Skip to contentSkip to conversation

Revert "iee1275/datetime: Fix off-by-1 error."

PATCH In review Glenn Washburn 3 participants 5 messages Jan 25, 2022 – Feb 11, 2022

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Revert "iee1275/datetime: Fix off-by-1 error."

This is causing the test grub_cmd_date to fail because the returned date is one day more than it should be.

This reverts commit 607d66116a67e5a13eb0d46076f26dedc988e6a4.

---
Hi all,

Reverting this commit allows the grub_cmd_date test to pass. It appears that this commit is (now) causing an off-by-1 error in QEMU and the latest OpenBIOS. What I'm unsure of is if the original commit is actually correct on real hardware and that potentially OpenBIOS has a bug.

Adrian and Daniel A, could you test the reverting of this commit on real hardware and see if date does in fact produce the expected date (and do current builds show a date one day ahead of what it should be)? Can anyone point to documentation saying that the original commit is in fact what should be done? If the issue is in OpenBIOS I'd like to have some documentation to back up a bug report.

Vladimir, do you have any thoughts on this?

Glenn

---
 grub-core/lib/ieee1275/datetime.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
grub-core/lib/ieee1275/datetime.c +2 -2
index b81fba2ed..74578f15a 100644
@@ -95,7 +95,7 @@ grub_get_datetime (struct grub_datetime *datetime)
 
   datetime->year = args.year;
   datetime->month = args.month;
-  datetime->day = args.day + 1;
+  datetime->day = args.day;
   datetime->hour = args.hour;
   datetime->minute = args.minute;
   datetime->second = args.second;
@@ -140,7 +140,7 @@ grub_set_datetime (struct grub_datetime *datetime)
 
   args.year = datetime->year;
   args.month = datetime->month;
-  args.day = datetime->day - 1;
+  args.day = datetime->day;
   args.hour = datetime->hour;
   args.minute = datetime->minute;
   args.second = datetime->second;
Signature · 1 line
2.27.0
Tested-by on v1 1/1
#1 lore raw

Hi,

I tested a pseries guest under:
- qemu + KVM on a Power8 host with a fairly modern qemu
- qemu + TCG

Here the 'date' command simply reports "error: no cmos found", so this patch will have no impact on those platforms.

I also tested an LPAR on a Power8 PowerVM machine. In this case, before the patch, grub printed:

grub> date

2022-02-09 04:31:10 Wednesday

whereas on booting:

[dja@sauce ~]$ date -u
Tue Feb 8 04:46:40 UTC 2022

After applying the patch, grub printed:

grub> date

2022-02-08 04:51:27 Tuesday

It seems the patch makes things better.

Kind regards,
Daniel

Quoted 53 lines · Glenn Washburn

This is causing the test grub_cmd_date to fail because the returned date is one day more than it should be.

This reverts commit 607d66116a67e5a13eb0d46076f26dedc988e6a4.

---
Hi all,

Reverting this commit allows the grub_cmd_date test to pass. It appears that this commit is (now) causing an off-by-1 error in QEMU and the latest OpenBIOS. What I'm unsure of is if the original commit is actually correct on real hardware and that potentially OpenBIOS has a bug.

Adrian and Daniel A, could you test the reverting of this commit on real hardware and see if date does in fact produce the expected date (and do current builds show a date one day ahead of what it should be)? Can anyone point to documentation saying that the original commit is in fact what should be done? If the issue is in OpenBIOS I'd like to have some documentation to back up a bug report.

Vladimir, do you have any thoughts on this?

Glenn

---
 grub-core/lib/ieee1275/datetime.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
grub-core/lib/ieee1275/datetime.c +2 -2
index b81fba2ed..74578f15a 100644
@@ -95,7 +95,7 @@ grub_get_datetime (struct grub_datetime *datetime)
 
   datetime->year = args.year;
   datetime->month = args.month;
-  datetime->day = args.day + 1;
+  datetime->day = args.day;
   datetime->hour = args.hour;
   datetime->minute = args.minute;
   datetime->second = args.second;
@@ -140,7 +140,7 @@ grub_set_datetime (struct grub_datetime *datetime)
 
   args.year = datetime->year;
   args.month = datetime->month;
-  args.day = datetime->day - 1;
+  args.day = datetime->day;
   args.hour = datetime->hour;
   args.minute = datetime->minute;
   args.second = datetime->second;
Signature · 1 line
2.27.0
#2 lore raw
Quoted 8 lines · On Tue, 08 Feb 2022 15:51:41 +1100 Daniel Axtens <[email protected]> wrote:

Hi,

I tested a pseries guest under:
- qemu + KVM on a Power8 host with a fairly modern qemu
- qemu + TCG

Here the 'date' command simply reports "error: no cmos found", so this patch will have no impact on those platforms.

I suspect its the same on KVM and TCG because the machine definition is the same and doesn't container either an RTC nor a CMOS. What confusing to me is, don't these machines have clocks?

Quoted 12 lines

I also tested an LPAR on a Power8 PowerVM machine. In this case, before the patch, grub printed:

grub> date

2022-02-09 04:31:10 Wednesday

whereas on booting:

[dja@sauce ~]$ date -u
Tue Feb 8 04:46:40 UTC 2022

Quoted 6 lines

After applying the patch, grub printed:

grub> date

2022-02-08 04:51:27 Tuesday

It seems the patch makes things better.

Great thanks Daniel for confirming this.

Glenn

Quoted 61 lines

Kind regards,
Daniel

Quoted 53 lines · Glenn Washburn

This is causing the test grub_cmd_date to fail because the returned date is one day more than it should be.

This reverts commit 607d66116a67e5a13eb0d46076f26dedc988e6a4.

---
Hi all,

Reverting this commit allows the grub_cmd_date test to pass. It appears that this commit is (now) causing an off-by-1 error in QEMU and the latest OpenBIOS. What I'm unsure of is if the original commit is actually correct on real hardware and that potentially OpenBIOS has a bug.

Adrian and Daniel A, could you test the reverting of this commit on real hardware and see if date does in fact produce the expected date (and do current builds show a date one day ahead of what it should be)? Can anyone point to documentation saying that the original commit is in fact what should be done? If the issue is in OpenBIOS I'd like to have some documentation to back up a bug report.

Vladimir, do you have any thoughts on this?

Glenn

---
 grub-core/lib/ieee1275/datetime.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
grub-core/lib/ieee1275/datetime.c +2 -2
index b81fba2ed..74578f15a 100644
@@ -95,7 +95,7 @@ grub_get_datetime (struct grub_datetime *datetime)
 
   datetime->year = args.year;
   datetime->month = args.month;
-  datetime->day = args.day + 1;
+  datetime->day = args.day;
   datetime->hour = args.hour;
   datetime->minute = args.minute;
   datetime->second = args.second;
@@ -140,7 +140,7 @@ grub_set_datetime (struct grub_datetime *datetime)
 
   args.year = datetime->year;
   args.month = datetime->month;
-  args.day = datetime->day - 1;
+  args.day = datetime->day;
   args.hour = datetime->hour;
   args.minute = datetime->minute;
   args.second = datetime->second;
Signature · 1 line
2.27.0
#3 lore raw
Quoted 3 lines

I suspect its the same on KVM and TCG because the machine definition is the same and doesn't container either an RTC nor a CMOS. What confusing to me is, don't these machines have clocks?

They do have clocks! Sadly I know very little about the clocks. date works fine in Linux, so I suspect there's something where SLOF doesn't expose whatever it is that we need in the form we expect, or something like that.

I guess someone should hunt it down but given that timeouts and such work (so I guess we have a working relative timer), I can't say it's likely to be a huge priority for anyone...

Kind regards,
Daniel

Quoted 85 lines
Quoted 13 lines

I also tested an LPAR on a Power8 PowerVM machine. In this case, before the patch, grub printed:

grub> date

2022-02-09 04:31:10 Wednesday

whereas on booting:

[dja@sauce ~]$ date -u
Tue Feb 8 04:46:40 UTC 2022

Quoted 6 lines

After applying the patch, grub printed:

grub> date

2022-02-08 04:51:27 Tuesday

It seems the patch makes things better.

Great thanks Daniel for confirming this.

Glenn

Quoted 61 lines

Kind regards,
Daniel

Quoted 53 lines · Glenn Washburn

This is causing the test grub_cmd_date to fail because the returned date is one day more than it should be.

This reverts commit 607d66116a67e5a13eb0d46076f26dedc988e6a4.

---
Hi all,

Reverting this commit allows the grub_cmd_date test to pass. It appears that this commit is (now) causing an off-by-1 error in QEMU and the latest OpenBIOS. What I'm unsure of is if the original commit is actually correct on real hardware and that potentially OpenBIOS has a bug.

Adrian and Daniel A, could you test the reverting of this commit on real hardware and see if date does in fact produce the expected date (and do current builds show a date one day ahead of what it should be)? Can anyone point to documentation saying that the original commit is in fact what should be done? If the issue is in OpenBIOS I'd like to have some documentation to back up a bug report.

Vladimir, do you have any thoughts on this?

Glenn

---
 grub-core/lib/ieee1275/datetime.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
grub-core/lib/ieee1275/datetime.c +2 -2
index b81fba2ed..74578f15a 100644
@@ -95,7 +95,7 @@ grub_get_datetime (struct grub_datetime *datetime)
 
   datetime->year = args.year;
   datetime->month = args.month;
-  datetime->day = args.day + 1;
+  datetime->day = args.day;
   datetime->hour = args.hour;
   datetime->minute = args.minute;
   datetime->second = args.second;
@@ -140,7 +140,7 @@ grub_set_datetime (struct grub_datetime *datetime)
 
   args.year = datetime->year;
   args.month = datetime->month;
-  args.day = datetime->day - 1;
+  args.day = datetime->day;
   args.hour = datetime->hour;
   args.minute = datetime->minute;
   args.second = datetime->second;
Signature · 1 line
2.27.0
Reviewed-by
#4 lore raw
In reply to Daniel Axtens
Quoted 20 lines · Daniel Axtens

Hi,

I tested a pseries guest under:
- qemu + KVM on a Power8 host with a fairly modern qemu
- qemu + TCG

Here the 'date' command simply reports "error: no cmos found", so this patch will have no impact on those platforms.

I also tested an LPAR on a Power8 PowerVM machine. In this case, before the patch, grub printed:

grub> date

2022-02-09 04:31:10 Wednesday

whereas on booting:

[dja@sauce ~]$ date -u
Tue Feb 8 04:46:40 UTC 2022

Quoted 8 lines

After applying the patch, grub printed:

grub> date

2022-02-08 04:51:27 Tuesday

It seems the patch makes things better.

Daniel, thank you for testing this!

Daniel