drm/xe: Add DRIVER_DATE to coredump error.
Subsystems (2)
Rodrigo Vivi
maintainerPATCH 1/2 drm/xe: Add DRIVER_DATE to coredump error.
This might be an useful debugging information if the driver date is properly updated with a certain cadence. With this we know from which point in time in our development the driver was from.
--- drivers/gpu/drm/xe/xe_devcoredump.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/gpu/drm/xe/xe_devcoredump.c +2 -0
@@ -10,6 +10,7 @@#include <generated/utsrelease.h> #include "xe_device.h" +#include "xe_drv.h" #include "xe_engine.h" #include "xe_force_wake.h" #include "xe_gt.h"
@@ -83,6 +84,7 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,drm_printf (&p,"**** Xe Device Coredump ****\n" );drm_printf (&p,"kernel: " UTS_RELEASE "\n" );drm_printf (&p,"module: " KBUILD_MODNAME "\n" ); +drm_printf (&p,"driver date: " DRIVER_DATE "\n" ); ts =ktime_to_timespec64 (ss->snapshot_time);drm_printf (&p,"Snapshot time: %lld.%09ld\n" , ts.tv_sec, ts.tv_nsec);
Signature · 1 line
2.41.0
Rodrigo Vivi
maintainerPATCH 2/2 drm/xe: Update DRIVER_DATE to 20230726
It is time to update the long and forgotten initial driver date.
--- drivers/gpu/drm/xe/xe_drv.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/xe/xe_drv.h +1 -1
@@ -10,7 +10,7 @@#define DRIVER_NAME "xe"#define DRIVER_DESC "Intel Xe Graphics" -#define DRIVER_DATE "20201103" +#define DRIVER_DATE "20230726"/* Interface history: *
Signature · 1 line
2.41.0
Re: [Intel-xe] [PATCH 2/2] drm/xe: Update DRIVER_DATE to 20230726
Quoted 2 lines · Rodrigo Vivi
It is time to update the long and forgotten initial driver date.
That old date is way off, AFAIK this date is long prior to Xe development starting.
Anyways:
Quoted 21 lines
Signed-off-by Rodrigo Vivi <[email protected]>--- drivers/gpu/drm/xe/xe_drv.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/gpu/drm/xe/xe_drv.h +1
-1@@ -10,7 +10,7 @@#define DRIVER_NAME "xe"#define DRIVER_DESC "Intel Xe Graphics" -#define DRIVER_DATE "20201103" +#define DRIVER_DATE "20230726"/* Interface history: * Signature · 1 line
2.41.0
Vivi, Rodrigo
maintainerRe: [Intel-xe] [PATCH 2/2] drm/xe: Update DRIVER_DATE to 20230726
Quoted 6 lines · Matthew Brost
Quoted 2 lines · Rodrigo Vivi
It is time to update the long and forgotten initial driver date.
That old date is way off, AFAIK this date is long prior to Xe development starting.
indeed, that was probably a copy and paste from something older that remained.
we should probably make this into a fix up or later squash in the original patch before the first pull request...
Quoted 26 lines
Anyways:
Reviewed-by Matthew Brost <[email protected]>Quoted 22 lines
Signed-off-by Rodrigo Vivi <[email protected]>---
drivers/gpu/drm/xe/xe_drv.h | 2 +-1 file changed, 1 insertion(+), 1 deletion(-)drivers/gpu/drm/xe/xe_drv.h +0
-0@@ -10,7 +10,7 @@
#define DRIVER_NAME "xe" #define DRIVER_DESC "Intel Xe Graphics"drivers/gpu/drm/xe/xe_drv.h +1
-1-#define DRIVER_DATE "20201103" +#define DRIVER_DATE "20230726"
/* Interface history:
*Signature · 1 line
2.41.0
Quoted 5 lines · Rodrigo Vivi
This might be an useful debugging information if the driver date is properly updated with a certain cadence. With this we know from which point in time in our development the driver was from.
Make sense to have this, wondering what the cadence should be...
Anyways:
Quoted 28 lines
Signed-off-by Rodrigo Vivi <[email protected]>--- drivers/gpu/drm/xe/xe_devcoredump.c | 2 ++ 1 file changed, 2 insertions(+)drivers/gpu/drm/xe/xe_devcoredump.c +2
-0@@ -10,6 +10,7 @@#include <generated/utsrelease.h> #include "xe_device.h" +#include "xe_drv.h" #include "xe_engine.h" #include "xe_force_wake.h" #include "xe_gt.h" @@ -83,6 +84,7 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,drm_printf (&p,"**** Xe Device Coredump ****\n" );drm_printf (&p,"kernel: " UTS_RELEASE "\n" );drm_printf (&p,"module: " KBUILD_MODNAME "\n" ); +drm_printf (&p,"driver date: " DRIVER_DATE "\n" ); ts =ktime_to_timespec64 (ss->snapshot_time);drm_printf (&p,"Snapshot time: %lld.%09ld\n" , ts.tv_sec, ts.tv_nsec);Signature · 1 line
2.41.0
On Wed, 2023-07-26 at 19:29 +0000, Matthew Brost wrote:
On Wed, Jul 26, 2023 at 03:25:19PM -0400, Rodrigo Vivi wrote:
This might be an useful debugging information if the driver date is properly updated with a certain cadence. With this we know from which point in time in our development the driver was from.
Make sense to have this, wondering what the cadence should be...
before we are merged we should probably change the date along with our rebases.
after we are in tree probably after the latest pull request towards the next version.
Anyways:
--- drivers/gpu/drm/xe/xe_devcoredump.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/gpu/drm/xe/xe_devcoredump.c +2 -0
@@ -10,6 +10,7 @@#include <generated/utsrelease.h> #include "xe_device.h" +#include "xe_drv.h" #include "xe_engine.h" #include "xe_force_wake.h" #include "xe_gt.h"
@@ -83,6 +84,7 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,drm_printf (&p,"**** Xe Device Coredump ****\n" );drm_printf (&p,"kernel: " UTS_RELEASE "\n" );drm_printf (&p,"module: " KBUILD_MODNAME "\n" ); +drm_printf (&p,"driver date: " DRIVER_DATE "\n" ); ts =ktime_to_timespec64 (ss->snapshot_time);drm_printf (&p,"Snapshot time: %lld.%09ld\n" , ts.tv_sec, ts.tv_nsec);
--
2.41.0
Quoted 8 lines · Rodrigo Vivi
On Wed, 2023-07-26 at 19:29 +0000, Matthew Brost wrote:
On Wed, Jul 26, 2023 at 03:25:19PM -0400, Rodrigo Vivi wrote:
This might be an useful debugging information if the driver date is properly updated with a certain cadence. With this we know from which point in time in our development the driver was from.
Quoted 5 lines
Make sense to have this, wondering what the cadence should be...
before we are merged we should probably change the date along with our rebases.
after we are in tree probably after the latest pull request towards the next version.
does it really make sense though? I think this exists only because of a pre-git era. What's the problem with using the git short hash that will automatically be updated and accurate?
Lucas De Marchi
Quoted 34 lines
Anyways:
---
drivers/gpu/drm/xe/xe_devcoredump.c| 2 ++1 file changed, 2 insertions(+)drivers/gpu/drm/xe/xe_devcoredump.c +2
-1@@ -10,6 +10,7 @@#include <generated/utsrelease.h> #include "xe_device.h" +#include "xe_drv.h" #include "xe_engine.h" #include "xe_force_wake.h" #include "xe_gt.h" @@ -83,6 +84,7 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,drm_printf (&p,"**** Xe Device Coredump ****\n" );drm_printf (&p,"kernel: " UTS_RELEASE "\n" );drm_printf (&p,"module: " KBUILD_MODNAME "\n" ); +drm_printf (&p,"driver date: " DRIVER_DATE "\n" ); ts =ktime_to_timespec64 (ss->snapshot_time);drm_printf (&p,"Snapshot time: %lld.%09ld\n" , ts.tv_sec, ts.tv_nsec); --2.41.0
Rodrigo Vivi
maintainerQuoted 11 lines · Lucas De Marchi
Quoted 10 lines · Rodrigo Vivi
On Wed, 2023-07-26 at 19:29 +0000, Matthew Brost wrote:
On Wed, Jul 26, 2023 at 03:25:19PM -0400, Rodrigo Vivi wrote:
This might be an useful debugging information if the driver date is properly updated with a certain cadence. With this we know from which point in time in our development the driver was from.Make sense to have this, wondering what the cadence should be...
Quoted 8 lines
Quoted 3 lines
before we are merged we should probably change the date along with our rebases.
after we are in tree probably after the latest pull request towards the next version.
does it really make sense though? I think this exists only because of a pre-git era. What's the problem with using the git short hash that will automatically be updated and accurate?
well, we still have the drm->date to fill so we need to maintain it.
maybe we could do like nouveau?!
#ifdef GIT_REVISION
.date = GIT_REVISION,
#else
.date = DRIVER_DATE,
#endif
Quoted 37 lines
Lucas De Marchi
Quoted 34 lines
Anyways:
---
drivers/gpu/drm/xe/xe_devcoredump.c| 2 ++1 file changed, 2 insertions(+)drivers/gpu/drm/xe/xe_devcoredump.c +2
-1@@ -10,6 +10,7 @@#include <generated/utsrelease.h> #include "xe_device.h" +#include "xe_drv.h" #include "xe_engine.h" #include "xe_force_wake.h" #include "xe_gt.h" @@ -83,6 +84,7 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,drm_printf (&p,"**** Xe Device Coredump ****\n" );drm_printf (&p,"kernel: " UTS_RELEASE "\n" );drm_printf (&p,"module: " KBUILD_MODNAME "\n" ); +drm_printf (&p,"driver date: " DRIVER_DATE "\n" ); ts =ktime_to_timespec64 (ss->snapshot_time);drm_printf (&p,"Snapshot time: %lld.%09ld\n" , ts.tv_sec, ts.tv_nsec); --2.41.0
Quoted 23 lines · Rodrigo Vivi
Quoted 11 lines · Lucas De Marchi
Quoted 10 lines · Rodrigo Vivi
On Wed, 2023-07-26 at 19:29 +0000, Matthew Brost wrote:
On Wed, Jul 26, 2023 at 03:25:19PM -0400, Rodrigo Vivi wrote:
This might be an useful debugging information if the driver date is properly updated with a certain cadence. With this we know from which point in time in our development the driver was from.Make sense to have this, wondering what the cadence should be...
Quoted 8 lines
Quoted 3 lines
before we are merged we should probably change the date along with our rebases.
after we are in tree probably after the latest pull request towards the next version.
does it really make sense though? I think this exists only because of a pre-git era. What's the problem with using the git short hash that will automatically be updated and accurate?
well, we still have the
drm->dateto fill so we need to maintain it.
Quoted 7 lines
maybe we could do like nouveau?!
#ifdef
GIT_REVISION
.date =GIT_REVISION,
#else
.date =DRIVER_DATE,
#endif
Sounds better. I don't see any value on DRIVER_DATE and having to update it. Just setting to some dummy value when GIT_REVISION is not available would be ok
$ git grep "\.date\s*=" -- drivers/gpu/drm/ | grep 2023 $ git grep DRIVER_DATE -- drivers/gpu/drm/ | grep 2023
Tells me nobody really maintains that.
Lucas De Marchi
Quoted 38 lines
Quoted 38 lines
Lucas De Marchi
Quoted 34 lines
Anyways:
---
drivers/gpu/drm/xe/xe_devcoredump.c| 2 ++1 file changed, 2 insertions(+)drivers/gpu/drm/xe/xe_devcoredump.c +2
-1@@ -10,6 +10,7 @@#include <generated/utsrelease.h> #include "xe_device.h" +#include "xe_drv.h" #include "xe_engine.h" #include "xe_force_wake.h" #include "xe_gt.h" @@ -83,6 +84,7 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,drm_printf (&p,"**** Xe Device Coredump ****\n" );drm_printf (&p,"kernel: " UTS_RELEASE "\n" );drm_printf (&p,"module: " KBUILD_MODNAME "\n" ); +drm_printf (&p,"driver date: " DRIVER_DATE "\n" ); ts =ktime_to_timespec64 (ss->snapshot_time);drm_printf (&p,"Snapshot time: %lld.%09ld\n" , ts.tv_sec, ts.tv_nsec); --2.41.0
Matthew Brost
maintainerQuoted 32 lines · Lucas De Marchi
Quoted 23 lines · Rodrigo Vivi
Quoted 11 lines · Lucas De Marchi
Quoted 10 lines · Rodrigo Vivi
On Wed, 2023-07-26 at 19:29 +0000, Matthew Brost wrote:
On Wed, Jul 26, 2023 at 03:25:19PM -0400, Rodrigo Vivi wrote:
This might be an useful debugging information if the driver date is properly updated with a certain cadence. With this we know from which point in time in our development the driver was from.Make sense to have this, wondering what the cadence should be...
Quoted 8 lines
Quoted 3 lines
before we are merged we should probably change the date along with our rebases.
after we are in tree probably after the latest pull request towards the next version.
does it really make sense though? I think this exists only because of a pre-git era. What's the problem with using the git short hash that will automatically be updated and accurate?
well, we still have the
drm->dateto fill so we need to maintain it.Quoted 7 lines
maybe we could do like nouveau?!
#ifdef
GIT_REVISION
.date =GIT_REVISION,
#else
.date =DRIVER_DATE,
#endif
Quoted 9 lines
Sounds better. I don't see any value on
DRIVER_DATEand having to update it. Just setting to some dummy value whenGIT_REVISIONis not available would be ok$ git grep "\.date\s*=" -- drivers/gpu/drm/ | grep 2023 $ git grep DRIVER_DATE -- drivers/gpu/drm/ | grep 2023Tells me nobody really maintains that.
Well I think we should attempt to maintain this, a date is helpful to give us a very quick idea of when the driver is from without having to dig through the git logs.
My opinion is both of these patches are valid but let's add a 3rd patch which also includes the GIT_REVISION in the core dump and perhaps a 4th that makes the date plus GIT_REVISION easily available in debugfs entry or something.
Matt
Quoted 40 lines
Lucas De Marchi
Quoted 38 lines
Quoted 38 lines
Lucas De Marchi
Quoted 34 lines
Anyways:
---
drivers/gpu/drm/xe/xe_devcoredump.c| 2 ++1 file changed, 2 insertions(+)drivers/gpu/drm/xe/xe_devcoredump.c +2
-1@@ -10,6 +10,7 @@#include <generated/utsrelease.h> #include "xe_device.h" +#include "xe_drv.h" #include "xe_engine.h" #include "xe_force_wake.h" #include "xe_gt.h" @@ -83,6 +84,7 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,drm_printf (&p,"**** Xe Device Coredump ****\n" );drm_printf (&p,"kernel: " UTS_RELEASE "\n" );drm_printf (&p,"module: " KBUILD_MODNAME "\n" ); +drm_printf (&p,"driver date: " DRIVER_DATE "\n" ); ts =ktime_to_timespec64 (ss->snapshot_time);drm_printf (&p,"Snapshot time: %lld.%09ld\n" , ts.tv_sec, ts.tv_nsec); --2.41.0
Quoted 40 lines · Matthew Brost
Quoted 39 lines · Lucas De Marchi
Quoted 23 lines · Rodrigo Vivi
Quoted 11 lines · Lucas De Marchi
Quoted 10 lines · Rodrigo Vivi
On Wed, 2023-07-26 at 19:29 +0000, Matthew Brost wrote:
On Wed, Jul 26, 2023 at 03:25:19PM -0400, Rodrigo Vivi wrote:
This might be an useful debugging information if the driver date is properly updated with a certain cadence. With this we know from which point in time in our development the driver was from.Make sense to have this, wondering what the cadence should be...
Quoted 8 lines
Quoted 3 lines
before we are merged we should probably change the date along with our rebases.
after we are in tree probably after the latest pull request towards the next version.
does it really make sense though? I think this exists only because of a pre-git era. What's the problem with using the git short hash that will automatically be updated and accurate?
well, we still have the
drm->dateto fill so we need to maintain it.Quoted 7 lines
maybe we could do like nouveau?!
#ifdef
GIT_REVISION
.date =GIT_REVISION,
#else
.date =DRIVER_DATE,
#endifSounds better. I don't see any value on
DRIVER_DATEand having to update it. Just setting to some dummy value whenGIT_REVISIONis not available would be ok$ git grep "\.date\s*=" -- drivers/gpu/drm/ | grep 2023 $ git grep DRIVER_DATE -- drivers/gpu/drm/ | grep 2023
Quoted 6 lines
Quoted 2 lines
Tells me nobody really maintains that.
Well I think we should attempt to maintain this, a date is helpful to give us a very quick idea of when the driver is from without having to dig through the git logs.
I don't see how the date helps with that. Nobody really maintains the date as the grep above shows. Even if we decided "we are going to be the outlier and really maintain this because we like", then it would be a date every .... 3 months? How is that any useful for whom got the coredump? Once we are merged upstream, having the *kernel* version is maybe useful, but the date doesn't really tell much
Lucas De Marchi
Quoted 48 lines
My opinion is both of these patches are valid but let's add a 3rd patch which also includes the
GIT_REVISIONin the core dump and perhaps a 4th that makes the date plusGIT_REVISIONeasily available in debugfs entry or something.Matt
Quoted 40 lines
Lucas De Marchi
Quoted 38 lines
Quoted 38 lines
Lucas De Marchi
Quoted 34 lines
Anyways:
---
drivers/gpu/drm/xe/xe_devcoredump.c| 2 ++1 file changed, 2 insertions(+)drivers/gpu/drm/xe/xe_devcoredump.c +2
-1@@ -10,6 +10,7 @@#include <generated/utsrelease.h> #include "xe_device.h" +#include "xe_drv.h" #include "xe_engine.h" #include "xe_force_wake.h" #include "xe_gt.h" @@ -83,6 +84,7 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,drm_printf (&p,"**** Xe Device Coredump ****\n" );drm_printf (&p,"kernel: " UTS_RELEASE "\n" );drm_printf (&p,"module: " KBUILD_MODNAME "\n" ); +drm_printf (&p,"driver date: " DRIVER_DATE "\n" ); ts =ktime_to_timespec64 (ss->snapshot_time);drm_printf (&p,"Snapshot time: %lld.%09ld\n" , ts.tv_sec, ts.tv_nsec); --2.41.0
Rodrigo Vivi
maintainerQuoted 34 lines · Lucas De Marchi
Quoted 33 lines · Matthew Brost
Quoted 32 lines · Lucas De Marchi
Quoted 30 lines · Rodrigo Vivi
Quoted 11 lines · Lucas De Marchi
Quoted 10 lines · Rodrigo Vivi
On Wed, 2023-07-26 at 19:29 +0000, Matthew Brost wrote:
On Wed, Jul 26, 2023 at 03:25:19PM -0400, Rodrigo Vivi wrote:
This might be an useful debugging information if the driver date is properly updated with a certain cadence. With this we know from which point in time in our development the driver was from.Make sense to have this, wondering what the cadence should be...
Quoted 8 lines
Quoted 3 lines
before we are merged we should probably change the date along with our rebases.
after we are in tree probably after the latest pull request towards the next version.
does it really make sense though? I think this exists only because of a pre-git era. What's the problem with using the git short hash that will automatically be updated and accurate?
well, we still have the
drm->dateto fill so we need to maintain it.maybe we could do like nouveau?!
#ifdef
GIT_REVISION
.date =GIT_REVISION,
#else
.date =DRIVER_DATE,
#endif
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Sounds better. I don't see any value on
DRIVER_DATEand having to update it. Just setting to some dummy value whenGIT_REVISIONis not available would be ok$ git grep "\.date\s*=" -- drivers/gpu/drm/ | grep 2023 $ git grep DRIVER_DATE -- drivers/gpu/drm/ | grep 2023Tells me nobody really maintains that.
Fair point. Maybe this could be used to convince the drm level to get entirely rid of drm->date?
Quoted 6 lines
Quoted 5 lines
Quoted 1 line
Well I think we should attempt to maintain this, a date is helpful to give us a very quick idea of when the driver is from without having to dig through the git logs.
Quoted 6 lines
I don't see how the date helps with that. Nobody really maintains the date as the grep above shows. Even if we decided "we are going to be the outlier and really maintain this because we like", then it would be a date every .... 3 months? How is that any useful for whom got the coredump? Once we are merged upstream, having the *kernel* version is maybe useful, but the date doesn't really tell much
Well, some bug reports we receive might come from end users without git and from trees that are ports and not necessarily the mainline, so even the version wouldn't tell where actually that code came from...
But well, without a good maintenance of the date, that is useless anyway indeed.
So, okay, let's drop this patch for now... let's move with the other patch as a fixup at last to not start with an insanely old version?
Quoted 51 lines
Lucas De Marchi
Quoted 48 lines
My opinion is both of these patches are valid but let's add a 3rd patch which also includes the
GIT_REVISIONin the core dump and perhaps a 4th that makes the date plusGIT_REVISIONeasily available in debugfs entry or something.Matt
Quoted 40 lines
Lucas De Marchi
Quoted 38 lines
Quoted 38 lines
Lucas De Marchi
Quoted 34 lines
Anyways:
---
drivers/gpu/drm/xe/xe_devcoredump.c| 2 ++1 file changed, 2 insertions(+)drivers/gpu/drm/xe/xe_devcoredump.c +2
-1@@ -10,6 +10,7 @@#include <generated/utsrelease.h> #include "xe_device.h" +#include "xe_drv.h" #include "xe_engine.h" #include "xe_force_wake.h" #include "xe_gt.h" @@ -83,6 +84,7 @@ static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,drm_printf (&p,"**** Xe Device Coredump ****\n" );drm_printf (&p,"kernel: " UTS_RELEASE "\n" );drm_printf (&p,"module: " KBUILD_MODNAME "\n" ); +drm_printf (&p,"driver date: " DRIVER_DATE "\n" ); ts =ktime_to_timespec64 (ss->snapshot_time);drm_printf (&p,"Snapshot time: %lld.%09ld\n" , ts.tv_sec, ts.tv_nsec); --2.41.0
Patchwork
CI[Intel-xe] ✗ CI.Patch_applied: failure for series starting with [1/2] drm/xe: Add DRIVER_DATE to coredump error. (rev2)
In reply to Rodrigo Vivi== Series Details ==
== Series Details ==
Series: series starting with [1/2] drm/xe: Add DRIVER_DATE to coredump error. (rev2)
URL : https://patchwork.freedesktop.org/series/121383/
State : failure
== Summary ==
=== Applying kernel patches on branch 'drm-xe-next' with base: === Base commit: b8202a008 drm/xe: Always use xe_vm_queue_rebind_worker helper === git am output follows ===
error: patch failed: drivers/gpu/drm/xe/xe_devcoredump.c:83
error: drivers/gpu/drm/xe/xe_devcoredump.c: patch does not apply
hint: Use 'git am --show-current-patch' to see the failed patch
Applying: drm/xe: Add DRIVER_DATE to coredump error.
Patch failed at 0001 drm/xe: Add DRIVER_DATE to coredump error. When you have resolved this problem, run "git am --continue". If you prefer to skip this patch, run "git am --skip" instead. To restore the original branch and stop patching, run "git am --abort".