Skip to contentSkip to conversation

HW_LOCK Security Patches

PATCH In review Peter Antoine 7 participants 39 messages Apr 23, 2015 – May 15, 2015

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

cover HW_LOCK Security Patches

The following series of patches fix a number of security holes in the i915 driver (actually in drm but...). The first three patches remove the actual security issues that have been found. The last two patches make the functions optional for all drivers. The only driver that has this feature turned on is the Nouveau driver, thus hopefully not breaking that driver.

The patch set has been tested on the Intel platforms but has not been tested on the Nouveau driver. Hopefully someone with a working card and the right combination of drmlib can verify that the patches do what they say.

There is a i-g-t test that goes with this patchset, but that test SHOULD NOT be run before the kernel is patches as the test will crash the driver and/or make the kernel panic.

Peter.

Peter Antoine (5):
drm: Kernel Crash in drm_unlock
drm: Fixes unsafe deference in locks.
drm: Possible lock priority escalation.
drm: Make HW_LOCK access functions optional.
drm: Make Legacy Context access functions optional.

 drivers/gpu/drm/drm_context.c         | 50 ++++++++++++++++++++++++++++++++---
 drivers/gpu/drm/drm_drv.c             | 12 +++++----
 drivers/gpu/drm/drm_lock.c            | 18 +++++++++++--
 drivers/gpu/drm/i915/i915_dma.c       |  3 +++
 drivers/gpu/drm/nouveau/nouveau_drm.c |  3 ++-
 include/drm/drmP.h                    | 23 ++++++++--------
 include/uapi/drm/i915_drm.h           |  1 +
 7 files changed, 87 insertions(+), 23 deletions(-)
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw

PATCH 1/5 drm: Kernel Crash in drm_unlock

This patch fixes a possible kernel crash when drm_unlock (DRM_IOCTL_UNLOCK) is called by a application that has not had a lock created by it. This crash can be caused by any application from all users.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_lock.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/gpu/drm/drm_lock.c +8 -0
index f861361..070dd5d 100644
@@ -159,6 +159,14 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 		return -EINVAL;
 	}
 
+	if (!master->lock.hw_lock) {
+		DRM_ERROR(
+			"Device has been unregistered. Hard exit. Process %d\n",
+			task_pid_nr(current));
+		send_sig(SIGTERM, current, 0);
+		return -EPERM;
+	}
+
 	if (drm_legacy_lock_free(&master->lock, lock->context)) {
 		/* FIXME: Should really bail out here. */
 	}
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#2 lore raw

Re: [Intel-gfx] [PATCH 1/5] drm: Kernel Crash in drm_unlock

Quoted 3 lines · Peter Antoine

This patch fixes a possible kernel crash when drm_unlock (DRM_IOCTL_UNLOCK) is called by a application that has not had a lock created by it. This crash can be caused by any application from all users.

Crashing the application is still a crash...
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#3 lore raw

Re: [PATCH 1/5] drm: Kernel Crash in drm_unlock

Before the patch the system required rebooting (driver crash and/or kernel panic). Now the application gets terminated.

This follows the pattern in the other parts of this code that checks that pointer.

Peter.

-----Original Message-----
From: Chris Wilson [mailto:[email protected]]
Sent: Thursday, April 23, 2015 3:20 PM
To: Antoine, Peter
Cc: [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [Intel-gfx] [PATCH 1/5] drm: Kernel Crash in drm_unlock

Quoted 3 lines · Peter Antoine

This patch fixes a possible kernel crash when drm_unlock (DRM_IOCTL_UNLOCK) is called by a application that has not had a lock created by it. This crash can be caused by any application from all users.

Crashing the application is still a crash...
-Chris

--
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#4 lore raw

Re: [Intel-gfx] [PATCH 1/5] drm: Kernel Crash in drm_unlock

Quoted 2 lines · Peter

Before the patch the system required rebooting (driver crash and/or kernel panic). Now the application gets terminated.

It should have an GPF which should not have required a reboot, but terminated the application. Unless you set it to automatically reboot. -Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#5 lore raw

Re: [PATCH 1/5] drm: Kernel Crash in drm_unlock

I picked up this work due to the following Jira ticket created by the security team (on Android) and was asked to give it a second look and found a few more issues with the hw lock code.

https://jira01.devtools.intel.com/browse/GMINL-5388
I/O control on /dev/dri/card0 crashes the kernel (0x4008642b)

It also stops Linux as it kills the driver, I guess it might be possible to reload the gfx driver. On a unpatched system the test that is included in the issue or the igt test that has been posted for the issue will show the problem.

I ran the test on an unpatched system here and the gui stopped and the keyboard stopped responding, so I rebooted. With the patched system I did not need to reboot.

Should I change the SIGTERM to SIGSEGV, not quite the same thing but tooling is better at handling a segfault than a SIGTERM and the application that calls this IOCTL is using an uninitialised hw lock so it is kind of the same as differencing an uninitialised pointer (kind of). Or, I could just remove it, but the bug has been in the code for at least two years (and known about), and I would guess that any code that is calling this is fuzzing the IOCTLs (as this is how the security team found it) and we should reward them with a application exit.

Peter.

Quoted 8 lines · Chris Wilson
Quoted 2 lines · Peter

Before the patch the system required rebooting (driver crash and/or kernel panic). Now the application gets terminated.

It should have an GPF which should not have required a reboot, but terminated the application. Unless you set it to automatically reboot. -Chris

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#6 lore raw

Re: [PATCH 1/5] drm: Kernel Crash in drm_unlock

Quoted 25 lines · Peter

I picked up this work due to the following Jira ticket created by the security team (on Android) and was asked to give it a second look and found a few more issues with the hw lock code.

https://jira01.devtools.intel.com/browse/GMINL-5388
I/O control on /dev/dri/card0 crashes the kernel (0x4008642b)

It also stops Linux as it kills the driver, I guess it might be possible to reload the gfx driver. On a unpatched system the test that is included in the issue or the igt test that has been posted for the issue will show the problem.

I ran the test on an unpatched system here and the gui stopped and the keyboard stopped responding, so I rebooted. With the patched system I did not need to reboot.

Should I change the SIGTERM to SIGSEGV, not quite the same thing but tooling is better at handling a segfault than a SIGTERM and the application that calls this IOCTL is using an uninitialised hw lock so it is kind of the same as differencing an uninitialised pointer (kind of). Or, I could just remove it, but the bug has been in the code for at least two years (and known about), and I would guess that any code that is calling this is fuzzing the IOCTLs (as this is how the security team found it) and we should reward them with a application exit.

Quoted 1 line

Peter.

SIGSEGV would be a better choice.

SIGTERM is normally sent by a user -- it's the default signal sent by kill(1). It's also commonly used to tell a long-running daemon process to tidy up and exit cleanly.

SIGSEGV commonly means "you accessed something that doesn't exist/isn't mapped/you don't have permissions for". There are specific subcases that can be indicated via the siginfo data; this is from the sigaction(1) manpage:

The following values can be placed in si_code for a SIGSEGV signal:

SEGV_MAPERR address not mapped to object

SEGV_ACCERR invalid permissions for mapped object

SIGBUS would also be a possibility but that's generally taken to mean that an access got all the way to some physical bus and then faulted, whereas SIGSEGV suggests the access was rejected during the virtual-to-physical mapping process.

.Dave.
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#7 lore raw

Re: [PATCH 1/5] drm: Kernel Crash in drm_unlock

Quoted 41 lines · Dave Gordon
Quoted 25 lines · Peter

I picked up this work due to the following Jira ticket created by the security team (on Android) and was asked to give it a second look and found a few more issues with the hw lock code.

https://jira01.devtools.intel.com/browse/GMINL-5388
I/O control on /dev/dri/card0 crashes the kernel (0x4008642b)

It also stops Linux as it kills the driver, I guess it might be possible to reload the gfx driver. On a unpatched system the test that is included in the issue or the igt test that has been posted for the issue will show the problem.

I ran the test on an unpatched system here and the gui stopped and the keyboard stopped responding, so I rebooted. With the patched system I did not need to reboot.

Should I change the SIGTERM to SIGSEGV, not quite the same thing but tooling is better at handling a segfault than a SIGTERM and the application that calls this IOCTL is using an uninitialised hw lock so it is kind of the same as differencing an uninitialised pointer (kind of). Or, I could just remove it, but the bug has been in the code for at least two years (and known about), and I would guess that any code that is calling this is fuzzing the IOCTLs (as this is how the security team found it) and we should reward them with a application exit.

Quoted 1 line

Peter.

SIGSEGV would be a better choice.

SIGTERM is normally sent by a user -- it's the default signal sent by kill(1). It's also commonly used to tell a long-running daemon process to tidy up and exit cleanly.

SIGSEGV commonly means "you accessed something that doesn't exist/isn't mapped/you don't have permissions for". There are specific subcases that can be indicated via the siginfo data; this is from the sigaction(1) manpage:

The following values can be placed in si_code for a SIGSEGV signal:

Quoted 8 lines

SEGV_MAPERR address not mapped to object

SEGV_ACCERR invalid permissions for mapped object

SIGBUS would also be a possibility but that's generally taken to mean that an access got all the way to some physical bus and then faulted, whereas SIGSEGV suggests the access was rejected during the virtual-to-physical mapping process.

None of the above. Just return -EINVAL, -EPERM, -EACCESS as appropriate. -Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#8 lore raw

Re: [PATCH 1/5] drm: Kernel Crash in drm_unlock

Quoted 44 lines · On Tue, Apr 28, 2015 at 10:52:32AM +0100, [email protected] wrote:
Quoted 43 lines · Dave Gordon
Quoted 25 lines · Peter

I picked up this work due to the following Jira ticket created by the security team (on Android) and was asked to give it a second look and found a few more issues with the hw lock code.

https://jira01.devtools.intel.com/browse/GMINL-5388
I/O control on /dev/dri/card0 crashes the kernel (0x4008642b)

It also stops Linux as it kills the driver, I guess it might be possible to reload the gfx driver. On a unpatched system the test that is included in the issue or the igt test that has been posted for the issue will show the problem.

I ran the test on an unpatched system here and the gui stopped and the keyboard stopped responding, so I rebooted. With the patched system I did not need to reboot.

Should I change the SIGTERM to SIGSEGV, not quite the same thing but tooling is better at handling a segfault than a SIGTERM and the application that calls this IOCTL is using an uninitialised hw lock so it is kind of the same as differencing an uninitialised pointer (kind of). Or, I could just remove it, but the bug has been in the code for at least two years (and known about), and I would guess that any code that is calling this is fuzzing the IOCTLs (as this is how the security team found it) and we should reward them with a application exit.

Quoted 1 line

Peter.

SIGSEGV would be a better choice.

SIGTERM is normally sent by a user -- it's the default signal sent by kill(1). It's also commonly used to tell a long-running daemon process to tidy up and exit cleanly.

SIGSEGV commonly means "you accessed something that doesn't exist/isn't mapped/you don't have permissions for". There are specific subcases that can be indicated via the siginfo data; this is from the sigaction(1) manpage:

The following values can be placed in si_code for a SIGSEGV signal:

SEGV_MAPERR address not mapped to object

Quoted 8 lines
Quoted 6 lines

SEGV_ACCERR invalid permissions for mapped object

SIGBUS would also be a possibility but that's generally taken to mean that an access got all the way to some physical bus and then faulted, whereas SIGSEGV suggests the access was rejected during the virtual-to-physical mapping process.

None of the above. Just return -EINVAL, -EPERM, -EACCESS as appropriate.

Seconded, we really don't want to be in the business of fixing up the drm design mistakes of the past 15 years. As long as we can fully lock out this particular dragon when running i915 we're imo good enough. The dri1 design of a kernel shim driver cooperating with the ums driver for hw ownership is fundamentally unfixable.

Also we can't change any of it for drivers actually using it since it'll break them, which is a big no-go.
-Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#9 lore raw

Re: [PATCH 1/5] drm: Kernel Crash in drm_unlock

Quoted 60 lines · Daniel Vetter
Quoted 44 lines · On Tue, Apr 28, 2015 at 10:52:32AM +0100, [email protected] wrote:
Quoted 43 lines · Dave Gordon
Quoted 25 lines · Peter

I picked up this work due to the following Jira ticket created by the security team (on Android) and was asked to give it a second look and found a few more issues with the hw lock code.

https://jira01.devtools.intel.com/browse/GMINL-5388
I/O control on /dev/dri/card0 crashes the kernel (0x4008642b)

It also stops Linux as it kills the driver, I guess it might be possible to reload the gfx driver. On a unpatched system the test that is included in the issue or the igt test that has been posted for the issue will show the problem.

I ran the test on an unpatched system here and the gui stopped and the keyboard stopped responding, so I rebooted. With the patched system I did not need to reboot.

Should I change the SIGTERM to SIGSEGV, not quite the same thing but tooling is better at handling a segfault than a SIGTERM and the application that calls this IOCTL is using an uninitialised hw lock so it is kind of the same as differencing an uninitialised pointer (kind of). Or, I could just remove it, but the bug has been in the code for at least two years (and known about), and I would guess that any code that is calling this is fuzzing the IOCTLs (as this is how the security team found it) and we should reward them with a application exit.

Quoted 1 line

Peter.

SIGSEGV would be a better choice.

SIGTERM is normally sent by a user -- it's the default signal sent by kill(1). It's also commonly used to tell a long-running daemon process to tidy up and exit cleanly.

SIGSEGV commonly means "you accessed something that doesn't exist/isn't mapped/you don't have permissions for". There are specific subcases that can be indicated via the siginfo data; this is from the sigaction(1) manpage:

The following values can be placed in si_code for a SIGSEGV signal:

SEGV_MAPERR address not mapped to object

Quoted 8 lines
Quoted 6 lines

SEGV_ACCERR invalid permissions for mapped object

SIGBUS would also be a possibility but that's generally taken to mean that an access got all the way to some physical bus and then faulted, whereas SIGSEGV suggests the access was rejected during the virtual-to-physical mapping process.

None of the above. Just return -EINVAL, -EPERM, -EACCESS as appropriate.

Seconded, we really don't want to be in the business of fixing up the drm design mistakes of the past 15 years. As long as we can fully lock out this particular dragon when running i915 we're imo good enough. The dri1 design of a kernel shim driver cooperating with the ums driver for hw ownership is fundamentally unfixable.

Quoted 3 lines

Also we can't change any of it for drivers actually using it since it'll break them, which is a big no-go.
-Daniel

I will remove it. But, If you are using this code path the driver/kernel will have crashed. It covers a NULL pointer deference, so we are not changing the API that anyone is actually using.

Peter.

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#10 lore raw

Re: [PATCH 1/5] drm: Kernel Crash in drm_unlock

Quoted 61 lines · Peter
Quoted 60 lines · Daniel Vetter
Quoted 44 lines · On Tue, Apr 28, 2015 at 10:52:32AM +0100, [email protected] wrote:
Quoted 43 lines · Dave Gordon
Quoted 25 lines · Peter

I picked up this work due to the following Jira ticket created by the security team (on Android) and was asked to give it a second look and found a few more issues with the hw lock code.

https://jira01.devtools.intel.com/browse/GMINL-5388
I/O control on /dev/dri/card0 crashes the kernel (0x4008642b)

It also stops Linux as it kills the driver, I guess it might be possible to reload the gfx driver. On a unpatched system the test that is included in the issue or the igt test that has been posted for the issue will show the problem.

I ran the test on an unpatched system here and the gui stopped and the keyboard stopped responding, so I rebooted. With the patched system I did not need to reboot.

Should I change the SIGTERM to SIGSEGV, not quite the same thing but tooling is better at handling a segfault than a SIGTERM and the application that calls this IOCTL is using an uninitialised hw lock so it is kind of the same as differencing an uninitialised pointer (kind of). Or, I could just remove it, but the bug has been in the code for at least two years (and known about), and I would guess that any code that is calling this is fuzzing the IOCTLs (as this is how the security team found it) and we should reward them with a application exit.

Quoted 1 line

Peter.

SIGSEGV would be a better choice.

SIGTERM is normally sent by a user -- it's the default signal sent by kill(1). It's also commonly used to tell a long-running daemon process to tidy up and exit cleanly.

SIGSEGV commonly means "you accessed something that doesn't exist/isn't mapped/you don't have permissions for". There are specific subcases that can be indicated via the siginfo data; this is from the sigaction(1) manpage:

The following values can be placed in si_code for a SIGSEGV signal:

SEGV_MAPERR address not mapped to object

Quoted 8 lines
Quoted 6 lines

SEGV_ACCERR invalid permissions for mapped object

SIGBUS would also be a possibility but that's generally taken to mean that an access got all the way to some physical bus and then faulted, whereas SIGSEGV suggests the access was rejected during the virtual-to-physical mapping process.

None of the above. Just return -EINVAL, -EPERM, -EACCESS as appropriate.

Seconded, we really don't want to be in the business of fixing up the drm design mistakes of the past 15 years. As long as we can fully lock out this particular dragon when running i915 we're imo good enough. The dri1 design of a kernel shim driver cooperating with the ums driver for hw ownership is fundamentally unfixable.

Quoted 7 lines
Quoted 3 lines

Also we can't change any of it for drivers actually using it since it'll break them, which is a big no-go.
-Daniel

I will remove it. But, If you are using this code path the driver/kernel will have crashed. It covers a NULL pointer deference, so we are not changing the API that anyone is actually using.

With ums/dri1 userspace can crash the kernel. That's pretty much by design, and it's pretty much unfixable. Trying to fix up the obvious ones like here is just cargo-culting. The only real option is to make absolutely sure we can't ever reach these codepaths at all from kms drivers like i915. Which tends to be a lot more work for auditing, but will be much more useful.
-Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#11 lore raw

Re: [PATCH 1/5] drm: Kernel Crash in drm_unlock

In reply to Dave Gordon
Quoted 26 lines · Dave Gordon
Quoted 25 lines · Peter

I picked up this work due to the following Jira ticket created by the security team (on Android) and was asked to give it a second look and found a few more issues with the hw lock code.

https://jira01.devtools.intel.com/browse/GMINL-5388
I/O control on /dev/dri/card0 crashes the kernel (0x4008642b)

It also stops Linux as it kills the driver, I guess it might be possible to reload the gfx driver. On a unpatched system the test that is included in the issue or the igt test that has been posted for the issue will show the problem.

I ran the test on an unpatched system here and the gui stopped and the keyboard stopped responding, so I rebooted. With the patched system I did not need to reboot.

Should I change the SIGTERM to SIGSEGV, not quite the same thing but tooling is better at handling a segfault than a SIGTERM and the application that calls this IOCTL is using an uninitialised hw lock so it is kind of the same as differencing an uninitialised pointer (kind of). Or, I could just remove it, but the bug has been in the code for at least two years (and known about), and I would guess that any code that is calling this is fuzzing the IOCTLs (as this is how the security team found it) and we should reward them with a application exit.

Quoted 3 lines
Quoted 1 line

Peter.

SIGSEGV would be a better choice.

[snip]

Nope, I've changed my mind about this. I thought the problematic case was just a process releasing the lock without having acquired it, but on further examination it's really that the DRM master process has gone away or otherwise deleted (or never created?) the lock. And THEN the (non-master?) process tries to release the (now) non-existent lock.

But more importantly, there's existing code in the acquire-lock path that sends SIGTERM for this case, so obviously the release-lock code should be as similar as possible.

.Dave.
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#12 lore raw

PATCH 2/5 drm: Fixes unsafe deference in locks.

In reply to Peter Antoine

This patch fixes an unsafe deference in the DRM_IOCTL_NEW_CTX. If the ioctl is called before the lock is created or after it has been destroyed. The code will deference a NULL pointer. This ioctl is a root ioctl so exploitation is limited.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_context.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/gpu/drm/drm_context.c +7 -1
index 9b23525..96350d1 100644
@@ -277,7 +277,13 @@ static int drm_context_switch_complete(struct drm_device *dev,
 {
 	dev->last_context = new;	/* PRE/POST: This is the _only_ writer. */
 
-	if (!_DRM_LOCK_IS_HELD(file_priv->master->lock.hw_lock->lock)) {
+	if (file_priv->master->lock.hw_lock == NULL) {
+		DRM_ERROR(
+			"Device has been unregistered. Hard exit. Process %d\n",
+			task_pid_nr(current));
+		send_sig(SIGTERM, current, 0);
+		return -EPERM;
+	} else if (!_DRM_LOCK_IS_HELD(file_priv->master->lock.hw_lock->lock)) {
 		DRM_ERROR("Lock isn't held after context switch\n");
 	}
 
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#13 lore raw

Re: [Intel-gfx] [PATCH 2/5] drm: Fixes unsafe deference in locks.

Quoted 4 lines · Peter Antoine

This patch fixes an unsafe deference in the DRM_IOCTL_NEW_CTX. If the ioctl is called before the lock is created or after it has been destroyed. The code will deference a NULL pointer. This ioctl is a root ioctl so exploitation is limited.

You've turned an application crash into an application crash... Just with a slightly less verbose kernel log.
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#14 lore raw

PATCH 3/5 drm: Possible lock priority escalation.

In reply to Peter Antoine

If an application that has a driver lock created, wants the lock the kernel context, it is not allowed to. If the call to drm_lock has a context of 0, it is rejected. If you set the context to _DRM_LOCK_CONT then call drm lock, it will pass the context == DRM_KERNEL_CONTEXT checks. But as the DRM_LOCK_CONT bits are not part of the context id this allows operations on the DRM_KERNEL_CONTEXT.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_context.c | 6 +++---
 drivers/gpu/drm/drm_lock.c    | 4 ++--
 2 files changed, 5 insertions(+), 5 deletions(-)
drivers/gpu/drm/drm_context.c +3 -3
index 96350d1..1febcd3 100644
@@ -123,7 +123,7 @@ void drm_legacy_ctxbitmap_flush(struct drm_device *dev, struct drm_file *file)
 
 	list_for_each_entry_safe(pos, tmp, &dev->ctxlist, head) {
 		if (pos->tag == file &&
-		    pos->handle != DRM_KERNEL_CONTEXT) {
+		    _DRM_LOCKING_CONTEXT(pos->handle) != DRM_KERNEL_CONTEXT) {
 			if (dev->driver->context_dtor)
 				dev->driver->context_dtor(dev, pos->handle);
 
@@ -342,7 +342,7 @@ int drm_legacy_addctx(struct drm_device *dev, void *data,
 	struct drm_ctx *ctx = data;
 
 	ctx->handle = drm_legacy_ctxbitmap_next(dev);
-	if (ctx->handle == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(ctx->handle) == DRM_KERNEL_CONTEXT) {
 		/* Skip kernel's context and get a new one. */
 		ctx->handle = drm_legacy_ctxbitmap_next(dev);
 	}
@@ -449,7 +449,7 @@ int drm_legacy_rmctx(struct drm_device *dev, void *data,
 	struct drm_ctx *ctx = data;
 
 	DRM_DEBUG("%d\n", ctx->handle);
-	if (ctx->handle != DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(ctx->handle) != DRM_KERNEL_CONTEXT) {
 		if (dev->driver->context_dtor)
 			dev->driver->context_dtor(dev, ctx->handle);
 		drm_legacy_ctxbitmap_free(dev, ctx->handle);
drivers/gpu/drm/drm_lock.c +2 -2
index 070dd5d..94500930 100644
@@ -63,7 +63,7 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 
 	++file_priv->lock_count;
 
-	if (lock->context == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
 		return -EINVAL;
@@ -153,7 +153,7 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
-	if (lock->context == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
 		return -EINVAL;
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#15 lore raw

Re: [Intel-gfx] [PATCH 3/5] drm: Possible lock priority escalation.

Quoted 37 lines · Peter Antoine

If an application that has a driver lock created, wants the lock the kernel context, it is not allowed to. If the call to drm_lock has a context of 0, it is rejected. If you set the context to _DRM_LOCK_CONT then call drm lock, it will pass the context == DRM_KERNEL_CONTEXT checks. But as the DRM_LOCK_CONT bits are not part of the context id this allows operations on the DRM_KERNEL_CONTEXT.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_context.c | 6 +++---
 drivers/gpu/drm/drm_lock.c    | 4 ++--
 2 files changed, 5 insertions(+), 5 deletions(-)
drivers/gpu/drm/drm_context.c +2 -2
index 96350d1..1febcd3 100644
@@ -123,7 +123,7 @@ void drm_legacy_ctxbitmap_flush(struct drm_device *dev, struct drm_file *file)
 
 	list_for_each_entry_safe(pos, tmp, &dev->ctxlist, head) {
 		if (pos->tag == file &&
-		    pos->handle != DRM_KERNEL_CONTEXT) {
+		    _DRM_LOCKING_CONTEXT(pos->handle) != DRM_KERNEL_CONTEXT) {
 			if (dev->driver->context_dtor)
 				dev->driver->context_dtor(dev, pos->handle);
 
@@ -342,7 +342,7 @@ int drm_legacy_addctx(struct drm_device *dev, void *data,
 	struct drm_ctx *ctx = data;
 
 	ctx->handle = drm_legacy_ctxbitmap_next(dev);
-	if (ctx->handle == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(ctx->handle) == DRM_KERNEL_CONTEXT) {
 		/* Skip kernel's context and get a new one. */
 		ctx->handle = drm_legacy_ctxbitmap_next(dev);
 	}

drivers/gpu/drm/drm_context.c int drm_legacy_rmctx(struct drm_device *dev, void *data, · line 455

Quoted 8 lines
drivers/gpu/drm/drm_context.c +1 -1
@@ -449,7 +449,7 @@ int drm_legacy_rmctx(struct drm_device *dev, void *data,
 	struct drm_ctx *ctx = data;
 
 	DRM_DEBUG("%d\n", ctx->handle);
-	if (ctx->handle != DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(ctx->handle) != DRM_KERNEL_CONTEXT) {
 		if (dev->driver->context_dtor)
 			dev->driver->context_dtor(dev, ctx->handle);
 		drm_legacy_ctxbitmap_free(dev, ctx->handle);

How about just fixing the end parameter passed to idr_alloc()? AFAICS that would take care of the context code.

Well, there are a few more issues with the code:
- not properly checking for negative return value from idr_alloc()
- leaking the ctx id on kmalloc() error
- pointless check for idr_alloc() returning 0 even though the min is 1

drivers/gpu/drm/drm_lock.c int drm_legacy_lock(struct drm_device *dev, void *data, · line 64

Quoted 7 lines
drivers/gpu/drm/drm_lock.c +0 -0
index 070dd5d..94500930 100644
@@ -63,7 +63,7 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 
 	++file_priv->lock_count;

While you're poking around this dungeopn, maybe you can kill lock_count? We never seem to decrement it, and it's only checked in drm_legacy_i_have_hw_lock().

Quoted 7 lines
drivers/gpu/drm/drm_lock.c +1 -1
-	if (lock->context == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
 		return -EINVAL;

drivers/gpu/drm/drm_lock.c int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_ · line 159

Quoted 8 lines
drivers/gpu/drm/drm_lock.c +1 -1
@@ -153,7 +153,7 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
-	if (lock->context == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
 		return -EINVAL;

These two changes look OK to me.

Quoted 7 lines
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 6 lines
Ville Syrjälä
Intel OTC
_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#16 lore raw

Re: [Intel-gfx] [PATCH 3/5] drm: Possible lock priority escalation.

Quoted 10 lines · Ville Syrjälä
Quoted 9 lines · Peter Antoine

If an application that has a driver lock created, wants the lock the kernel context, it is not allowed to. If the call to drm_lock has a context of 0, it is rejected. If you set the context to _DRM_LOCK_CONT then call drm lock, it will pass the context == DRM_KERNEL_CONTEXT checks. But as the DRM_LOCK_CONT bits are not part of the context id this allows operations on the DRM_KERNEL_CONTEXT.

Issue: VIZ-5485

If you're touching code with drm_legacy_ prefix of in such a file you've ended up in the horrible corners of the dri1 dungeons and should head back out pronto ;-)

If we can actually run into this code on production i915 then we need to improve the locks at the door of these dungeons for kms drivers, not try to fix up the mess behind them. That's just plain impossible.

If you want to make really sure we get this right some simple drm igt tests to make sure these codepaths are really dead for kms driver might be good. But otherwise we really can only annotate this as wontfix in code security issue scanners.
-Daniel

Quoted 85 lines
Quoted 28 lines
---
 drivers/gpu/drm/drm_context.c | 6 +++---
 drivers/gpu/drm/drm_lock.c    | 4 ++--
 2 files changed, 5 insertions(+), 5 deletions(-)
drivers/gpu/drm/drm_context.c +2 -2
index 96350d1..1febcd3 100644
@@ -123,7 +123,7 @@ void drm_legacy_ctxbitmap_flush(struct drm_device *dev, struct drm_file *file)
 
 	list_for_each_entry_safe(pos, tmp, &dev->ctxlist, head) {
 		if (pos->tag == file &&
-		    pos->handle != DRM_KERNEL_CONTEXT) {
+		    _DRM_LOCKING_CONTEXT(pos->handle) != DRM_KERNEL_CONTEXT) {
 			if (dev->driver->context_dtor)
 				dev->driver->context_dtor(dev, pos->handle);
 
@@ -342,7 +342,7 @@ int drm_legacy_addctx(struct drm_device *dev, void *data,
 	struct drm_ctx *ctx = data;
 
 	ctx->handle = drm_legacy_ctxbitmap_next(dev);
-	if (ctx->handle == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(ctx->handle) == DRM_KERNEL_CONTEXT) {
 		/* Skip kernel's context and get a new one. */
 		ctx->handle = drm_legacy_ctxbitmap_next(dev);
 	}

drivers/gpu/drm/drm_context.c int drm_legacy_rmctx(struct drm_device *dev, void *data, · line 455

Quoted 8 lines
drivers/gpu/drm/drm_context.c +1 -1
@@ -449,7 +449,7 @@ int drm_legacy_rmctx(struct drm_device *dev, void *data,
 	struct drm_ctx *ctx = data;
 
 	DRM_DEBUG("%d\n", ctx->handle);
-	if (ctx->handle != DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(ctx->handle) != DRM_KERNEL_CONTEXT) {
 		if (dev->driver->context_dtor)
 			dev->driver->context_dtor(dev, ctx->handle);
 		drm_legacy_ctxbitmap_free(dev, ctx->handle);

How about just fixing the end parameter passed to idr_alloc()? AFAICS that would take care of the context code.

Well, there are a few more issues with the code:
- not properly checking for negative return value from idr_alloc()
- leaking the ctx id on kmalloc() error
- pointless check for idr_alloc() returning 0 even though the min is 1

drivers/gpu/drm/drm_lock.c int drm_legacy_lock(struct drm_device *dev, void *data, · line 64

Quoted 7 lines
drivers/gpu/drm/drm_lock.c +0 -0
index 070dd5d..94500930 100644
@@ -63,7 +63,7 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 
 	++file_priv->lock_count;

While you're poking around this dungeopn, maybe you can kill lock_count? We never seem to decrement it, and it's only checked in drm_legacy_i_have_hw_lock().

Quoted 7 lines
drivers/gpu/drm/drm_lock.c +1 -1
-	if (lock->context == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
 		return -EINVAL;

drivers/gpu/drm/drm_lock.c int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_ · line 159

Quoted 8 lines
drivers/gpu/drm/drm_lock.c +1 -1
@@ -153,7 +153,7 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
-	if (lock->context == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
 		return -EINVAL;

These two changes look OK to me.

Quoted 7 lines
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 2 lines
Ville Syrjälä
Intel OTC
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#17 lore raw

Re: [PATCH 3/5] drm: Possible lock priority escalation.

Quoted 20 lines · Daniel Vetter
Quoted 10 lines · Ville Syrjälä
Quoted 9 lines · Peter Antoine

If an application that has a driver lock created, wants the lock the kernel context, it is not allowed to. If the call to drm_lock has a context of 0, it is rejected. If you set the context to _DRM_LOCK_CONT then call drm lock, it will pass the context == DRM_KERNEL_CONTEXT checks. But as the DRM_LOCK_CONT bits are not part of the context id this allows operations on the DRM_KERNEL_CONTEXT.

Issue: VIZ-5485

If you're touching code with drm_legacy_ prefix of in such a file you've ended up in the horrible corners of the dri1 dungeons and should head back out pronto ;-)

If we can actually run into this code on production i915 then we need to improve the locks at the door of these dungeons for kms drivers, not try to fix up the mess behind them. That's just plain impossible.

Quoted 6 lines

If you want to make really sure we get this right some simple drm igt tests to make sure these codepaths are really dead for kms driver might be good. But otherwise we really can only annotate this as wontfix in code security issue scanners.
-Daniel

There is a test that covers this fix. This is a simple three line fix that stops a userspace driver locking the kernel context. Yes they are other problems with this code, but why are they stopping this patch that does a simple fix from going in?

I'll happily drop this patch if it causes more problems that it fixes.

Peter.

Quoted 77 lines
Quoted 77 lines
Quoted 28 lines
---
 drivers/gpu/drm/drm_context.c | 6 +++---
 drivers/gpu/drm/drm_lock.c    | 4 ++--
 2 files changed, 5 insertions(+), 5 deletions(-)
drivers/gpu/drm/drm_context.c +2 -2
index 96350d1..1febcd3 100644
@@ -123,7 +123,7 @@ void drm_legacy_ctxbitmap_flush(struct drm_device *dev, struct drm_file *file)
 
 	list_for_each_entry_safe(pos, tmp, &dev->ctxlist, head) {
 		if (pos->tag == file &&
-		    pos->handle != DRM_KERNEL_CONTEXT) {
+		    _DRM_LOCKING_CONTEXT(pos->handle) != DRM_KERNEL_CONTEXT) {
 			if (dev->driver->context_dtor)
 				dev->driver->context_dtor(dev, pos->handle);
 
@@ -342,7 +342,7 @@ int drm_legacy_addctx(struct drm_device *dev, void *data,
 	struct drm_ctx *ctx = data;
 
 	ctx->handle = drm_legacy_ctxbitmap_next(dev);
-	if (ctx->handle == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(ctx->handle) == DRM_KERNEL_CONTEXT) {
 		/* Skip kernel's context and get a new one. */
 		ctx->handle = drm_legacy_ctxbitmap_next(dev);
 	}

drivers/gpu/drm/drm_context.c int drm_legacy_rmctx(struct drm_device *dev, void *data, · line 455

Quoted 8 lines
drivers/gpu/drm/drm_context.c +1 -1
@@ -449,7 +449,7 @@ int drm_legacy_rmctx(struct drm_device *dev, void *data,
 	struct drm_ctx *ctx = data;
 
 	DRM_DEBUG("%d\n", ctx->handle);
-	if (ctx->handle != DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(ctx->handle) != DRM_KERNEL_CONTEXT) {
 		if (dev->driver->context_dtor)
 			dev->driver->context_dtor(dev, ctx->handle);
 		drm_legacy_ctxbitmap_free(dev, ctx->handle);

How about just fixing the end parameter passed to idr_alloc()? AFAICS that would take care of the context code.

Well, there are a few more issues with the code:
- not properly checking for negative return value from idr_alloc()
- leaking the ctx id on kmalloc() error
- pointless check for idr_alloc() returning 0 even though the min is 1

drivers/gpu/drm/drm_lock.c int drm_legacy_lock(struct drm_device *dev, void *data, · line 64

Quoted 7 lines
drivers/gpu/drm/drm_lock.c +0 -0
index 070dd5d..94500930 100644
@@ -63,7 +63,7 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 
 	++file_priv->lock_count;

While you're poking around this dungeopn, maybe you can kill lock_count? We never seem to decrement it, and it's only checked in drm_legacy_i_have_hw_lock().

Quoted 7 lines
drivers/gpu/drm/drm_lock.c +1 -1
-	if (lock->context == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
 		return -EINVAL;

drivers/gpu/drm/drm_lock.c int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_ · line 159

Quoted 8 lines
drivers/gpu/drm/drm_lock.c +1 -1
@@ -153,7 +153,7 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
-	if (lock->context == DRM_KERNEL_CONTEXT) {
+	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
 		return -EINVAL;

These two changes look OK to me.

Quoted 3 lines
Signature · 1 line
1.9.1
Quoted 9 lines
Quoted 8 lines
Quoted 4 lines

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Signature · 2 lines
Ville Syrjälä
Intel OTC

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#18 lore raw

Re: [Intel-gfx] [PATCH 3/5] drm: Possible lock priority escalation.

Quoted 27 lines · Peter
Quoted 20 lines · Daniel Vetter
Quoted 10 lines · Ville Syrjälä
Quoted 9 lines · Peter Antoine

If an application that has a driver lock created, wants the lock the kernel context, it is not allowed to. If the call to drm_lock has a context of 0, it is rejected. If you set the context to _DRM_LOCK_CONT then call drm lock, it will pass the context == DRM_KERNEL_CONTEXT checks. But as the DRM_LOCK_CONT bits are not part of the context id this allows operations on the DRM_KERNEL_CONTEXT.

Issue: VIZ-5485

If you're touching code with drm_legacy_ prefix of in such a file you've ended up in the horrible corners of the dri1 dungeons and should head back out pronto ;-)

If we can actually run into this code on production i915 then we need to improve the locks at the door of these dungeons for kms drivers, not try to fix up the mess behind them. That's just plain impossible.

Quoted 6 lines

If you want to make really sure we get this right some simple drm igt tests to make sure these codepaths are really dead for kms driver might be good. But otherwise we really can only annotate this as wontfix in code security issue scanners.
-Daniel

Quoted 6 lines

There is a test that covers this fix. This is a simple three line fix that stops a userspace driver locking the kernel context. Yes they are other problems with this code, but why are they stopping this patch that does a simple fix from going in?

I'll happily drop this patch if it causes more problems that it fixes.

Because we don't want to fix the legacy context crap but instead outright disable it for everyone (well except nouveau) running in kms code. drm legacy code really is broken by design, there's no way to fix it.

And worst case we'll end up breaking some old machines in some and then have to deal with the regression. Which means I won't apply these patches if we can somehow just disable all that code for i915. -Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#19 lore raw

PATCH 4/5 drm: Make HW_LOCK access functions optional.

In reply to Peter Antoine

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_lock.c            |  6 ++++++
 drivers/gpu/drm/i915/i915_dma.c       |  3 +++
 drivers/gpu/drm/nouveau/nouveau_drm.c |  3 ++-
 include/drm/drmP.h                    | 23 ++++++++++++-----------
 include/uapi/drm/i915_drm.h           |  1 +
 5 files changed, 24 insertions(+), 12 deletions(-)
drivers/gpu/drm/drm_lock.c +6 -0
index 94500930..b61d4c7 100644
@@ -61,6 +61,9 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 	struct drm_master *master = file_priv->master;
 	int ret = 0;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	++file_priv->lock_count;
 
 	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
@@ -153,6 +156,9 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
drivers/gpu/drm/i915/i915_dma.c +3 -0
index e44116f..c771ef0 100644
@@ -163,6 +163,9 @@ static int i915_getparam(struct drm_device *dev, void *data,
 		if (!value)
 			return -ENODEV;
 		break;
+	case I915_PARAM_HAS_LEGACY_CONTEXT:
+		value = drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT);
+		break;
 	default:
 		DRM_DEBUG("Unknown parameter %d\n", param->param);
 		return -EINVAL;
drivers/gpu/drm/nouveau/nouveau_drm.c +2 -1
index 8763deb..936b423 100644
@@ -940,7 +940,8 @@ static struct drm_driver
 driver_stub = {
 	.driver_features =
 		DRIVER_USE_AGP |
-		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER,
+		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER |
+		DRIVER_KMS_LEGACY_CONTEXT,
 
 	.load = nouveau_drm_load,
 	.unload = nouveau_drm_unload,
include/drm/drmP.h +12 -11
index 62c40777..367e42f 100644
@@ -137,17 +137,18 @@ void drm_err(const char *format, ...);
 /*@{*/
 
 /* driver capabilities and requirements mask */
-#define DRIVER_USE_AGP     0x1
-#define DRIVER_PCI_DMA     0x8
-#define DRIVER_SG          0x10
-#define DRIVER_HAVE_DMA    0x20
-#define DRIVER_HAVE_IRQ    0x40
-#define DRIVER_IRQ_SHARED  0x80
-#define DRIVER_GEM         0x1000
-#define DRIVER_MODESET     0x2000
-#define DRIVER_PRIME       0x4000
-#define DRIVER_RENDER      0x8000
-#define DRIVER_ATOMIC      0x10000
+#define DRIVER_USE_AGP			0x1
+#define DRIVER_PCI_DMA			0x8
+#define DRIVER_SG			0x10
+#define DRIVER_HAVE_DMA			0x20
+#define DRIVER_HAVE_IRQ			0x40
+#define DRIVER_IRQ_SHARED		0x80
+#define DRIVER_GEM			0x1000
+#define DRIVER_MODESET			0x2000
+#define DRIVER_PRIME			0x4000
+#define DRIVER_RENDER			0x8000
+#define DRIVER_ATOMIC			0x10000
+#define DRIVER_KMS_LEGACY_CONTEXT	0x20000
 
 /***********************************************************************/
 /** \name Macros to make printk easier */
include/uapi/drm/i915_drm.h +1 -0
index 4851d66..0ad611a2 100644
@@ -350,6 +350,7 @@ typedef struct drm_i915_irq_wait {
 #define I915_PARAM_REVISION              32
 #define I915_PARAM_SUBSLICE_TOTAL	 33
 #define I915_PARAM_EU_TOTAL		 34
+#define I915_PARAM_HAS_LEGACY_CONTEXT	 35
 
 typedef struct drm_i915_getparam {
 	int param;
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#20 lore raw

Re: [PATCH 4/5] drm: Make HW_LOCK access functions optional.

Quoted 42 lines · Peter Antoine

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_lock.c            |  6 ++++++
 drivers/gpu/drm/i915/i915_dma.c       |  3 +++
 drivers/gpu/drm/nouveau/nouveau_drm.c |  3 ++-
 include/drm/drmP.h                    | 23 ++++++++++++-----------
 include/uapi/drm/i915_drm.h           |  1 +
 5 files changed, 24 insertions(+), 12 deletions(-)
drivers/gpu/drm/drm_lock.c +6 -0
index 94500930..b61d4c7 100644
@@ -61,6 +61,9 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 	struct drm_master *master = file_priv->master;
 	int ret = 0;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	++file_priv->lock_count;
 
 	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
@@ -153,6 +156,9 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
drivers/gpu/drm/i915/i915_dma.c +0 -0
index e44116f..c771ef0 100644

drivers/gpu/drm/i915/i915_dma.c static int i915_getparam(struct drm_device *dev, void *data, · line 168

Quoted 6 lines
drivers/gpu/drm/i915/i915_dma.c +3 -0
@@ -163,6 +163,9 @@ static int i915_getparam(struct drm_device *dev, void *data,
 		if (!value)
 			return -ENODEV;
 		break;
+	case I915_PARAM_HAS_LEGACY_CONTEXT:
+		value = drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT);
+		break;

Seems pointless to add a parameter that'll always be false.

Quoted 8 lines
drivers/gpu/drm/i915/i915_dma.c +0 -0
 	default:
 		DRM_DEBUG("Unknown parameter %d\n", param->param);
 		return -EINVAL;
drivers/gpu/drm/nouveau/nouveau_drm.c +0 -0
index 8763deb..936b423 100644

drivers/gpu/drm/nouveau/nouveau_drm.c static struct drm_driver · line 944

Quoted 6 lines
drivers/gpu/drm/nouveau/nouveau_drm.c +2 -1
@@ -940,7 +940,8 @@ static struct drm_driver
 driver_stub = {
 	.driver_features =
 		DRIVER_USE_AGP |
-		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER,
+		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER |
+		DRIVER_KMS_LEGACY_CONTEXT,

Why is this here? AFAICS only the via driver cares about legacy contexts, and only dri1 drivers care about the hw lock.

Quoted 24 lines
drivers/gpu/drm/nouveau/nouveau_drm.c +0 -0
 	.load = nouveau_drm_load,
 	.unload = nouveau_drm_unload,
include/drm/drmP.h +2 -11
index 62c40777..367e42f 100644
@@ -137,17 +137,18 @@ void drm_err(const char *format, ...);
 /*@{*/
 
 /* driver capabilities and requirements mask */
-#define DRIVER_USE_AGP     0x1
-#define DRIVER_PCI_DMA     0x8
-#define DRIVER_SG          0x10
-#define DRIVER_HAVE_DMA    0x20
-#define DRIVER_HAVE_IRQ    0x40
-#define DRIVER_IRQ_SHARED  0x80
-#define DRIVER_GEM         0x1000
-#define DRIVER_MODESET     0x2000
-#define DRIVER_PRIME       0x4000
-#define DRIVER_RENDER      0x8000
-#define DRIVER_ATOMIC      0x10000
+#define DRIVER_USE_AGP			0x1
+#define DRIVER_PCI_DMA			0x8

include/drm/drmP.h void drm_err(const char *format, ...); · line 151

Quoted 10 lines
include/drm/drmP.h +10 -0
+#define DRIVER_SG			0x10
+#define DRIVER_HAVE_DMA			0x20
+#define DRIVER_HAVE_IRQ			0x40
+#define DRIVER_IRQ_SHARED		0x80
+#define DRIVER_GEM			0x1000
+#define DRIVER_MODESET			0x2000
+#define DRIVER_PRIME			0x4000
+#define DRIVER_RENDER			0x8000
+#define DRIVER_ATOMIC			0x10000
+#define DRIVER_KMS_LEGACY_CONTEXT	0x20000

Why is there KMS in the name?

I was thinking just checking for GEM, but I think there was some gem+dri1 userland for i915 at some point in time. ums and dri1 are now dead as far as i915 is concerned, so in theory it should be fine. But I'm not sure if some other driver might have the same baggage.

I suppose one option would be to check for MODESET instead. kms+dri1 doesn't sound like an entirely sane combination to me.

Quoted 22 lines
include/drm/drmP.h +0 -0
 /***********************************************************************/
 /** \name Macros to make printk easier */
include/uapi/drm/i915_drm.h +1 -0
index 4851d66..0ad611a2 100644
@@ -350,6 +350,7 @@ typedef struct drm_i915_irq_wait {
 #define I915_PARAM_REVISION              32
 #define I915_PARAM_SUBSLICE_TOTAL	 33
 #define I915_PARAM_EU_TOTAL		 34
+#define I915_PARAM_HAS_LEGACY_CONTEXT	 35
 
 typedef struct drm_i915_getparam {
 	int param;
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 6 lines
Ville Syrjälä
Intel OTC
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#21 lore raw

Re: [PATCH 4/5] drm: Make HW_LOCK access functions optional.

Hi,

(replies inline)

-----Original Message-----
From: Ville Syrjälä [mailto:[email protected]]
Sent: Monday, April 27, 2015 6:04 PM
To: Antoine, Peter
Cc: [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [Intel-gfx] [PATCH 4/5] drm: Make HW_LOCK access functions optional.

Quoted 42 lines · Peter Antoine

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_lock.c            |  6 ++++++
 drivers/gpu/drm/i915/i915_dma.c       |  3 +++
 drivers/gpu/drm/nouveau/nouveau_drm.c |  3 ++-
 include/drm/drmP.h                    | 23 ++++++++++++-----------
 include/uapi/drm/i915_drm.h           |  1 +
 5 files changed, 24 insertions(+), 12 deletions(-)
drivers/gpu/drm/drm_lock.c +6 -1
index 94500930..b61d4c7 100644
@@ -61,6 +61,9 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 	struct drm_master *master = file_priv->master;
 	int ret = 0;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	++file_priv->lock_count;
 
 	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) { @@ 
-153,6 +156,9 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context); diff --git 

a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c index e44116f..c771ef0 100644

drivers/gpu/drm/i915/i915_dma.c static int i915_getparam(struct drm_device *dev, void *data, · line 168

Quoted 6 lines
drivers/gpu/drm/i915/i915_dma.c +3 -0
@@ -163,6 +163,9 @@ static int i915_getparam(struct drm_device *dev, void *data,
 		if (!value)
 			return -ENODEV;
 		break;
+	case I915_PARAM_HAS_LEGACY_CONTEXT:
+		value = drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT);
+		break;

Seems pointless to add a parameter that'll always be false.

There is some history to these changes, the HW_LOCK functions were removed previously but causes an issue with the Nouveau drivers. So that the functions where put back in. So the parameter has been added to allow for that driver to turn the legacy context on as it is needed.

Quoted 9 lines
drivers/gpu/drm/i915/i915_dma.c +0 -0
 	default:
 		DRM_DEBUG("Unknown parameter %d\n", param->param);
 		return -EINVAL;
drivers/gpu/drm/nouveau/nouveau_drm.c +0 -0
index 8763deb..936b423 100644

drivers/gpu/drm/nouveau/nouveau_drm.c static struct drm_driver driver_stub = { · line 943

Quoted 5 lines
drivers/gpu/drm/nouveau/nouveau_drm.c +2 -1
@@ -940,7 +940,8 @@ static struct drm_driver  driver_stub = {
 	.driver_features =
 		DRIVER_USE_AGP |
-		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER,
+		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER |
+		DRIVER_KMS_LEGACY_CONTEXT,

Why is this here? AFAICS only the via driver cares about legacy contexts, and only dri1 drivers care about the hw lock.

See above.

Quoted 23 lines
drivers/gpu/drm/nouveau/nouveau_drm.c +0 -0
 	.load = nouveau_drm_load,
 	.unload = nouveau_drm_unload,
include/drm/drmP.h index renamed from include/drm/drmP.h +0 -0

62c40777..367e42f 100644

include/drm/drmP.h +2 -11
@@ -137,17 +137,18 @@ void drm_err(const char *format, ...);  /*@{*/
 
 /* driver capabilities and requirements mask */
-#define DRIVER_USE_AGP     0x1
-#define DRIVER_PCI_DMA     0x8
-#define DRIVER_SG          0x10
-#define DRIVER_HAVE_DMA    0x20
-#define DRIVER_HAVE_IRQ    0x40
-#define DRIVER_IRQ_SHARED  0x80
-#define DRIVER_GEM         0x1000
-#define DRIVER_MODESET     0x2000
-#define DRIVER_PRIME       0x4000
-#define DRIVER_RENDER      0x8000
-#define DRIVER_ATOMIC      0x10000
+#define DRIVER_USE_AGP			0x1
+#define DRIVER_PCI_DMA			0x8

include/drm/drmP.h void drm_err(const char *format, ...); /*@{*/ · line 150

Quoted 10 lines
include/drm/drmP.h +10 -0
+#define DRIVER_SG			0x10
+#define DRIVER_HAVE_DMA			0x20
+#define DRIVER_HAVE_IRQ			0x40
+#define DRIVER_IRQ_SHARED		0x80
+#define DRIVER_GEM			0x1000
+#define DRIVER_MODESET			0x2000
+#define DRIVER_PRIME			0x4000
+#define DRIVER_RENDER			0x8000
+#define DRIVER_ATOMIC			0x10000
+#define DRIVER_KMS_LEGACY_CONTEXT	0x20000

Why is there KMS in the name?

By suggestion of Daniel.

I was thinking just checking for GEM, but I think there was some gem+dri1 userland for i915 at some point in time. ums and dri1 are now dead as far as i915 is concerned, so in theory it should be fine. But I'm not sure if some other driver might have the same baggage.

Other drivers have the same baggage.

I suppose one option would be to check for MODESET instead. kms+dri1 doesn't sound like an entirely sane combination to me.

Can't use the MODESET as this was how it was turned off in the previous incarnation and was reverted by Dave Airle.

Peter.

Quoted 20 lines
include/drm/drmP.h +0 -0
 /*********************************************************************
 **/
 /** \name Macros to make printk easier */ diff --git 

a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index 4851d66..0ad611a2 100644

include/uapi/drm/i915_drm.h +1 -1
@@ -350,6 +350,7 @@ typedef struct drm_i915_irq_wait {
 #define I915_PARAM_REVISION              32
 #define I915_PARAM_SUBSLICE_TOTAL	 33
 #define I915_PARAM_EU_TOTAL		 34
+#define I915_PARAM_HAS_LEGACY_CONTEXT	 35
 
 typedef struct drm_i915_getparam {
 	int param;
--

1.9.1

Quoted 4 lines

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

--
Ville Syrjälä
Intel OTC
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#22 lore raw

Re: [PATCH 4/5] drm: Make HW_LOCK access functions optional.

Quoted 126 lines · Peter

Hi,

(replies inline)

-----Original Message-----
From: Ville Syrjälä [mailto:[email protected]]
Sent: Monday, April 27, 2015 6:04 PM
To: Antoine, Peter
Cc: [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [Intel-gfx] [PATCH 4/5] drm: Make HW_LOCK access functions optional.

Quoted 42 lines · Peter Antoine

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_lock.c            |  6 ++++++
 drivers/gpu/drm/i915/i915_dma.c       |  3 +++
 drivers/gpu/drm/nouveau/nouveau_drm.c |  3 ++-
 include/drm/drmP.h                    | 23 ++++++++++++-----------
 include/uapi/drm/i915_drm.h           |  1 +
 5 files changed, 24 insertions(+), 12 deletions(-)
drivers/gpu/drm/drm_lock.c +6 -1
index 94500930..b61d4c7 100644
@@ -61,6 +61,9 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 	struct drm_master *master = file_priv->master;
 	int ret = 0;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	++file_priv->lock_count;
 
 	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) { @@ 
-153,6 +156,9 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context); diff --git 

a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c index e44116f..c771ef0 100644

drivers/gpu/drm/i915/i915_dma.c static int i915_getparam(struct drm_device *dev, void *data, · line 168

Quoted 6 lines
drivers/gpu/drm/i915/i915_dma.c +3 -0
@@ -163,6 +163,9 @@ static int i915_getparam(struct drm_device *dev, void *data,
 		if (!value)
 			return -ENODEV;
 		break;
+	case I915_PARAM_HAS_LEGACY_CONTEXT:
+		value = drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT);
+		break;

Seems pointless to add a parameter that'll always be false.

There is some history to these changes, the HW_LOCK functions were removed previously but causes an issue with the Nouveau drivers. So that the functions where put back in. So the parameter has been added to allow for that driver to turn the legacy context on as it is needed.

Quoted 9 lines
drivers/gpu/drm/i915/i915_dma.c +0 -0
 	default:
 		DRM_DEBUG("Unknown parameter %d\n", param->param);
 		return -EINVAL;
drivers/gpu/drm/nouveau/nouveau_drm.c +0 -0
index 8763deb..936b423 100644

drivers/gpu/drm/nouveau/nouveau_drm.c static struct drm_driver driver_stub = { · line 943

Quoted 5 lines
drivers/gpu/drm/nouveau/nouveau_drm.c +2 -1
@@ -940,7 +940,8 @@ static struct drm_driver  driver_stub = {
 	.driver_features =
 		DRIVER_USE_AGP |
-		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER,
+		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER |
+		DRIVER_KMS_LEGACY_CONTEXT,

Why is this here? AFAICS only the via driver cares about legacy contexts, and only dri1 drivers care about the hw lock.

See above.

Quoted 23 lines
drivers/gpu/drm/nouveau/nouveau_drm.c +0 -0
 	.load = nouveau_drm_load,
 	.unload = nouveau_drm_unload,
include/drm/drmP.h index renamed from include/drm/drmP.h +0 -0

62c40777..367e42f 100644

include/drm/drmP.h +2 -11
@@ -137,17 +137,18 @@ void drm_err(const char *format, ...);  /*@{*/
 
 /* driver capabilities and requirements mask */
-#define DRIVER_USE_AGP     0x1
-#define DRIVER_PCI_DMA     0x8
-#define DRIVER_SG          0x10
-#define DRIVER_HAVE_DMA    0x20
-#define DRIVER_HAVE_IRQ    0x40
-#define DRIVER_IRQ_SHARED  0x80
-#define DRIVER_GEM         0x1000
-#define DRIVER_MODESET     0x2000
-#define DRIVER_PRIME       0x4000
-#define DRIVER_RENDER      0x8000
-#define DRIVER_ATOMIC      0x10000
+#define DRIVER_USE_AGP			0x1
+#define DRIVER_PCI_DMA			0x8

include/drm/drmP.h void drm_err(const char *format, ...); /*@{*/ · line 150

Quoted 10 lines
include/drm/drmP.h +10 -0
+#define DRIVER_SG			0x10
+#define DRIVER_HAVE_DMA			0x20
+#define DRIVER_HAVE_IRQ			0x40
+#define DRIVER_IRQ_SHARED		0x80
+#define DRIVER_GEM			0x1000
+#define DRIVER_MODESET			0x2000
+#define DRIVER_PRIME			0x4000
+#define DRIVER_RENDER			0x8000
+#define DRIVER_ATOMIC			0x10000
+#define DRIVER_KMS_LEGACY_CONTEXT	0x20000

Why is there KMS in the name?

By suggestion of Daniel.

I was thinking just checking for GEM, but I think there was some gem+dri1 userland for i915 at some point in time. ums and dri1 are now dead as far as i915 is concerned, so in theory it should be fine. But I'm not sure if some other driver might have the same baggage.

Quoted 5 lines

Other drivers have the same baggage.

I suppose one option would be to check for MODESET instead. kms+dri1 doesn't sound like an entirely sane combination to me.

Can't use the MODESET as this was how it was turned off in the previous incarnation and was reverted by Dave Airle.

Reference?

Quoted 31 lines

Peter.

Quoted 20 lines
include/drm/drmP.h +0 -0
 /*********************************************************************
 **/
 /** \name Macros to make printk easier */ diff --git 

a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index 4851d66..0ad611a2 100644

include/uapi/drm/i915_drm.h +1 -1
@@ -350,6 +350,7 @@ typedef struct drm_i915_irq_wait {
 #define I915_PARAM_REVISION              32
 #define I915_PARAM_SUBSLICE_TOTAL	 33
 #define I915_PARAM_EU_TOTAL		 34
+#define I915_PARAM_HAS_LEGACY_CONTEXT	 35
 
 typedef struct drm_i915_getparam {
 	int param;
--

1.9.1

Quoted 4 lines

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

--
Ville Syrjälä
Intel OTC

Signature · 6 lines
Ville Syrjälä
Intel OTC
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#23 lore raw

Re: [PATCH 4/5] drm: Make HW_LOCK access functions optional.

reply at end.

Quoted 127 lines · Ville Syrjälä
Quoted 126 lines · Peter

Hi,

(replies inline)

-----Original Message-----
From: Ville Syrjälä [mailto:[email protected]]
Sent: Monday, April 27, 2015 6:04 PM
To: Antoine, Peter
Cc: [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [Intel-gfx] [PATCH 4/5] drm: Make HW_LOCK access functions optional.

Quoted 42 lines · Peter Antoine

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_lock.c            |  6 ++++++
 drivers/gpu/drm/i915/i915_dma.c       |  3 +++
 drivers/gpu/drm/nouveau/nouveau_drm.c |  3 ++-
 include/drm/drmP.h                    | 23 ++++++++++++-----------
 include/uapi/drm/i915_drm.h           |  1 +
 5 files changed, 24 insertions(+), 12 deletions(-)
drivers/gpu/drm/drm_lock.c +6 -1
index 94500930..b61d4c7 100644
@@ -61,6 +61,9 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 	struct drm_master *master = file_priv->master;
 	int ret = 0;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	++file_priv->lock_count;
 
 	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) { @@ 
-153,6 +156,9 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	if (_DRM_LOCKING_CONTEXT(lock->context) == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context); diff --git 

a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c index e44116f..c771ef0 100644

drivers/gpu/drm/i915/i915_dma.c static int i915_getparam(struct drm_device *dev, void *data, · line 168

Quoted 6 lines
drivers/gpu/drm/i915/i915_dma.c +3 -0
@@ -163,6 +163,9 @@ static int i915_getparam(struct drm_device *dev, void *data,
 		if (!value)
 			return -ENODEV;
 		break;
+	case I915_PARAM_HAS_LEGACY_CONTEXT:
+		value = drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT);
+		break;

Seems pointless to add a parameter that'll always be false.

There is some history to these changes, the HW_LOCK functions were removed previously but causes an issue with the Nouveau drivers. So that the functions where put back in. So the parameter has been added to allow for that driver to turn the legacy context on as it is needed.

Quoted 9 lines
drivers/gpu/drm/i915/i915_dma.c +0 -0
 	default:
 		DRM_DEBUG("Unknown parameter %d\n", param->param);
 		return -EINVAL;
drivers/gpu/drm/nouveau/nouveau_drm.c +0 -0
index 8763deb..936b423 100644

drivers/gpu/drm/nouveau/nouveau_drm.c static struct drm_driver driver_stub = { · line 943

Quoted 5 lines
drivers/gpu/drm/nouveau/nouveau_drm.c +2 -1
@@ -940,7 +940,8 @@ static struct drm_driver  driver_stub = {
 	.driver_features =
 		DRIVER_USE_AGP |
-		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER,
+		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER |
+		DRIVER_KMS_LEGACY_CONTEXT,

Why is this here? AFAICS only the via driver cares about legacy contexts, and only dri1 drivers care about the hw lock.

See above.

Quoted 23 lines
drivers/gpu/drm/nouveau/nouveau_drm.c +0 -0
 	.load = nouveau_drm_load,
 	.unload = nouveau_drm_unload,
include/drm/drmP.h index renamed from include/drm/drmP.h +0 -0

62c40777..367e42f 100644

include/drm/drmP.h +2 -11
@@ -137,17 +137,18 @@ void drm_err(const char *format, ...);  /*@{*/
 
 /* driver capabilities and requirements mask */
-#define DRIVER_USE_AGP     0x1
-#define DRIVER_PCI_DMA     0x8
-#define DRIVER_SG          0x10
-#define DRIVER_HAVE_DMA    0x20
-#define DRIVER_HAVE_IRQ    0x40
-#define DRIVER_IRQ_SHARED  0x80
-#define DRIVER_GEM         0x1000
-#define DRIVER_MODESET     0x2000
-#define DRIVER_PRIME       0x4000
-#define DRIVER_RENDER      0x8000
-#define DRIVER_ATOMIC      0x10000
+#define DRIVER_USE_AGP			0x1
+#define DRIVER_PCI_DMA			0x8

include/drm/drmP.h void drm_err(const char *format, ...); /*@{*/ · line 150

Quoted 10 lines
include/drm/drmP.h +10 -0
+#define DRIVER_SG			0x10
+#define DRIVER_HAVE_DMA			0x20
+#define DRIVER_HAVE_IRQ			0x40
+#define DRIVER_IRQ_SHARED		0x80
+#define DRIVER_GEM			0x1000
+#define DRIVER_MODESET			0x2000
+#define DRIVER_PRIME			0x4000
+#define DRIVER_RENDER			0x8000
+#define DRIVER_ATOMIC			0x10000
+#define DRIVER_KMS_LEGACY_CONTEXT	0x20000

Why is there KMS in the name?

By suggestion of Daniel.

I was thinking just checking for GEM, but I think there was some gem+dri1 userland for i915 at some point in time. ums and dri1 are now dead as far as i915 is concerned, so in theory it should be fine. But I'm not sure if some other driver might have the same baggage.

Quoted 7 lines
Quoted 5 lines

Other drivers have the same baggage.

I suppose one option would be to check for MODESET instead. kms+dri1 doesn't sound like an entirely sane combination to me.

Can't use the MODESET as this was how it was turned off in the previous incarnation and was reverted by Dave Airle.

Reference?

From the next commit [5/5] as it is the one that actually turns off the functions that were turned off before.

These changes are based on the two patches:
commit c21eb21cb50d58e7cbdcb8b9e7ff68b85cfa5095
Author: Dave Airlie <[email protected]>

And the commit that the above patch reverts:
commit 7c510133d93dd6f15ca040733ba7b2891ed61fd1
Author: Daniel Vetter <[email protected]>

Quoted 24 lines
Quoted 24 lines

Peter.

Quoted 20 lines
include/drm/drmP.h +0 -0
 /*********************************************************************
 **/
 /** \name Macros to make printk easier */ diff --git 

a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index 4851d66..0ad611a2 100644

include/uapi/drm/i915_drm.h +1 -1
@@ -350,6 +350,7 @@ typedef struct drm_i915_irq_wait {
 #define I915_PARAM_REVISION              32
 #define I915_PARAM_SUBSLICE_TOTAL	 33
 #define I915_PARAM_EU_TOTAL		 34
+#define I915_PARAM_HAS_LEGACY_CONTEXT	 35
 
 typedef struct drm_i915_getparam {
 	int param;
--

1.9.1

Quoted 9 lines
Quoted 8 lines
Quoted 4 lines

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

--
Ville Syrjälä
Intel OTC

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#24 lore raw

Re: [PATCH 4/5] drm: Make HW_LOCK access functions optional.

Quoted 51 lines · Peter
Quoted 40 lines
Quoted 40 lines
Quoted 20 lines
include/drm/drmP.h index renamed from include/drm/drmP.h +0 -0

62c40777..367e42f 100644

include/drm/drmP.h +2 -11
@@ -137,17 +137,18 @@ void drm_err(const char *format, ...);  /*@{*/
 
 /* driver capabilities and requirements mask */
-#define DRIVER_USE_AGP     0x1
-#define DRIVER_PCI_DMA     0x8
-#define DRIVER_SG          0x10
-#define DRIVER_HAVE_DMA    0x20
-#define DRIVER_HAVE_IRQ    0x40
-#define DRIVER_IRQ_SHARED  0x80
-#define DRIVER_GEM         0x1000
-#define DRIVER_MODESET     0x2000
-#define DRIVER_PRIME       0x4000
-#define DRIVER_RENDER      0x8000
-#define DRIVER_ATOMIC      0x10000
+#define DRIVER_USE_AGP			0x1
+#define DRIVER_PCI_DMA			0x8

include/drm/drmP.h void drm_err(const char *format, ...); /*@{*/ · line 150

Quoted 10 lines
include/drm/drmP.h +10 -0
+#define DRIVER_SG			0x10
+#define DRIVER_HAVE_DMA			0x20
+#define DRIVER_HAVE_IRQ			0x40
+#define DRIVER_IRQ_SHARED		0x80
+#define DRIVER_GEM			0x1000
+#define DRIVER_MODESET			0x2000
+#define DRIVER_PRIME			0x4000
+#define DRIVER_RENDER			0x8000
+#define DRIVER_ATOMIC			0x10000
+#define DRIVER_KMS_LEGACY_CONTEXT	0x20000

Why is there KMS in the name?

By suggestion of Daniel.

I was thinking just checking for GEM, but I think there was some gem+dri1 userland for i915 at some point in time. ums and dri1 are now dead as far as i915 is concerned, so in theory it should be fine. But I'm not sure if some other driver might have the same baggage.

Quoted 7 lines
Quoted 5 lines

Other drivers have the same baggage.

I suppose one option would be to check for MODESET instead. kms+dri1 doesn't sound like an entirely sane combination to me.

Can't use the MODESET as this was how it was turned off in the previous incarnation and was reverted by Dave Airle.

Reference?

From the next commit [5/5] as it is the one that actually turns off the functions that were turned off before.

Quoted 7 lines

These changes are based on the two patches:
commit c21eb21cb50d58e7cbdcb8b9e7ff68b85cfa5095
Author: Dave Airlie <[email protected]>

And the commit that the above patch reverts:
commit 7c510133d93dd6f15ca040733ba7b2891ed61fd1
Author: Daniel Vetter <[email protected]>

Looking at ancient libdrm sources makes me think nouveau just used to create and destroy the context, but not actually use it for anything. So nopping out the ioctls should be good enough AFAICS. Or am I missing something?

Signature · 6 lines
Ville Syrjälä
Intel OTC
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#25 lore raw

Re: [PATCH 4/5] drm: Make HW_LOCK access functions optional.

Quoted 56 lines · Ville Syrjälä
Quoted 55 lines · Peter
Quoted 40 lines
Quoted 40 lines
Quoted 20 lines
include/drm/drmP.h index renamed from include/drm/drmP.h +0 -0

62c40777..367e42f 100644

include/drm/drmP.h +2 -11
@@ -137,17 +137,18 @@ void drm_err(const char *format, ...);  /*@{*/
 
 /* driver capabilities and requirements mask */
-#define DRIVER_USE_AGP     0x1
-#define DRIVER_PCI_DMA     0x8
-#define DRIVER_SG          0x10
-#define DRIVER_HAVE_DMA    0x20
-#define DRIVER_HAVE_IRQ    0x40
-#define DRIVER_IRQ_SHARED  0x80
-#define DRIVER_GEM         0x1000
-#define DRIVER_MODESET     0x2000
-#define DRIVER_PRIME       0x4000
-#define DRIVER_RENDER      0x8000
-#define DRIVER_ATOMIC      0x10000
+#define DRIVER_USE_AGP			0x1
+#define DRIVER_PCI_DMA			0x8

include/drm/drmP.h void drm_err(const char *format, ...); /*@{*/ · line 150

Quoted 10 lines
include/drm/drmP.h +10 -0
+#define DRIVER_SG			0x10
+#define DRIVER_HAVE_DMA			0x20
+#define DRIVER_HAVE_IRQ			0x40
+#define DRIVER_IRQ_SHARED		0x80
+#define DRIVER_GEM			0x1000
+#define DRIVER_MODESET			0x2000
+#define DRIVER_PRIME			0x4000
+#define DRIVER_RENDER			0x8000
+#define DRIVER_ATOMIC			0x10000
+#define DRIVER_KMS_LEGACY_CONTEXT	0x20000

Why is there KMS in the name?

By suggestion of Daniel.

I was thinking just checking for GEM, but I think there was some gem+dri1 userland for i915 at some point in time. ums and dri1 are now dead as far as i915 is concerned, so in theory it should be fine. But I'm not sure if some other driver might have the same baggage.

Quoted 7 lines
Quoted 5 lines

Other drivers have the same baggage.

I suppose one option would be to check for MODESET instead. kms+dri1 doesn't sound like an entirely sane combination to me.

Can't use the MODESET as this was how it was turned off in the previous incarnation and was reverted by Dave Airle.

Reference?

From the next commit [5/5] as it is the one that actually turns off the functions that were turned off before.

These changes are based on the two patches:
commit c21eb21cb50d58e7cbdcb8b9e7ff68b85cfa5095
Author: Dave Airlie <[email protected]>

Quoted 9 lines
Quoted 3 lines

And the commit that the above patch reverts:
commit 7c510133d93dd6f15ca040733ba7b2891ed61fd1
Author: Daniel Vetter <[email protected]>

Looking at ancient libdrm sources makes me think nouveau just used to create and destroy the context, but not actually use it for anything. So nopping out the ioctls should be good enough AFAICS. Or am I missing something?

An old version of libdrm that still requires support needs them, it's the reason that David Airlie reverted the patch that Daniel did to remove the functions. Do they still need support, I don't know? David Airlie is on the cc list.

A discussion was had and this was the way that it was suggested it be done. This seems a good half-way house, the actual security holes that have been found have been fixed and the functions (that seem to have a lot more security issues in them) are turned off for the drivers that don't use them, and if a driver does require them, it will be a one line change to reintroduce them. Are we carrying code we don't need to support, probably.

Peter.
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#26 lore raw

Re: [PATCH 4/5] drm: Make HW_LOCK access functions optional.

Quoted 53 lines · Peter
Quoted 52 lines · Ville Syrjälä
Quoted 51 lines · Peter
Quoted 40 lines
Quoted 40 lines
Quoted 20 lines
include/drm/drmP.h index renamed from include/drm/drmP.h +0 -0

62c40777..367e42f 100644

include/drm/drmP.h +2 -11
@@ -137,17 +137,18 @@ void drm_err(const char *format, ...);  /*@{*/
 
 /* driver capabilities and requirements mask */
-#define DRIVER_USE_AGP     0x1
-#define DRIVER_PCI_DMA     0x8
-#define DRIVER_SG          0x10
-#define DRIVER_HAVE_DMA    0x20
-#define DRIVER_HAVE_IRQ    0x40
-#define DRIVER_IRQ_SHARED  0x80
-#define DRIVER_GEM         0x1000
-#define DRIVER_MODESET     0x2000
-#define DRIVER_PRIME       0x4000
-#define DRIVER_RENDER      0x8000
-#define DRIVER_ATOMIC      0x10000
+#define DRIVER_USE_AGP			0x1
+#define DRIVER_PCI_DMA			0x8

include/drm/drmP.h void drm_err(const char *format, ...); /*@{*/ · line 150

Quoted 10 lines
include/drm/drmP.h +10 -0
+#define DRIVER_SG			0x10
+#define DRIVER_HAVE_DMA			0x20
+#define DRIVER_HAVE_IRQ			0x40
+#define DRIVER_IRQ_SHARED		0x80
+#define DRIVER_GEM			0x1000
+#define DRIVER_MODESET			0x2000
+#define DRIVER_PRIME			0x4000
+#define DRIVER_RENDER			0x8000
+#define DRIVER_ATOMIC			0x10000
+#define DRIVER_KMS_LEGACY_CONTEXT	0x20000

Why is there KMS in the name?

By suggestion of Daniel.

I was thinking just checking for GEM, but I think there was some gem+dri1 userland for i915 at some point in time. ums and dri1 are now dead as far as i915 is concerned, so in theory it should be fine. But I'm not sure if some other driver might have the same baggage.

Quoted 7 lines
Quoted 5 lines

Other drivers have the same baggage.

I suppose one option would be to check for MODESET instead. kms+dri1 doesn't sound like an entirely sane combination to me.

Can't use the MODESET as this was how it was turned off in the previous incarnation and was reverted by Dave Airle.

Reference?

From the next commit [5/5] as it is the one that actually turns off the functions that were turned off before.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

These changes are based on the two patches:
commit c21eb21cb50d58e7cbdcb8b9e7ff68b85cfa5095
Author: Dave Airlie <[email protected]>

And the commit that the above patch reverts:
commit 7c510133d93dd6f15ca040733ba7b2891ed61fd1
Author: Daniel Vetter <[email protected]>

These two commits definitely should be referenced from the commit message, othwerise no one will understand the history.

Quoted 11 lines
Quoted 5 lines

Looking at ancient libdrm sources makes me think nouveau just used to create and destroy the context, but not actually use it for anything. So nopping out the ioctls should be good enough AFAICS. Or am I missing something?

An old version of libdrm that still requires support needs them, it's the reason that David Airlie reverted the patch that Daniel did to remove the functions. Do they still need support, I don't know? David Airlie is on the cc list.

Quoted 7 lines

A discussion was had and this was the way that it was suggested it be done. This seems a good half-way house, the actual security holes that have been found have been fixed and the functions (that seem to have a lot more security issues in them) are turned off for the drivers that don't use them, and if a driver does require them, it will be a one line change to reintroduce them. Are we carrying code we don't need to support, probably.

Iirc it was in the ddx, and it was actually using the mmap code. Leftovers from ums, but unfortunately X crashes if we take them away. If I recall correctly nouveau was in staging still, but per Linus staging or not doesn't matter when distros are shipping with the code already. I did dig out the details way back out of curiosity, but lost them since then again. -Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
maintainer
#27 lore raw

Re: [PATCH 4/5] drm: Make HW_LOCK access functions optional.

Quoted 7 lines

Iirc it was in the ddx, and it was actually using the mmap code. Leftovers from ums, but unfortunately X crashes if we take them away. If I recall correctly nouveau was in staging still, but per Linus staging or not doesn't matter when distros are shipping with the code already. I did dig out the details way back out of curiosity, but lost them since then again. -Daniel

There were two different nouveau problems

a) old libdrm used contexts, it called drmCreateContext and drmDestroyContext from what I could see.

b) old DDX used DRI1 functions, DRIOpenDRMMaster in the X server when it wanted to use open, that function added maps with locks and mapped them.

I get the impression this is case (b).

Dave.
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#28 lore raw

PATCH 5/5 drm: Make Legacy Context access functions optional.

In reply to Peter Antoine

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

These changes are based on the two patches:
commit c21eb21cb50d58e7cbdcb8b9e7ff68b85cfa5095
Author: Dave Airlie <[email protected]>

And the commit that the above patch reverts:
commit 7c510133d93dd6f15ca040733ba7b2891ed61fd1
Author: Daniel Vetter <[email protected]>

This should now turn off the context feature.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_context.c | 36 ++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/drm_drv.c     | 12 +++++++-----
 2 files changed, 43 insertions(+), 5 deletions(-)
drivers/gpu/drm/drm_context.c +36 -0
index 1febcd3..09af26c 100644
@@ -53,6 +53,9 @@ struct drm_ctx_list {
  */
 void drm_legacy_ctxbitmap_free(struct drm_device * dev, int ctx_handle)
 {
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->struct_mutex);
 	idr_remove(&dev->ctx_idr, ctx_handle);
 	mutex_unlock(&dev->struct_mutex);
@@ -87,6 +90,9 @@ static int drm_legacy_ctxbitmap_next(struct drm_device * dev)
  */
 int drm_legacy_ctxbitmap_init(struct drm_device * dev)
 {
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	idr_init(&dev->ctx_idr);
 	return 0;
 }
@@ -101,6 +107,9 @@ int drm_legacy_ctxbitmap_init(struct drm_device * dev)
  */
 void drm_legacy_ctxbitmap_cleanup(struct drm_device * dev)
 {
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->struct_mutex);
 	idr_destroy(&dev->ctx_idr);
 	mutex_unlock(&dev->struct_mutex);
@@ -119,6 +128,9 @@ void drm_legacy_ctxbitmap_flush(struct drm_device *dev, struct drm_file *file)
 {
 	struct drm_ctx_list *pos, *tmp;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->ctxlist_mutex);
 
 	list_for_each_entry_safe(pos, tmp, &dev->ctxlist, head) {
@@ -161,6 +173,9 @@ int drm_legacy_getsareactx(struct drm_device *dev, void *data,
 	struct drm_local_map *map;
 	struct drm_map_list *_entry;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->struct_mutex);
 
 	map = idr_find(&dev->ctx_idr, request->ctx_id);
@@ -205,6 +220,9 @@ int drm_legacy_setsareactx(struct drm_device *dev, void *data,
 	struct drm_local_map *map = NULL;
 	struct drm_map_list *r_list = NULL;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->struct_mutex);
 	list_for_each_entry(r_list, &dev->maplist, head) {
 		if (r_list->map
@@ -311,6 +329,9 @@ int drm_legacy_resctx(struct drm_device *dev, void *data,
 	struct drm_ctx ctx;
 	int i;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	if (res->count >= DRM_RESERVED_CONTEXTS) {
 		memset(&ctx, 0, sizeof(ctx));
 		for (i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
@@ -341,6 +362,9 @@ int drm_legacy_addctx(struct drm_device *dev, void *data,
 	struct drm_ctx_list *ctx_entry;
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	ctx->handle = drm_legacy_ctxbitmap_next(dev);
 	if (_DRM_LOCKING_CONTEXT(ctx->handle) == DRM_KERNEL_CONTEXT) {
 		/* Skip kernel's context and get a new one. */
@@ -384,6 +408,9 @@ int drm_legacy_getctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	/* This is 0, because we don't handle any context flags */
 	ctx->flags = 0;
 
@@ -406,6 +433,9 @@ int drm_legacy_switchctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	DRM_DEBUG("%d\n", ctx->handle);
 	return drm_context_switch(dev, dev->last_context, ctx->handle);
 }
@@ -426,6 +456,9 @@ int drm_legacy_newctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	DRM_DEBUG("%d\n", ctx->handle);
 	drm_context_switch_complete(dev, file_priv, ctx->handle);
 
@@ -448,6 +481,9 @@ int drm_legacy_rmctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	DRM_DEBUG("%d\n", ctx->handle);
 	if (_DRM_LOCKING_CONTEXT(ctx->handle) != DRM_KERNEL_CONTEXT) {
 		if (dev->driver->context_dtor)
drivers/gpu/drm/drm_drv.c +7 -5
index 48f7359..dc662e8 100644
@@ -596,11 +596,13 @@ struct drm_device *drm_dev_alloc(struct drm_driver *driver,
 	if (drm_ht_create(&dev->map_hash, 12))
 		goto err_minors;
 
-	ret = drm_legacy_ctxbitmap_init(dev);
-	if (ret) {
-		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
-		goto err_ht;
-	}
+	if (drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		ret = drm_legacy_ctxbitmap_init(dev);
+		if (ret) {
+			DRM_ERROR(
+				"Cannot allocate memory for context bitmap.\n");
+			goto err_ht;
+		}
 
 	if (drm_core_check_feature(dev, DRIVER_GEM)) {
 		ret = drm_gem_init(dev);
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Tested-by on v1 5/5
#29 lore raw

Re: [PATCH 5/5] drm: Make Legacy Context access functions optional.

Task id: 6256 -------------------------------------Summary-------------------------------------
Platform Delta drm-intel-nightly Series Applied

PNV                                  276/276              276/276
ILK                                  302/302              302/302
SNB                                  318/318              318/318
IVB                                  341/341              341/341
BYT                                  287/287              287/287
BDW                                  318/318              318/318

-------------------------------------Detailed-------------------------------------
Platform Test drm-intel-nightly Series Applied
Note: You need to pay more attention to line start with '*'
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#30 lore raw

v2 cover HW_LOCK kernel patched.

In reply to Peter Antoine

This is a resubmission of the security fixed for a Kernel Crash found in the drm part of the driver. It was caused by a attempt to Lock/Unlock a HW_LOCK that was uninitialised. As this feature is only required on some legacy drivers this patchset now only turns these features off.

There is an igt test to follow, just working out a way of not crashing the systems that have this feature-set turned on (probably need to create a context and if that succeeds then exit as the feature is on.

V2: Remove the patches that cover the Kernel Crash and the Lock Escalation.
Remove the PARAM that was used by the test to detect the presences of this feature.

Peter Antoine (2):
drm: Make HW_LOCK access functions optional.
drm: Make Legacy Context access functions optional.

 drivers/gpu/drm/drm_context.c         | 36 +++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/drm_drv.c             | 12 +++++++-----
 drivers/gpu/drm/drm_lock.c            |  6 ++++++
 drivers/gpu/drm/nouveau/nouveau_drm.c |  3 ++-
 include/drm/drmP.h                    | 23 +++++++++++-----------
 5 files changed, 63 insertions(+), 17 deletions(-)
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#31 lore raw

PATCH v2 1/2 drm: Make HW_LOCK access functions optional.

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_lock.c            |  6 ++++++
 drivers/gpu/drm/nouveau/nouveau_drm.c |  3 ++-
 include/drm/drmP.h                    | 23 ++++++++++++-----------
 3 files changed, 20 insertions(+), 12 deletions(-)
drivers/gpu/drm/drm_lock.c +6 -0
index f861361..21eb180 100644
@@ -61,6 +61,9 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 	struct drm_master *master = file_priv->master;
 	int ret = 0;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	++file_priv->lock_count;
 
 	if (lock->context == DRM_KERNEL_CONTEXT) {
@@ -153,6 +156,9 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	if (lock->context == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
drivers/gpu/drm/nouveau/nouveau_drm.c +2 -1
index 8904933..9624b38 100644
@@ -941,7 +941,8 @@ static struct drm_driver
 driver_stub = {
 	.driver_features =
 		DRIVER_USE_AGP |
-		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER,
+		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER |
+		DRIVER_KMS_LEGACY_CONTEXT,
 
 	.load = nouveau_drm_load,
 	.unload = nouveau_drm_unload,
include/drm/drmP.h +12 -11
index df6d997..3874942 100644
@@ -137,17 +137,18 @@ void drm_err(const char *format, ...);
 /*@{*/
 
 /* driver capabilities and requirements mask */
-#define DRIVER_USE_AGP     0x1
-#define DRIVER_PCI_DMA     0x8
-#define DRIVER_SG          0x10
-#define DRIVER_HAVE_DMA    0x20
-#define DRIVER_HAVE_IRQ    0x40
-#define DRIVER_IRQ_SHARED  0x80
-#define DRIVER_GEM         0x1000
-#define DRIVER_MODESET     0x2000
-#define DRIVER_PRIME       0x4000
-#define DRIVER_RENDER      0x8000
-#define DRIVER_ATOMIC      0x10000
+#define DRIVER_USE_AGP			0x1
+#define DRIVER_PCI_DMA			0x8
+#define DRIVER_SG			0x10
+#define DRIVER_HAVE_DMA			0x20
+#define DRIVER_HAVE_IRQ			0x40
+#define DRIVER_IRQ_SHARED		0x80
+#define DRIVER_GEM			0x1000
+#define DRIVER_MODESET			0x2000
+#define DRIVER_PRIME			0x4000
+#define DRIVER_RENDER			0x8000
+#define DRIVER_ATOMIC			0x10000
+#define DRIVER_KMS_LEGACY_CONTEXT	0x20000
 
 /***********************************************************************/
 /** \name Macros to make printk easier */
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#32 lore raw

Re: [PATCH v2 1/2] drm: Make HW_LOCK access functions optional.

Quoted 2 lines · Peter Antoine

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

Is there a reference for why nouveau needs hw locks too? Also have you done an audit of mesa history and X history to make sure there's no other driver accidentally using it with a modern kms driver?

Quoted 13 lines

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_lock.c            |  6 ++++++
 drivers/gpu/drm/nouveau/nouveau_drm.c |  3 ++-
 include/drm/drmP.h                    | 23 ++++++++++++-----------
 3 files changed, 20 insertions(+), 12 deletions(-)
drivers/gpu/drm/drm_lock.c +0 -0
index f861361..21eb180 100644

drivers/gpu/drm/drm_lock.c int drm_legacy_lock(struct drm_device *dev, void *data, · line 64

Quoted 4 lines
drivers/gpu/drm/drm_lock.c +1 -0
@@ -61,6 +61,9 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 	struct drm_master *master = file_priv->master;
 	int ret = 0;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))

You also need to allow these for all legacy drivers, i.e. without DRIVER_MODESET.
-Daniel

Quoted 66 lines
drivers/gpu/drm/drm_lock.c +5 -0
+		return -EINVAL;
+
 	++file_priv->lock_count;
 
 	if (lock->context == DRM_KERNEL_CONTEXT) {
@@ -153,6 +156,9 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	if (lock->context == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
drivers/gpu/drm/nouveau/nouveau_drm.c +2 -1
index 8904933..9624b38 100644
@@ -941,7 +941,8 @@ static struct drm_driver
 driver_stub = {
 	.driver_features =
 		DRIVER_USE_AGP |
-		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER,
+		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER |
+		DRIVER_KMS_LEGACY_CONTEXT,
 
 	.load = nouveau_drm_load,
 	.unload = nouveau_drm_unload,
include/drm/drmP.h +12 -11
index df6d997..3874942 100644
@@ -137,17 +137,18 @@ void drm_err(const char *format, ...);
 /*@{*/
 
 /* driver capabilities and requirements mask */
-#define DRIVER_USE_AGP     0x1
-#define DRIVER_PCI_DMA     0x8
-#define DRIVER_SG          0x10
-#define DRIVER_HAVE_DMA    0x20
-#define DRIVER_HAVE_IRQ    0x40
-#define DRIVER_IRQ_SHARED  0x80
-#define DRIVER_GEM         0x1000
-#define DRIVER_MODESET     0x2000
-#define DRIVER_PRIME       0x4000
-#define DRIVER_RENDER      0x8000
-#define DRIVER_ATOMIC      0x10000
+#define DRIVER_USE_AGP			0x1
+#define DRIVER_PCI_DMA			0x8
+#define DRIVER_SG			0x10
+#define DRIVER_HAVE_DMA			0x20
+#define DRIVER_HAVE_IRQ			0x40
+#define DRIVER_IRQ_SHARED		0x80
+#define DRIVER_GEM			0x1000
+#define DRIVER_MODESET			0x2000
+#define DRIVER_PRIME			0x4000
+#define DRIVER_RENDER			0x8000
+#define DRIVER_ATOMIC			0x10000
+#define DRIVER_KMS_LEGACY_CONTEXT	0x20000
 
 /***********************************************************************/
 /** \name Macros to make printk easier */
Signature · 1 line
1.9.1
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#33 lore raw

Re: [PATCH v2 1/2] drm: Make HW_LOCK access functions optional.

Adding dri-devel, I've forgotten to do that ...
-Daniel

Quoted 101 lines · Daniel Vetter
Quoted 2 lines · Peter Antoine

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

Is there a reference for why nouveau needs hw locks too? Also have you done an audit of mesa history and X history to make sure there's no other driver accidentally using it with a modern kms driver?

Quoted 13 lines

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_lock.c            |  6 ++++++
 drivers/gpu/drm/nouveau/nouveau_drm.c |  3 ++-
 include/drm/drmP.h                    | 23 ++++++++++++-----------
 3 files changed, 20 insertions(+), 12 deletions(-)
drivers/gpu/drm/drm_lock.c +0 -0
index f861361..21eb180 100644

drivers/gpu/drm/drm_lock.c int drm_legacy_lock(struct drm_device *dev, void *data, · line 64

Quoted 4 lines
drivers/gpu/drm/drm_lock.c +1 -0
@@ -61,6 +61,9 @@ int drm_legacy_lock(struct drm_device *dev, void *data,
 	struct drm_master *master = file_priv->master;
 	int ret = 0;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))

You also need to allow these for all legacy drivers, i.e. without DRIVER_MODESET.
-Daniel

Quoted 66 lines
drivers/gpu/drm/drm_lock.c +5 -0
+		return -EINVAL;
+
 	++file_priv->lock_count;
 
 	if (lock->context == DRM_KERNEL_CONTEXT) {
@@ -153,6 +156,9 @@ int drm_legacy_unlock(struct drm_device *dev, void *data, struct drm_file *file_
 	struct drm_lock *lock = data;
 	struct drm_master *master = file_priv->master;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	if (lock->context == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
 			  task_pid_nr(current), lock->context);
drivers/gpu/drm/nouveau/nouveau_drm.c +2 -1
index 8904933..9624b38 100644
@@ -941,7 +941,8 @@ static struct drm_driver
 driver_stub = {
 	.driver_features =
 		DRIVER_USE_AGP |
-		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER,
+		DRIVER_GEM | DRIVER_MODESET | DRIVER_PRIME | DRIVER_RENDER |
+		DRIVER_KMS_LEGACY_CONTEXT,
 
 	.load = nouveau_drm_load,
 	.unload = nouveau_drm_unload,
include/drm/drmP.h +12 -11
index df6d997..3874942 100644
@@ -137,17 +137,18 @@ void drm_err(const char *format, ...);
 /*@{*/
 
 /* driver capabilities and requirements mask */
-#define DRIVER_USE_AGP     0x1
-#define DRIVER_PCI_DMA     0x8
-#define DRIVER_SG          0x10
-#define DRIVER_HAVE_DMA    0x20
-#define DRIVER_HAVE_IRQ    0x40
-#define DRIVER_IRQ_SHARED  0x80
-#define DRIVER_GEM         0x1000
-#define DRIVER_MODESET     0x2000
-#define DRIVER_PRIME       0x4000
-#define DRIVER_RENDER      0x8000
-#define DRIVER_ATOMIC      0x10000
+#define DRIVER_USE_AGP			0x1
+#define DRIVER_PCI_DMA			0x8
+#define DRIVER_SG			0x10
+#define DRIVER_HAVE_DMA			0x20
+#define DRIVER_HAVE_IRQ			0x40
+#define DRIVER_IRQ_SHARED		0x80
+#define DRIVER_GEM			0x1000
+#define DRIVER_MODESET			0x2000
+#define DRIVER_PRIME			0x4000
+#define DRIVER_RENDER			0x8000
+#define DRIVER_ATOMIC			0x10000
+#define DRIVER_KMS_LEGACY_CONTEXT	0x20000
 
 /***********************************************************************/
 /** \name Macros to make printk easier */
Signature · 1 line
1.9.1
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
#34 lore raw

PATCH v2 2/2 drm: Make Legacy Context access functions optional.

In reply to Peter Antoine

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

These changes are based on the two patches:
commit c21eb21cb50d58e7cbdcb8b9e7ff68b85cfa5095
Author: Dave Airlie <[email protected]>

And the commit that the above patch reverts:
commit 7c510133d93dd6f15ca040733ba7b2891ed61fd1
Author: Daniel Vetter <[email protected]>

This should now turn off the context feature.

Issue: VIZ-5485

---
 drivers/gpu/drm/drm_context.c | 36 ++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/drm_drv.c     | 12 +++++++-----
 2 files changed, 43 insertions(+), 5 deletions(-)
drivers/gpu/drm/drm_context.c +36 -0
index 9b23525..574be2a 100644
@@ -53,6 +53,9 @@ struct drm_ctx_list {
  */
 void drm_legacy_ctxbitmap_free(struct drm_device * dev, int ctx_handle)
 {
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->struct_mutex);
 	idr_remove(&dev->ctx_idr, ctx_handle);
 	mutex_unlock(&dev->struct_mutex);
@@ -87,6 +90,9 @@ static int drm_legacy_ctxbitmap_next(struct drm_device * dev)
  */
 int drm_legacy_ctxbitmap_init(struct drm_device * dev)
 {
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	idr_init(&dev->ctx_idr);
 	return 0;
 }
@@ -101,6 +107,9 @@ int drm_legacy_ctxbitmap_init(struct drm_device * dev)
  */
 void drm_legacy_ctxbitmap_cleanup(struct drm_device * dev)
 {
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->struct_mutex);
 	idr_destroy(&dev->ctx_idr);
 	mutex_unlock(&dev->struct_mutex);
@@ -119,6 +128,9 @@ void drm_legacy_ctxbitmap_flush(struct drm_device *dev, struct drm_file *file)
 {
 	struct drm_ctx_list *pos, *tmp;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->ctxlist_mutex);
 
 	list_for_each_entry_safe(pos, tmp, &dev->ctxlist, head) {
@@ -161,6 +173,9 @@ int drm_legacy_getsareactx(struct drm_device *dev, void *data,
 	struct drm_local_map *map;
 	struct drm_map_list *_entry;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->struct_mutex);
 
 	map = idr_find(&dev->ctx_idr, request->ctx_id);
@@ -205,6 +220,9 @@ int drm_legacy_setsareactx(struct drm_device *dev, void *data,
 	struct drm_local_map *map = NULL;
 	struct drm_map_list *r_list = NULL;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->struct_mutex);
 	list_for_each_entry(r_list, &dev->maplist, head) {
 		if (r_list->map
@@ -305,6 +323,9 @@ int drm_legacy_resctx(struct drm_device *dev, void *data,
 	struct drm_ctx ctx;
 	int i;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	if (res->count >= DRM_RESERVED_CONTEXTS) {
 		memset(&ctx, 0, sizeof(ctx));
 		for (i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
@@ -335,6 +356,9 @@ int drm_legacy_addctx(struct drm_device *dev, void *data,
 	struct drm_ctx_list *ctx_entry;
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	ctx->handle = drm_legacy_ctxbitmap_next(dev);
 	if (ctx->handle == DRM_KERNEL_CONTEXT) {
 		/* Skip kernel's context and get a new one. */
@@ -378,6 +402,9 @@ int drm_legacy_getctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	/* This is 0, because we don't handle any context flags */
 	ctx->flags = 0;
 
@@ -400,6 +427,9 @@ int drm_legacy_switchctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	DRM_DEBUG("%d\n", ctx->handle);
 	return drm_context_switch(dev, dev->last_context, ctx->handle);
 }
@@ -420,6 +450,9 @@ int drm_legacy_newctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	DRM_DEBUG("%d\n", ctx->handle);
 	drm_context_switch_complete(dev, file_priv, ctx->handle);
 
@@ -442,6 +475,9 @@ int drm_legacy_rmctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	DRM_DEBUG("%d\n", ctx->handle);
 	if (ctx->handle != DRM_KERNEL_CONTEXT) {
 		if (dev->driver->context_dtor)
drivers/gpu/drm/drm_drv.c +7 -5
index 3b3c4f5..1de2df8 100644
@@ -584,11 +584,13 @@ struct drm_device *drm_dev_alloc(struct drm_driver *driver,
 	if (drm_ht_create(&dev->map_hash, 12))
 		goto err_minors;
 
-	ret = drm_legacy_ctxbitmap_init(dev);
-	if (ret) {
-		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
-		goto err_ht;
-	}
+	if (drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		ret = drm_legacy_ctxbitmap_init(dev);
+		if (ret) {
+			DRM_ERROR(
+				"Cannot allocate memory for context bitmap.\n");
+			goto err_ht;
+		}
 
 	if (drm_core_check_feature(dev, DRIVER_GEM)) {
 		ret = drm_gem_init(dev);
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#35 lore raw

Re: [PATCH v2 2/2] drm: Make Legacy Context access functions optional.

Quoted 7 lines · Peter Antoine

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

These changes are based on the two patches:
commit c21eb21cb50d58e7cbdcb8b9e7ff68b85cfa5095
Author: Dave Airlie <[email protected]>

Quoted 8 lines

And the commit that the above patch reverts:
commit 7c510133d93dd6f15ca040733ba7b2891ed61fd1
Author: Daniel Vetter <[email protected]>

This should now turn off the context feature.

Issue: VIZ-5485

Please cc drm core patches to dri-devel. Review below. -Daniel

Quoted 10 lines
---
 drivers/gpu/drm/drm_context.c | 36 ++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/drm_drv.c     | 12 +++++++-----
 2 files changed, 43 insertions(+), 5 deletions(-)
drivers/gpu/drm/drm_context.c +0 -0
index 9b23525..574be2a 100644

drivers/gpu/drm/drm_context.c struct drm_ctx_list { · line 57

Quoted 5 lines
drivers/gpu/drm/drm_context.c +2 -0
@@ -53,6 +53,9 @@ struct drm_ctx_list {
  */
 void drm_legacy_ctxbitmap_free(struct drm_device * dev, int ctx_handle)
 {
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;

This doesn't compile too well. Also like with the previous patch, drivers without DRIVER_MODESET still need these ioctls.

drivers/gpu/drm/drm_context.c static int drm_legacy_ctxbitmap_next(struct drm_device * dev) · line 94

Quoted 10 lines
drivers/gpu/drm/drm_context.c +3 -0
+
 	mutex_lock(&dev->struct_mutex);
 	idr_remove(&dev->ctx_idr, ctx_handle);
 	mutex_unlock(&dev->struct_mutex);
@@ -87,6 +90,9 @@ static int drm_legacy_ctxbitmap_next(struct drm_device * dev)
  */
 int drm_legacy_ctxbitmap_init(struct drm_device * dev)
 {
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;

This is redundant with the check you've added around the caller. Looking at other places wrapping callers of _legacy_ functions with these checks is the more common pattern in drm.

drivers/gpu/drm/drm_context.c int drm_legacy_ctxbitmap_init(struct drm_device * dev) · line 111

Quoted 10 lines
drivers/gpu/drm/drm_context.c +3 -0
+
 	idr_init(&dev->ctx_idr);
 	return 0;
 }
@@ -101,6 +107,9 @@ int drm_legacy_ctxbitmap_init(struct drm_device * dev)
  */
 void drm_legacy_ctxbitmap_cleanup(struct drm_device * dev)
 {
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;

Same issue with compiling. And probably better to move the check out into callers.

drivers/gpu/drm/drm_context.c void drm_legacy_ctxbitmap_flush(struct drm_device *dev, struct drm_file *file) · line 132

Quoted 10 lines
drivers/gpu/drm/drm_context.c +3 -0
+
 	mutex_lock(&dev->struct_mutex);
 	idr_destroy(&dev->ctx_idr);
 	mutex_unlock(&dev->struct_mutex);
@@ -119,6 +128,9 @@ void drm_legacy_ctxbitmap_flush(struct drm_device *dev, struct drm_file *file)
 {
 	struct drm_ctx_list *pos, *tmp;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;

Same as above.

Quoted 89 lines
drivers/gpu/drm/drm_context.c +25 -0
+
 	mutex_lock(&dev->ctxlist_mutex);
 
 	list_for_each_entry_safe(pos, tmp, &dev->ctxlist, head) {
@@ -161,6 +173,9 @@ int drm_legacy_getsareactx(struct drm_device *dev, void *data,
 	struct drm_local_map *map;
 	struct drm_map_list *_entry;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->struct_mutex);
 
 	map = idr_find(&dev->ctx_idr, request->ctx_id);
@@ -205,6 +220,9 @@ int drm_legacy_setsareactx(struct drm_device *dev, void *data,
 	struct drm_local_map *map = NULL;
 	struct drm_map_list *r_list = NULL;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	mutex_lock(&dev->struct_mutex);
 	list_for_each_entry(r_list, &dev->maplist, head) {
 		if (r_list->map
@@ -305,6 +323,9 @@ int drm_legacy_resctx(struct drm_device *dev, void *data,
 	struct drm_ctx ctx;
 	int i;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	if (res->count >= DRM_RESERVED_CONTEXTS) {
 		memset(&ctx, 0, sizeof(ctx));
 		for (i = 0; i < DRM_RESERVED_CONTEXTS; i++) {
@@ -335,6 +356,9 @@ int drm_legacy_addctx(struct drm_device *dev, void *data,
 	struct drm_ctx_list *ctx_entry;
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	ctx->handle = drm_legacy_ctxbitmap_next(dev);
 	if (ctx->handle == DRM_KERNEL_CONTEXT) {
 		/* Skip kernel's context and get a new one. */
@@ -378,6 +402,9 @@ int drm_legacy_getctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	/* This is 0, because we don't handle any context flags */
 	ctx->flags = 0;
 
@@ -400,6 +427,9 @@ int drm_legacy_switchctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	DRM_DEBUG("%d\n", ctx->handle);
 	return drm_context_switch(dev, dev->last_context, ctx->handle);
 }
@@ -420,6 +450,9 @@ int drm_legacy_newctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	DRM_DEBUG("%d\n", ctx->handle);
 	drm_context_switch_complete(dev, file_priv, ctx->handle);
 
@@ -442,6 +475,9 @@ int drm_legacy_rmctx(struct drm_device *dev, void *data,
 {
 	struct drm_ctx *ctx = data;
 
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;
+
 	DRM_DEBUG("%d\n", ctx->handle);
 	if (ctx->handle != DRM_KERNEL_CONTEXT) {
 		if (dev->driver->context_dtor)
drivers/gpu/drm/drm_drv.c +0 -0
index 3b3c4f5..1de2df8 100644

drivers/gpu/drm/drm_drv.c struct drm_device *drm_dev_alloc(struct drm_driver *driver, · line 587

Quoted 9 lines
drivers/gpu/drm/drm_drv.c +1 -5
@@ -584,11 +584,13 @@ struct drm_device *drm_dev_alloc(struct drm_driver *driver,
 	if (drm_ht_create(&dev->map_hash, 12))
 		goto err_minors;
 
-	ret = drm_legacy_ctxbitmap_init(dev);
-	if (ret) {
-		DRM_ERROR("Cannot allocate memory for context bitmap.\n");
-		goto err_ht;
-	}
+	if (drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))

Coding style requires {} since it's multi-line. And confusing since the if (ret) gets always executed.

Quoted 12 lines
drivers/gpu/drm/drm_drv.c +6 -0
+		ret = drm_legacy_ctxbitmap_init(dev);
+		if (ret) {
+			DRM_ERROR(
+				"Cannot allocate memory for context bitmap.\n");
+			goto err_ht;
+		}
 
 	if (drm_core_check_feature(dev, DRIVER_GEM)) {
 		ret = drm_gem_init(dev);
Signature · 1 line
1.9.1
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#36 lore raw

Re: [PATCH v2 2/2] drm: Make Legacy Context access functions optional.

Quoted 25 lines · Daniel Vetter
Quoted 7 lines · Peter Antoine

As these functions are only used by one driver and there are security holes in these functions. Make the functions optional.

These changes are based on the two patches:
commit c21eb21cb50d58e7cbdcb8b9e7ff68b85cfa5095
Author: Dave Airlie <[email protected]>

Quoted 8 lines

And the commit that the above patch reverts:
commit 7c510133d93dd6f15ca040733ba7b2891ed61fd1
Author: Daniel Vetter <[email protected]>

This should now turn off the context feature.

Issue: VIZ-5485

Please cc drm core patches to dri-devel. Review below. -Daniel

Quoted 5 lines
---
 drivers/gpu/drm/drm_context.c | 36 ++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/drm_drv.c     | 12 +++++++-----
 2 files changed, 43 insertions(+), 5 deletions(-)
Quoted 13 lines

drivers/gpu/drm/drm_context.c struct drm_ctx_list { · line 57

Quoted 10 lines
drivers/gpu/drm/drm_context.c +2 -0
index 9b23525..574be2a 100644
@@ -53,6 +53,9 @@ struct drm_ctx_list {
  */
 void drm_legacy_ctxbitmap_free(struct drm_device * dev, int ctx_handle)
 {
+	if (!drm_core_check_feature(dev, DRIVER_KMS_LEGACY_CONTEXT))
+		return -EINVAL;

This doesn't compile too well. Also like with the previous patch, drivers without DRIVER_MODESET still need these ioctls.

Only via actually needs them AFAIK, but continuing to allow them for the rest of the old drivers is no worse than what we have today.

And I think nouveau just needs a nop create/destroy ioctls, so I think we should really just make those two nops for nouveau (or maybe for all modern drivers?) and reject the rest of the ioctls even for nouveau.

Signature · 6 lines
Ville Syrjälä
Intel OTC
_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
Tested-by on v2 2/2
#37 lore raw

Re: [PATCH v2 2/2] drm: Make Legacy Context access functions optional.

In reply to Peter Antoine

Task id: 6394 -------------------------------------Summary-------------------------------------
Platform Delta drm-intel-nightly Series Applied
PNV -2 272/272 270/272
ILK 302/302 302/302
SNB -1 315/315 314/315

IVB                                  343/343              343/343
BYT                                  287/287              287/287
BDW                                  317/317              317/317
-------------------------------------Detailed-------------------------------------
Platform  Test                                drm-intel-nightly          Series Applied
*PNV  igt@gem_tiled_pread_pwrite      PASS(4)      FAIL(1)
*PNV  igt@gem_userptr_blits@coherency-sync      PASS(4)      CRASH(1)
 SNB  igt@pm_rpm@dpms-mode-unset-non-lpsp      DMESG_WARN(7)PASS(1)      DMESG_WARN(1)

(dmesg patch applied)WARNING:at_drivers/gpu/drm/i915/intel_uncore.c:#assert_device_not_suspended[i915]()@WARNING:.* at .* assert_device_not_suspended+0x
Note: You need to pay more attention to line start with '*'
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#38 lore raw

Re: [PATCH v2 0/2] HW_LOCK kernel patched.

In reply to Peter Antoine
Quoted 5 lines · Peter Antoine

This is a resubmission of the security fixed for a Kernel Crash found in the drm part of the driver. It was caused by a attempt to Lock/Unlock a HW_LOCK that was uninitialised. As this feature is only required on some legacy drivers this patchset now only turns these features off.

Quoted 7 lines

There is an igt test to follow, just working out a way of not crashing the systems that have this feature-set turned on (probably need to create a context and if that succeeds then exit as the feature is on.

V2: Remove the patches that cover the Kernel Crash and the Lock Escalation.
Remove the PARAM that was used by the test to detect the presences of this feature.

Process nit: When resending the entire patch series anew please start a new thread. That avoids confusion with the discussion on the old patches (I've managed to wreak havoc and merge a mix of old&new patches this way already). in-reply-to resending is just for resending parts of the series when you do a quick update of a patch.
-Daniel

Quoted 15 lines

Peter Antoine (2):
drm: Make HW_LOCK access functions optional.
drm: Make Legacy Context access functions optional.

 drivers/gpu/drm/drm_context.c         | 36 +++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/drm_drv.c             | 12 +++++++-----
 drivers/gpu/drm/drm_lock.c            |  6 ++++++
 drivers/gpu/drm/nouveau/nouveau_drm.c |  3 ++-
 include/drm/drmP.h                    | 23 +++++++++++-----------
 5 files changed, 63 insertions(+), 17 deletions(-)
Signature · 1 line
1.9.1
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx