linux-stable.git/drivers/gpu/drm/amd/amdkfd, branch v4.6.7

drm/amdkfd: destroy dbgmgr in notifier release

2016-07-27T15:42:17+00:00

commit bc4755a4bd1845ef6e88ac8c62f12e05bb530256 upstream.

amdkfd need to destroy the debug manager in case amdkfd's notifier
function is called before the unbind function, because in that case,
the unbind function will exit without destroying debug manager.

Signed-off-by: Oded Gabbay 
Signed-off-by: Greg Kroah-Hartman

drm/amdkfd: unbind only existing processes

2016-07-27T15:42:17+00:00

commit 121b78e679ee3ffab780115e260b2775d0cc1f73 upstream.

When unbinding a process from a device (initiated by amd_iommu_v2), the
driver needs to make sure that process still exists in the process table.
There is a possibility that amdkfd's own notifier handler -
kfd_process_notifier_release() - was called before the unbind function
and it already removed the process from the process table.

v2:
Because there can be only one process with the specified pasid, and
because *p can't be NULL inside the hash_for_each_rcu macro, it is more
reasonable to just put the whole code inside the if statement that
compares the pasid value. That way, when we exit hash_for_each_rcu, we
simply exit the function as well.

Signed-off-by: Oded Gabbay 
Signed-off-by: Greg Kroah-Hartman

drivers/gpu/drm/amd/amdkfd: use in_compat_syscall to check open() caller type

2016-03-22T22:36:02+00:00

amdkfd wants to know syscall type, not task type.  Check directly.

Unfortunately, amdkfd is making nasty assumptions that a process'
bitness is a well-defined constant thing.  This isn't the case on x86.
I don't know how much this matters, but this patch has no effect on
generated code on x86, so amdkfd is equally broken with and without this
patch.

Signed-off-by: Andy Lutomirski 
Cc: Oded Gabbay 
Cc: David Airlie 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

drm/amdkfd: uninitialized variable in dbgdev_wave_control_set_registers()

2016-03-15T12:09:37+00:00

At the end of the function we expect "status" to be zero, but it's
either -EINVAL or uninitialized.

Fixes: 788bf83db301 ('drm/amdkfd: Add wave control operation to debugger')
Signed-off-by: Dan Carpenter 
Signed-off-by: Oded Gabbay

drm/amdkfd: Track when module's init is complete

2016-02-27T20:52:40+00:00

Current dependencies between amdkfd and radeon/amdgpu force the loading
of amdkfd _before_ radeon and/or amdgpu are loaded. When all these kernel
drivers are built as modules, this ordering is enforced by the kernel
built-in mechanism of loading dependent modules.

However, there is no such mechanism in case where all these drivers are
compiled inside the kernel image (not as modules). The current way to
enforce loading of amdkfd before radeon/amdgpu, is to put amdkfd before
radeon/amdgpu in the drm Makefile, but that method is way too fragile.

In addition, there is no kernel mechanism to check whether a kernel
driver that is built inside the kernel image, has already been loaded.

To solve this, this patch adds to kfd_module.c a new static variable,
amdkfd_init_completed, that is set to 1 only when amdkfd's
module initialization function has been completed (successfully).

kgd2kfd_init(), which is the initialization function of the
kgd-->kfd interface, and which is the first function in amdkfd called by
radeon/amdgpu, will return successfully only if amdkfd_init_completed is
equal 1.

If amdkfd_init_completed is not equal to 1, kgd2kfd_init() will
return -EPROBE_DEFER to signal radeon/amdgpu they need to defer
their loading until amdkfd is loaded.

Signed-off-by: Oded Gabbay 
Reviewed-by: Alex Deucher

drm/amdkfd: Remove unnecessary cast in kfree

2016-01-28T12:40:11+00:00

Remove an unnecassary cast in the argument to kfree.

Found using Coccinelle. The semantic patch used to find this is as follows:

//
@@
type T;
expression *f;
@@

- kfree((T *)(f));
+ kfree(f);
//

Signed-off-by: Amitoj Kaur Chawla 
Signed-off-by: Oded Gabbay

amdkfd: Copy from the proper user command pointer

2016-01-15T20:14:17+00:00

8f1d57c17248 ("amdkfd: don't open-code memdup_user()") mistakenly uses
an uninitialized local pointer, gcc complains:

  drivers/gpu/drm/amd/amdkfd/kfd_chardev.c: In function ‘kfd_ioctl_dbg_address_watch’:
  drivers/gpu/drm/amd/amdkfd/kfd_chardev.c:562:12: warning: ‘args_buff’ may be used uninitialized in this function [-Wmaybe-uninitialized]
    args_buff = memdup_user(args_buff,
                ^

Fix it.

Signed-off-by: Borislav Petkov 
Cc: Al Viro 
Signed-off-by: Al Viro

amdkfd: don't open-code memdup_user()

2016-01-06T13:25:25+00:00

Signed-off-by: Al Viro

amdkfd: use instead of

2015-08-30T09:36:58+00:00

The latter is a default version of  and not for driver use.

Signed-off-by: Christoph Hellwig 
Signed-off-by: Oded Gabbay

drm/amdkfd: fix bug when initializing sdma vm

2015-07-30T06:26:15+00:00

A logical AND operation was used during mask and shift, instead of a
bitwise AND operation. This patch fixes this bug by changing the
operation to bitwise AND.

Signed-off-by: Oded Gabbay 
Reviewed-by: Alex Deucher