Skip to contentSkip to conversation

Drivers: hv: hv_balloon: do not online pages in offline blocks

PATCH In review Vitaly Kuznetsov 2 participants 5 messages Mar 25, 2015 – Mar 27, 2015 Hyper-V/Azure core and drivers

lore
Share & tools
lore
#0 lore raw

cover Drivers: hv: hv_balloon: do not online pages in offline blocks

Patch 1/1 is rather a hackaround. The root cause of the issue is the fact that we're adding not fully backed memory blocks and allow to online them. It is relatively easy to forbid that and add only fully backed 128Mb regions. The whole driver code will be much simpler: hv_bring_pgs_online() and hv_online_page() will go away, ha_region_mutex won't probably be required,... Please let me know if you think I should follow this way.

Vitaly Kuznetsov (1):
Drivers: hv: hv_balloon: do not online pages in offline blocks

 drivers/hv/hv_balloon.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
Signature · 1 line
1.9.3
#1 lore raw

PATCH 1/1 Drivers: hv: hv_balloon: do not online pages in offline blocks

Currently we add memory in 128Mb blocks but the request from host can be aligned differently. In such case we add a partially backed block and when this block goes online we skip onlining pages which are not backed (hv_online_page() callback serves this purpose). When we receive next request for the same host add region we online pages which were not backed before with hv_bring_pgs_online(). However, we don't check if the the block in question was onlined and online this tail unconditionally. This is bad as we avoid all online_pages() logic: these pages are not accounted, we don't send notifications (and hv_balloon is not the only receiver of them),... And, first of all, nobody asked as to online these pages. Solve the issue by checking if the last previously backed page was onlined and onlining the tail only in case it was.

---
 drivers/hv/hv_balloon.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
drivers/hv/hv_balloon.c +11 -1
index 014256a..99afef9 100644
@@ -778,7 +778,17 @@ static unsigned long handle_pg_range(unsigned long pg_start,
 			pgs_ol = has->ha_end_pfn - start_pfn;
 			if (pgs_ol > pfn_cnt)
 				pgs_ol = pfn_cnt;
-			hv_bring_pgs_online(start_pfn, pgs_ol);
+
+			/*
+			 * Check if the corresponding memory block is already
+			 * online by checking its last previously backed page.
+			 * In case it is we need to bring rest (which was not
+			 * backed previously) online too.
+			 */
+			if (start_pfn > has->start_pfn &&
+			    !PageReserved(pfn_to_page(start_pfn - 1)))
+				hv_bring_pgs_online(start_pfn, pgs_ol);
+
 			has->covered_end_pfn +=  pgs_ol;
 			pfn_cnt -= pgs_ol;
 		}
Signature · 1 line
1.9.3
#2 lore raw
In reply to Vitaly Kuznetsov
Quoted 11 lines

-----Original Message-----
From: Vitaly Kuznetsov [mailto:[email protected]]
Sent: Wednesday, March 25, 2015 11:02 AM
To: KY Srinivasan
Cc: Haiyang Zhang; [email protected]; linux- [email protected]; Dexuan Cui
Subject: [PATCH 0/1] Drivers: hv: hv_balloon: do not online pages in offline blocks

Quoted 9 lines

Patch 1/1 is rather a hackaround. The root cause of the issue is the fact that we're adding not fully backed memory blocks and allow to online them. It is relatively easy to forbid that and add only fully backed 128Mb regions. The whole
driver code will be much simpler: hv_bring_pgs_online() and hv_online_page()
will go away, ha_region_mutex won't probably be required,... Please let me know
if you think I should follow this way.

Vitaly,

The main issue I had to address here was that Windows has a 2M alignment for memory hot add while Linux has a 128M alignment requirement. Furthermore, we will also have to deal with the case where only a partial 128MB block might be hot-added. The strategy I chose of selectively onlining pages in a properly aligned block that would be hot added was to address the above two issues.

Regards,

K. Y

Quoted 9 lines

Vitaly Kuznetsov (1):
Drivers: hv: hv_balloon: do not online pages in offline blocks

 drivers/hv/hv_balloon.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

--
1.9.3

#3 lore raw
Quoted 19 lines · KY Srinivasan
Quoted 9 lines

-----Original Message-----
From: Vitaly Kuznetsov [mailto:[email protected]]
Sent: Wednesday, March 25, 2015 11:02 AM
To: KY Srinivasan
Cc: Haiyang Zhang; [email protected]; linux- [email protected]; Dexuan Cui
Subject: [PATCH 0/1] Drivers: hv: hv_balloon: do not online pages in offline blocks

Quoted 9 lines

Patch 1/1 is rather a hackaround. The root cause of the issue is the fact that we're adding not fully backed memory blocks and allow to online them. It is relatively easy to forbid that and add only fully backed 128Mb regions. The whole
driver code will be much simpler: hv_bring_pgs_online() and hv_online_page()
will go away, ha_region_mutex won't probably be required,... Please let me know
if you think I should follow this way.

Quoted 7 lines

Vitaly,

The main issue I had to address here was that Windows has a 2M alignment for memory hot add while Linux has a 128M alignment requirement. Furthermore, we will also have to deal with the case where only a partial 128MB block might be hot-added. The strategy I chose of selectively onlining pages in a properly aligned block that would be hot added was
to address the above two issues.

Yes, we can keep things this was, basically everyhting works. I just wanted to highlight the other possible solution for the issue which is also a major simplification for the driver.

So what do you think about my patch? We definitely should avoid onlining pages in offline blocks.

Quoted 13 lines

Regards,

K. Y

Quoted 9 lines

Vitaly Kuznetsov (1):
Drivers: hv: hv_balloon: do not online pages in offline blocks

 drivers/hv/hv_balloon.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

--
1.9.3

Signature · 1 line
  Vitaly
#4 lore raw
Quoted 47 lines

-----Original Message-----
From: Vitaly Kuznetsov [mailto:[email protected]]
Sent: Thursday, March 26, 2015 2:40 AM
To: KY Srinivasan
Cc: Haiyang Zhang; [email protected]; linux- [email protected]; Dexuan Cui
Subject: Re: [PATCH 0/1] Drivers: hv: hv_balloon: do not online pages in offline blocks

Quoted 7 lines · KY Srinivasan
Quoted 7 lines

-----Original Message-----
From: Vitaly Kuznetsov [mailto:[email protected]]
Sent: Wednesday, March 25, 2015 11:02 AM
To: KY Srinivasan
Cc: Haiyang Zhang; [email protected]; linux- [email protected]; Dexuan Cui
Subject: [PATCH 0/1] Drivers: hv: hv_balloon: do not online pages in

offline

Quoted 3 lines
Quoted 3 lines

blocks

Patch 1/1 is rather a hackaround. The root cause of the issue is the fact

that

Quoted 1 line
Quoted 1 line

we're adding not fully backed memory blocks and allow to online them. It

is

Quoted 5 lines
Quoted 5 lines

relatively easy to forbid that and add only fully backed 128Mb regions. The whole
driver code will be much simpler: hv_bring_pgs_online() and hv_online_page()
will go away, ha_region_mutex won't probably be required,... Please let

me

Quoted 6 lines
Quoted 2 lines

know
if you think I should follow this way.

Vitaly,

The main issue I had to address here was that Windows has a 2M alignment

for memory

Quoted 1 line

hot add while Linux has a 128M alignment requirement. Furthermore, we

will also have to deal with

Quoted 1 line

the case where only a partial 128MB block might be hot-added. The

strategy I chose of selectively

Quoted 2 lines

onlining pages in a properly aligned block that would be hot added was to address the above two issues.

Quoted 6 lines

Yes, we can keep things this was, basically everyhting works. I just wanted to highlight the other possible solution for the issue which is also a major simplification for the driver.

So what do you think about my patch? We definitely should avoid onlining pages in offline blocks.

Yes; looks good.

K. Y

Quoted 17 lines
Quoted 14 lines

Regards,

K. Y

Quoted 9 lines

Vitaly Kuznetsov (1):
Drivers: hv: hv_balloon: do not online pages in offline blocks

 drivers/hv/hv_balloon.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

--
1.9.3

--
Vitaly