000000011356ca3c code 2
[   44.338548] Core dump to |/bin/false pipe failed

/lib/kdump-lib-initramfs.sh: line 98: 469 Bus error $CORE_COLLECTOR /proc/vmcore $_mp/$KDUMP_PATH/$HOST_IP-$DATEDIR/vmcore-incomplete
kdump: saving vmcore failed

* Root cause *
After analyzing, it turns out that in the current implementation,
when hot-removing lmb, the KOBJ_REMOVE event ejects before the dt updating as the code __remove_memory() comes before drmem_update_dt().
So in kdump kernel, when read_from_oldmem() resorts to pSeries_lpar_hpte_insert() to install hpte, but fails with -2 due to non-exist pfn. And finally, low_hash_fault() raise SIGBUS to process, as it can be observed "Bus error"

From a viewpoint of listener and publisher, the publisher notifies the listener before data is ready. This introduces a problem where udev launches kexec-tools (due to KOBJ_REMOVE) and loads a stale dt before updating. And in capture kernel, makedumpfile will access the memory based on the stale dt info, and hit a SIGBUS error due to an un-existed lmb.

* Fix *
In order to fix this issue, update dt before __remove_memory(), and
accordingly the same rule in hot-add path.

This will introduce extra dt updating payload for each involved lmb when hotplug. But it should be fine since drmem_update_dt() is memory based operation and hotplug is not a hot path.

Cc Michael Ellerman <[email protected]>
Cc Hari Bathini <[email protected]>
Cc Nathan Lynch <[email protected]>

To: [email protected]
Cc: [email protected]
---
v2 -> v3: rebase onto ppc next-test branch

---
 arch/powerpc/platforms/pseries/hotplug-memory.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
arch/powerpc/platforms/pseries/hotplug-memory.c +0 -0
index 1a3ac3b..def8cb3f 100644

arch/powerpc/platforms/pseries/hotplug-memory.c static int dlpar_remove_lmb(struct drmem_lmb *lmb) · line 375

Quoted 4 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +1 -0
@@ -372,6 +372,7 @@ static int dlpar_remove_lmb(struct drmem_lmb *lmb)
 	invalidate_lmb_associativity_index(lmb);
 	lmb_clear_nid(lmb);
 	lmb->flags &= ~DRCONF_MEM_ASSIGNED;
+	drmem_update_dt();

No error checking?

arch/powerpc/platforms/pseries/hotplug-memory.c static int dlpar_add_lmb(struct drmem_lmb *lmb) · line 611

Quoted 7 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +1 -0
 	__remove_memory(nid, base_addr, block_sz);
 
@@ -607,6 +608,7 @@ static int dlpar_add_lmb(struct drmem_lmb *lmb)
 
 	lmb_set_nid(lmb);
 	lmb->flags |= DRCONF_MEM_ASSIGNED;
+	drmem_update_dt();

And here ..

arch/powerpc/platforms/pseries/hotplug-memory.c static int dlpar_add_lmb(struct drmem_lmb *lmb) · line 630

Quoted 8 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +1 -0
 	block_sz = memory_block_size_bytes();
 
@@ -625,6 +627,7 @@ static int dlpar_add_lmb(struct drmem_lmb *lmb)
 		invalidate_lmb_associativity_index(lmb);
 		lmb_clear_nid(lmb);
 		lmb->flags &= ~DRCONF_MEM_ASSIGNED;
+		drmem_update_dt();

And here ..

arch/powerpc/platforms/pseries/hotplug-memory.c int dlpar_memory(struct pseries_hp_errorlog *hp_elog) · line 884

Quoted 11 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +0 -3
 		__remove_memory(nid, base_addr, block_sz);
 	}
@@ -877,9 +880,6 @@ int dlpar_memory(struct pseries_hp_errorlog *hp_elog)
 		break;
 	}
 
-	if (!rc)
-		rc = drmem_update_dt();
-
 	unlock_device_hotplug();
 	return rc;

Whereas previously we did check it.

cheers

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#3 lore raw

Re: [PATCHv3 2/2] powerpc/pseries: update device tree before ejecting hotplug uevents

Quoted 7 lines · Michael Ellerman
Quoted 1 line · Pingfan Liu

A bug is observed on pseries by taking the following steps on rhel:

                                                                ^
                                                                RHEL

I assume it happens on mainline too?

Yes, it does.

Quoted 1 line

[...]

Quoted 11 lines

arch/powerpc/platforms/pseries/hotplug-memory.c static int dlpar_remove_lmb(struct drmem_lmb *lmb) · line 375

Quoted 9 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +1 -0
index 1a3ac3b..def8cb3f 100644
@@ -372,6 +372,7 @@ static int dlpar_remove_lmb(struct drmem_lmb *lmb)
      invalidate_lmb_associativity_index(lmb);
      lmb_clear_nid(lmb);
      lmb->flags &= ~DRCONF_MEM_ASSIGNED;
+     drmem_update_dt();

No error checking?

Hmm, here should be a more careful design. Please see the comment at the end.

Quoted 28 lines

arch/powerpc/platforms/pseries/hotplug-memory.c static int dlpar_add_lmb(struct drmem_lmb *lmb) · line 611

Quoted 8 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +1 -0
      __remove_memory(nid, base_addr, block_sz);
 
@@ -607,6 +608,7 @@ static int dlpar_add_lmb(struct drmem_lmb *lmb)
 
      lmb_set_nid(lmb);
      lmb->flags |= DRCONF_MEM_ASSIGNED;
+     drmem_update_dt();

And here ..

arch/powerpc/platforms/pseries/hotplug-memory.c static int dlpar_add_lmb(struct drmem_lmb *lmb) · line 630

Quoted 8 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +1 -0
      block_sz = memory_block_size_bytes();
 
@@ -625,6 +627,7 @@ static int dlpar_add_lmb(struct drmem_lmb *lmb)
              invalidate_lmb_associativity_index(lmb);
              lmb_clear_nid(lmb);
              lmb->flags &= ~DRCONF_MEM_ASSIGNED;
+             drmem_update_dt();

And here ..

Quoted 6 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +0 -0
              __remove_memory(nid, base_addr, block_sz);
      }
@@ -877,9 +880,6 @@ int dlpar_memory(struct pseries_hp_errorlog *hp_elog)
              break;
      }
 
Quoted 7 lines

arch/powerpc/platforms/pseries/hotplug-memory.c int dlpar_memory(struct pseries_hp_errorlog *hp_elog) · line 884

Quoted 5 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +0 -3
-     if (!rc)
-             rc = drmem_update_dt();
-
      unlock_device_hotplug();
      return rc;

Whereas previously we did check it.

drmem_update_dt() fails iff allocating memory fail. And in the failed case, even the original code does not roll back the effect of __add_memory()/__remove_memory().

And I plan to do the following in V4: if drmem_update_dt() fails in dlpar_add_lmb(), then bails out immediately.

Thanks,
Pingfan

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#4 lore raw

Re: [PATCHv3 2/2] powerpc/pseries: update device tree before ejecting hotplug uevents

Quoted 59 lines · Pingfan Liu
Quoted 7 lines · Michael Ellerman
Quoted 1 line · Pingfan Liu

A bug is observed on pseries by taking the following steps on rhel:

                                                                ^
                                                                RHEL

I assume it happens on mainline too?

Yes, it does.

Quoted 1 line

[...]

Quoted 11 lines

arch/powerpc/platforms/pseries/hotplug-memory.c static int dlpar_remove_lmb(struct drmem_lmb *lmb) · line 375

Quoted 9 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +1 -0
index 1a3ac3b..def8cb3f 100644
@@ -372,6 +372,7 @@ static int dlpar_remove_lmb(struct drmem_lmb *lmb)
      invalidate_lmb_associativity_index(lmb);
      lmb_clear_nid(lmb);
      lmb->flags &= ~DRCONF_MEM_ASSIGNED;
+     drmem_update_dt();

No error checking?

Hmm, here should be a more careful design. Please see the comment at the end.

Quoted 28 lines

arch/powerpc/platforms/pseries/hotplug-memory.c static int dlpar_add_lmb(struct drmem_lmb *lmb) · line 611

Quoted 8 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +1 -0
      __remove_memory(nid, base_addr, block_sz);
 
@@ -607,6 +608,7 @@ static int dlpar_add_lmb(struct drmem_lmb *lmb)
 
      lmb_set_nid(lmb);
      lmb->flags |= DRCONF_MEM_ASSIGNED;
+     drmem_update_dt();

And here ..

arch/powerpc/platforms/pseries/hotplug-memory.c static int dlpar_add_lmb(struct drmem_lmb *lmb) · line 630

Quoted 8 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +1 -0
      block_sz = memory_block_size_bytes();
 
@@ -625,6 +627,7 @@ static int dlpar_add_lmb(struct drmem_lmb *lmb)
              invalidate_lmb_associativity_index(lmb);
              lmb_clear_nid(lmb);
              lmb->flags &= ~DRCONF_MEM_ASSIGNED;
+             drmem_update_dt();

And here ..

Quoted 6 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +0 -0
              __remove_memory(nid, base_addr, block_sz);
      }
@@ -877,9 +880,6 @@ int dlpar_memory(struct pseries_hp_errorlog *hp_elog)
              break;
      }
 
Quoted 7 lines

arch/powerpc/platforms/pseries/hotplug-memory.c int dlpar_memory(struct pseries_hp_errorlog *hp_elog) · line 884

Quoted 5 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +0 -3
-     if (!rc)
-             rc = drmem_update_dt();
-
      unlock_device_hotplug();
      return rc;

Whereas previously we did check it.

Quoted 1 line

drmem_update_dt() fails iff allocating memory fail.

That's true currently, but it might change in future.

Quoted 2 lines

And in the failed case, even the original code does not roll back the effect of __add_memory()/__remove_memory().

Yeah hard to know what the desired behaviour is. If something fails we at least need to print a message though, not silently swallow it.

Quoted 2 lines

And I plan to do the following in V4: if drmem_update_dt() fails in dlpar_add_lmb(), then bails out immediately.

That sounds reasonable.

cheers

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#5 lore raw

Re: [PATCHv3 2/2] powerpc/pseries: update device tree before ejecting hotplug uevents

In reply to Pingfan Liu
Quoted 33 lines · Pingfan Liu

A bug is observed on pseries by taking the following steps on rhel:
-1. drmgr -c mem -r -q 5
-2. echo c > /proc/sysrq-trigger

And then, the failure looks like:
kdump: saving to /sysroot//var/crash/127.0.0.1-2020-01-16-02:06:14/
kdump: saving vmcore-dmesg.txt
kdump: saving vmcore-dmesg.txt complete
kdump: saving vmcore
Checking for memory holes : [ 0.0 %] / Checking for memory holes : [100.0 %] | Excluding unnecessary pages : [100.0 %] \ Copying data : [ 0.3 %] - eta: 38s[ 44.337636] hash-mmu: mm: Hashing failure ! EA=0x7fffba400000 access=0x8000000000000004 current=makedumpfile

[   44.337663] hash-mmu:     trap=0x300 vsid=0x13a109c ssize=1 base psize=2 psize 2 pte=0xc000000050000504
[   44.337677] hash-mmu: mm: Hashing failure ! EA=0x7fffba400000 access=0x8000000000000004 current=makedumpfile
[   44.337692] hash-mmu:     trap=0x300 vsid=0x13a109c ssize=1 base psize=2 psize 2 pte=0xc000000050000504
[   44.337708] makedumpfile[469]: unhandled signal 7 at 00007fffba400000 nip 00007fffbbc4d7fc lr 000000011356ca3c code 2
[   44.338548] Core dump to |/bin/false pipe failed

/lib/kdump-lib-initramfs.sh: line 98: 469 Bus error $CORE_COLLECTOR /proc/vmcore $_mp/$KDUMP_PATH/$HOST_IP-$DATEDIR/vmcore-incomplete
kdump: saving vmcore failed

* Root cause *
After analyzing, it turns out that in the current implementation,
when hot-removing lmb, the KOBJ_REMOVE event ejects before the dt updating as the code __remove_memory() comes before drmem_update_dt().
So in kdump kernel, when read_from_oldmem() resorts to pSeries_lpar_hpte_insert() to install hpte, but fails with -2 due to non-exist pfn. And finally, low_hash_fault() raise SIGBUS to process, as it can be observed "Bus error"

From a viewpoint of listener and publisher, the publisher notifies the listener before data is ready. This introduces a problem where udev launches kexec-tools (due to KOBJ_REMOVE) and loads a stale dt before updating. And in capture kernel, makedumpfile will access the memory based on the stale dt info, and hit a SIGBUS error due to an un-existed lmb.

Quoted 7 lines

* Fix *
In order to fix this issue, update dt before __remove_memory(), and
accordingly the same rule in hot-add path.

This will introduce extra dt updating payload for each involved lmb when hotplug. But it should be fine since drmem_update_dt() is memory based operation and hotplug is not a hot path.

This is great analysis but the performance implications of the change are grave. The add/remove paths here are already O(n) where n is the quantity of memory assigned to the LP, this change would make it O(n^2):

dlpar_memory_add_by_count
for_each_drmem_lmb <--

    dlpar_add_lmb
      drmem_update_dt(_v1|_v2)
        for_each_drmem_lmb       <--

Memory add/remove isn't a hot path but quadratic runtime complexity isn't acceptable. Its current performance is bad enough that I have internal bugs open on it.

Not to mention we leak memory every time drmem_update_dt is called because we can't safely free device tree properties :-(

Also note that this sort of reverts (fixes?) 063b8b1251fd ("powerpc/pseries/memory-hotplug: Only update DT once per memory DLPAR request").

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#6 lore raw

Re: [PATCHv3 2/2] powerpc/pseries: update device tree before ejecting hotplug uevents

Quoted 47 lines · Nathan Lynch
Quoted 33 lines · Pingfan Liu

A bug is observed on pseries by taking the following steps on rhel:
-1. drmgr -c mem -r -q 5
-2. echo c > /proc/sysrq-trigger

And then, the failure looks like:
kdump: saving to /sysroot//var/crash/127.0.0.1-2020-01-16-02:06:14/
kdump: saving vmcore-dmesg.txt
kdump: saving vmcore-dmesg.txt complete
kdump: saving vmcore
Checking for memory holes : [ 0.0 %] / Checking for memory holes : [100.0 %] | Excluding unnecessary pages : [100.0 %] \ Copying data : [ 0.3 %] - eta: 38s[ 44.337636] hash-mmu: mm: Hashing failure ! EA=0x7fffba400000 access=0x8000000000000004 current=makedumpfile

[   44.337663] hash-mmu:     trap=0x300 vsid=0x13a109c ssize=1 base psize=2 psize 2 pte=0xc000000050000504
[   44.337677] hash-mmu: mm: Hashing failure ! EA=0x7fffba400000 access=0x8000000000000004 current=makedumpfile
[   44.337692] hash-mmu:     trap=0x300 vsid=0x13a109c ssize=1 base psize=2 psize 2 pte=0xc000000050000504
[   44.337708] makedumpfile[469]: unhandled signal 7 at 00007fffba400000 nip 00007fffbbc4d7fc lr 000000011356ca3c code 2
[   44.338548] Core dump to |/bin/false pipe failed

/lib/kdump-lib-initramfs.sh: line 98: 469 Bus error $CORE_COLLECTOR /proc/vmcore $_mp/$KDUMP_PATH/$HOST_IP-$DATEDIR/vmcore-incomplete
kdump: saving vmcore failed

* Root cause *
After analyzing, it turns out that in the current implementation,
when hot-removing lmb, the KOBJ_REMOVE event ejects before the dt updating as the code __remove_memory() comes before drmem_update_dt().
So in kdump kernel, when read_from_oldmem() resorts to pSeries_lpar_hpte_insert() to install hpte, but fails with -2 due to non-exist pfn. And finally, low_hash_fault() raise SIGBUS to process, as it can be observed "Bus error"

From a viewpoint of listener and publisher, the publisher notifies the listener before data is ready. This introduces a problem where udev launches kexec-tools (due to KOBJ_REMOVE) and loads a stale dt before updating. And in capture kernel, makedumpfile will access the memory based on the stale dt info, and hit a SIGBUS error due to an un-existed lmb.

Quoted 7 lines

* Fix *
In order to fix this issue, update dt before __remove_memory(), and
accordingly the same rule in hot-add path.

This will introduce extra dt updating payload for each involved lmb when hotplug. But it should be fine since drmem_update_dt() is memory based operation and hotplug is not a hot path.

This is great analysis but the performance implications of the change are grave. The add/remove paths here are already O(n) where n is the quantity of memory assigned to the LP, this change would make it O(n^2):

Quoted 8 lines

dlpar_memory_add_by_count
for_each_drmem_lmb <--

    dlpar_add_lmb
      drmem_update_dt(_v1|_v2)
        for_each_drmem_lmb       <--

Memory add/remove isn't a hot path but quadratic runtime complexity isn't acceptable. Its current performance is bad enough that I have

Yes, the quadratic runtime complexity sounds terrible.
And I am curious about the bug. Does the system have thousands of lmb?

Quoted 4 lines

internal bugs open on it.

Not to mention we leak memory every time drmem_update_dt is called because we can't safely free device tree properties :-(

Do you know what block us to free it?

Quoted 4 lines

Also note that this sort of reverts (fixes?) 063b8b1251fd ("powerpc/pseries/memory-hotplug: Only update DT once per memory DLPAR request").

Yes. And now, I think I need to bring up another method to fix it.

Thanks,
Pingfan

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#7 lore raw

Re: [PATCHv3 2/2] powerpc/pseries: update device tree before ejecting hotplug uevents

Quoted 16 lines · Pingfan Liu
Quoted 15 lines · Nathan Lynch
Quoted 3 lines · Pingfan Liu

This will introduce extra dt updating payload for each involved lmb when hotplug. But it should be fine since drmem_update_dt() is memory based operation and hotplug is not a hot path.

This is great analysis but the performance implications of the change are grave. The add/remove paths here are already O(n) where n is the quantity of memory assigned to the LP, this change would make it O(n^2):

dlpar_memory_add_by_count
for_each_drmem_lmb <--

    dlpar_add_lmb
      drmem_update_dt(_v1|_v2)
        for_each_drmem_lmb       <--
Quoted 4 lines
Quoted 2 lines

Memory add/remove isn't a hot path but quadratic runtime complexity isn't acceptable. Its current performance is bad enough that I have

Yes, the quadratic runtime complexity sounds terrible.
And I am curious about the bug. Does the system have thousands of lmb?

Yes.

Quoted 3 lines
Quoted 2 lines

Not to mention we leak memory every time drmem_update_dt is called because we can't safely free device tree properties :-(

Do you know what block us to free it?

It's a longstanding problem. References to device tree properties aren't counted or tracked so there's no way to safely free them unless the node itself is released. But the ibm,dynamic-reconfiguration-memory node does not ever go away and its properties are only subject to updates.

Maybe there's a way to address the specific case of ibm,dynamic-reconfiguration-memory and the ibm,dynamic-memory(-v2) properties, instead of tackling the general problem.

Regardless of all that, the drmem code needs better data structures and lookup functions.

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#8 lore raw
In reply to Pingfan Liu
Quoted 6 lines · Pingfan Liu

This patch prepares for the incoming patch which swaps the order of KOBJ_ uevent and dt's updating.

It has no functional effect, just groups lmb operation and memblock's in order to insert dt updating operation easily, and makes it easier to review.

...

Quoted 28 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +4 -6
index 5d545b7..1a3ac3b 100644
@@ -355,7 +355,8 @@ static int dlpar_add_lmb(struct drmem_lmb *);
 static int dlpar_remove_lmb(struct drmem_lmb *lmb)
 {
 	unsigned long block_sz;
-	int rc;
+	phys_addr_t base_addr;
+	int rc, nid;
 
 	if (!lmb_is_removable(lmb))
 		return -EINVAL;
@@ -364,17 +365,19 @@ static int dlpar_remove_lmb(struct drmem_lmb *lmb)
 	if (rc)
 		return rc;
 
+	base_addr = lmb->base_addr;
+	nid = lmb->nid;
 	block_sz = pseries_memory_block_size();
 
-	__remove_memory(lmb->nid, lmb->base_addr, block_sz);
-
-	/* Update memory regions for memory remove */
-	memblock_remove(lmb->base_addr, block_sz);
-
 	invalidate_lmb_associativity_index(lmb);

arch/powerpc/platforms/pseries/hotplug-memory.c static int dlpar_remove_lmb(struct drmem_lmb *lmb) · line 382

Quoted 10 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +5 -0
 	lmb_clear_nid(lmb);
 	lmb->flags &= ~DRCONF_MEM_ASSIGNED;
 
+	__remove_memory(nid, base_addr, block_sz);
+
+	/* Update memory regions for memory remove */
+	memblock_remove(base_addr, block_sz);
+
 	return 0;
 }

I don't understand; the commit message should not claim this has no functional effect when it changes the order of operations like this. Maybe this is an improvement over the current behavior, but it's not explained why it would be.

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#9 lore raw
Quoted 42 lines · Nathan Lynch
Quoted 6 lines · Pingfan Liu

This patch prepares for the incoming patch which swaps the order of KOBJ_ uevent and dt's updating.

It has no functional effect, just groups lmb operation and memblock's in order to insert dt updating operation easily, and makes it easier to review.

...

Quoted 31 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +4 -6
index 5d545b7..1a3ac3b 100644
@@ -355,7 +355,8 @@ static int dlpar_add_lmb(struct drmem_lmb *);
 static int dlpar_remove_lmb(struct drmem_lmb *lmb)
 {
      unsigned long block_sz;
-     int rc;
+     phys_addr_t base_addr;
+     int rc, nid;
 
      if (!lmb_is_removable(lmb))
              return -EINVAL;
@@ -364,17 +365,19 @@ static int dlpar_remove_lmb(struct drmem_lmb *lmb)
      if (rc)
              return rc;
 
+     base_addr = lmb->base_addr;
+     nid = lmb->nid;
      block_sz = pseries_memory_block_size();
 
-     __remove_memory(lmb->nid, lmb->base_addr, block_sz);
-
-     /* Update memory regions for memory remove */
-     memblock_remove(lmb->base_addr, block_sz);
-
      invalidate_lmb_associativity_index(lmb);
      lmb_clear_nid(lmb);
      lmb->flags &= ~DRCONF_MEM_ASSIGNED;
 
Quoted 12 lines

arch/powerpc/platforms/pseries/hotplug-memory.c static int dlpar_remove_lmb(struct drmem_lmb *lmb) · line 382

Quoted 7 lines
arch/powerpc/platforms/pseries/hotplug-memory.c +5 -0
+     __remove_memory(nid, base_addr, block_sz);
+
+     /* Update memory regions for memory remove */
+     memblock_remove(base_addr, block_sz);
+
      return 0;
 }

I don't understand; the commit message should not claim this has no functional effect when it changes the order of operations like this. Maybe this is an improvement over the current behavior, but it's not explained why it would be.

One group of functions, which name contains lmb, are powerpc specific, and used to form dt.

The other group __remove_memory() and memblock_remove() are integrated with linux mm.

And [2/2] arrange dt-updating just before __remove_memory()

Thanks,
Pingfan

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec