[PATCH RFC] drm/amdgpu: Block userspace mapping of IO

[PATCH RFC] drm/amdgpu: Block userspace mapping of IO

[Ujwal Kundur]

This is a RFC patch for blocking userspace mapping of IO register(s)
before ioremap() calls are made. Out of the available IRQ sources, CRTC
seemed the most appropriate for this task, however I'm not quite sure
about that as well as the type, which I've set to 0.

If I understand correctly, we actually want to block certain ioctls from
userspace that can interfere with ioremap but I don't see a dedicated
source for that.

Signed-off-by: Ujwal Kundur <ujwal.kundur@gmail.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index a30111d2c3ea..365af52af6e5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -47,6 +47,7 @@
 #include <linux/vga_switcheroo.h>
 #include <linux/efi.h>
 #include "amdgpu.h"
+#include "amdgpu_irq.h"
 #include "amdgpu_trace.h"
 #include "amdgpu_i2c.h"
 #include "atom.h"
@@ -4367,7 +4368,9 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	ratelimit_set_flags(&adev->throttling_logging_rs, RATELIMIT_MSG_ON_RELEASE);
 
 	/* Registers mapping */
-	/* TODO: block userspace mapping of io register */
+	/* Block userspace mapping of io register */
+	amdgpu_irq_put(adev, &adev->crtc_irq, 0);
+
 	if (adev->asic_type >= CHIP_BONAIRE) {
 		adev->rmmio_base = pci_resource_start(adev->pdev, 5);
 		adev->rmmio_size = pci_resource_len(adev->pdev, 5);
@@ -4380,6 +4383,9 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 		atomic_set(&adev->pm.pwr_state[i], POWER_STATE_UNKNOWN);
 
 	adev->rmmio = ioremap(adev->rmmio_base, adev->rmmio_size);
+
+	amdgpu_irq_get(adev, &adev->crtc_irq, 0);
+
 	if (!adev->rmmio)
 		return -ENOMEM;
 
-- 
2.20.1

Re: [PATCH RFC] drm/amdgpu: Block userspace mapping of IO

[Christian König]

Am 16.04.25 um 09:28 schrieb Ujwal Kundur:
> This is a RFC patch for blocking userspace mapping of IO register(s)
> before ioremap() calls are made. Out of the available IRQ sources, CRTC
> seemed the most appropriate for this task, however I'm not quite sure
> about that as well as the type, which I've set to 0.

Hui what? Why do you think that grabbing a reference to an interrupt would block userspace mapping of IO registers?

Christian.

>
> If I understand correctly, we actually want to block certain ioctls from
> userspace that can interfere with ioremap but I don't see a dedicated
> source for that.
>
> Signed-off-by: Ujwal Kundur <ujwal.kundur@gmail.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index a30111d2c3ea..365af52af6e5 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -47,6 +47,7 @@
>  #include <linux/vga_switcheroo.h>
>  #include <linux/efi.h>
>  #include "amdgpu.h"
> +#include "amdgpu_irq.h"
>  #include "amdgpu_trace.h"
>  #include "amdgpu_i2c.h"
>  #include "atom.h"
> @@ -4367,7 +4368,9 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>  	ratelimit_set_flags(&adev->throttling_logging_rs, RATELIMIT_MSG_ON_RELEASE);
>  
>  	/* Registers mapping */
> -	/* TODO: block userspace mapping of io register */
> +	/* Block userspace mapping of io register */
> +	amdgpu_irq_put(adev, &adev->crtc_irq, 0);
> +
>  	if (adev->asic_type >= CHIP_BONAIRE) {
>  		adev->rmmio_base = pci_resource_start(adev->pdev, 5);
>  		adev->rmmio_size = pci_resource_len(adev->pdev, 5);
> @@ -4380,6 +4383,9 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>  		atomic_set(&adev->pm.pwr_state[i], POWER_STATE_UNKNOWN);
>  
>  	adev->rmmio = ioremap(adev->rmmio_base, adev->rmmio_size);
> +
> +	amdgpu_irq_get(adev, &adev->crtc_irq, 0);
> +
>  	if (!adev->rmmio)
>  		return -ENOMEM;
>  

Re: [PATCH RFC] drm/amdgpu: Block userspace mapping of IO

[Ujwal Kundur]

Thanks for your response.

> Hui what? Why do you think that grabbing a reference to an interrupt would block userspace mapping of IO registers?

It looks like I am missing a lot of pieces to do this, I'll try again
once I have a better understanding.

Sorry about that and thanks again for your time.

--- Ujwal.

Re: [PATCH RFC] drm/amdgpu: Block userspace mapping of IO

[Christian König]

Am 16.04.25 um 13:43 schrieb Ujwal Kundur:
> Thanks for your response.
>
>> Hui what? Why do you think that grabbing a reference to an interrupt would block userspace mapping of IO registers?
> It looks like I am missing a lot of pieces to do this, I'll try again
> once I have a better understanding.
>
> Sorry about that and thanks again for your time.

No problem. If you have questions about the driver stack then just ask.

Regards,
Christian.

>
> --- Ujwal.