[PATCH v7] drm/amdgpu: replace PASID IDR with XArray

[PATCH v7] drm/amdgpu: replace PASID IDR with XArray

[Mikhail Gavrilov]

Replace the PASID IDR + spinlock with XArray as noted in the TODO
left by commit dccd79bb1c7f ("drm/amdgpu: fix the idr allocation
flags").

The IDR conversion still has an IRQ safety issue:
amdgpu_pasid_free() can be called from hardirq context via the fence
signal path, but amdgpu_pasid_idr_lock is taken with plain spin_lock()
in process context, creating a potential deadlock:

     CPU0
     ----
     spin_lock(&amdgpu_pasid_idr_lock)   // process context, IRQs on
     <Interrupt>
       spin_lock(&amdgpu_pasid_idr_lock) // deadlock

   The hardirq call chain is:

     sdma_v6_0_process_trap_irq
      -> amdgpu_fence_process
       -> dma_fence_signal
        -> drm_sched_job_done
         -> dma_fence_signal
          -> amdgpu_pasid_free_cb
           -> amdgpu_pasid_free

Use XArray with XA_FLAGS_LOCK_IRQ (all xa operations use IRQ-safe
locking internally) and XA_FLAGS_ALLOC1 (zero is not a valid PASID).
Both xa_alloc_cyclic() and xa_erase() then handle locking
consistently, fixing the IRQ safety issue and removing the need for
an explicit spinlock.

Suggested-by: Lijo Lazar <lijo.lazar@amd.com>
Fixes: e6d765de3d6b ("drm/amdgpu: prevent immediate PASID reuse case")
Signed-off-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
---

v7: Rebased on amd-staging-drm-next which already includes
    dccd79bb1c7f ("drm/amdgpu: fix the idr allocation flags").
    Updated commit message to reflect that sleeping-under-spinlock
    is already fixed and the xarray conversion now addresses the
    remaining IRQ safety issue.  Inverted error check to
    if (r < 0) return r; per Christian König.
v6: Use DEFINE_XARRAY_FLAGS with XA_FLAGS_LOCK_IRQ | XA_FLAGS_ALLOC1
    so all xa operations use IRQ-safe locking internally.  Drop
    Cc: stable since the regression was never released to any stable
    kernel. (Christian König)
    https://lore.kernel.org/all/20260331111733.118553-1-mikhail.v.gavrilov@gmail.com/
v5: Use explicit xa_lock_irqsave/__xa_erase for amdgpu_pasid_free()
    since xa_erase() only uses plain xa_lock() which is not safe from
    hardirq context.
    https://lore.kernel.org/all/20260330191120.105065-1-mikhail.v.gavrilov@gmail.com/
v4: Use xa_alloc_cyclic/xa_erase directly instead of explicit
    xa_lock_irqsave, as suggested by Lijo Lazar.
    https://lore.kernel.org/all/20260330162038.25073-1-mikhail.v.gavrilov@gmail.com/
v3: Replace IDR with XArray instead of fixing the spinlock, as
    suggested by Lijo Lazar.
    https://lore.kernel.org/all/20260330110346.16548-1-mikhail.v.gavrilov@gmail.com/
v2: Added second patch fixing the {HARDIRQ-ON-W} -> {IN-HARDIRQ-W}
    lock inconsistency (spin_lock -> spin_lock_irqsave).
    https://lore.kernel.org/all/20260330053025.19203-1-mikhail.v.gavrilov@gmail.com/
v1: Fixed sleeping-under-spinlock (idr_alloc_cyclic with GFP_KERNEL)
    using idr_preload/GFP_NOWAIT.
    https://lore.kernel.org/all/20260328213900.19255-1-mikhail.v.gavrilov@gmail.com/

 drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c | 34 ++++++++++---------------
 1 file changed, 13 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
index e495a8fa13fd..a6ac3b4ce0df 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
@@ -22,7 +22,7 @@
  */
 #include "amdgpu_ids.h"
 
-#include <linux/idr.h>
+#include <linux/xarray.h>
 #include <linux/dma-fence-array.h>
 
 
@@ -40,8 +40,8 @@
  * VMs are looked up from the PASID per amdgpu_device.
  */
 
-static DEFINE_IDR(amdgpu_pasid_idr);
-static DEFINE_SPINLOCK(amdgpu_pasid_idr_lock);
+static DEFINE_XARRAY_FLAGS(amdgpu_pasid_xa, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ALLOC1);
+static u32 amdgpu_pasid_xa_next;
 
 /* Helper to free pasid from a fence callback */
 struct amdgpu_pasid_cb {
@@ -62,22 +62,19 @@ struct amdgpu_pasid_cb {
  */
 int amdgpu_pasid_alloc(unsigned int bits)
 {
-	int pasid;
+	u32 pasid;
+	int r;
 
 	if (bits == 0)
 		return -EINVAL;
 
-	spin_lock(&amdgpu_pasid_idr_lock);
-	/* TODO: Need to replace the idr with an xarry, and then
-	 * handle the internal locking with ATOMIC safe paths.
-	 */
-	pasid = idr_alloc_cyclic(&amdgpu_pasid_idr, NULL, 1,
-				 1U << bits, GFP_ATOMIC);
-	spin_unlock(&amdgpu_pasid_idr_lock);
-
-	if (pasid >= 0)
-		trace_amdgpu_pasid_allocated(pasid);
+	r = xa_alloc_cyclic(&amdgpu_pasid_xa, &pasid, xa_mk_value(0),
+			    XA_LIMIT(1, (1U << bits) - 1),
+			    &amdgpu_pasid_xa_next, GFP_KERNEL);
+	if (r < 0)
+		return r;
 
+	trace_amdgpu_pasid_allocated(pasid);
 	return pasid;
 }
 
@@ -88,10 +85,7 @@ int amdgpu_pasid_alloc(unsigned int bits)
 void amdgpu_pasid_free(u32 pasid)
 {
 	trace_amdgpu_pasid_freed(pasid);
-
-	spin_lock(&amdgpu_pasid_idr_lock);
-	idr_remove(&amdgpu_pasid_idr, pasid);
-	spin_unlock(&amdgpu_pasid_idr_lock);
+	xa_erase(&amdgpu_pasid_xa, pasid);
 }
 
 static void amdgpu_pasid_free_cb(struct dma_fence *fence,
@@ -634,7 +628,5 @@ void amdgpu_vmid_mgr_fini(struct amdgpu_device *adev)
  */
 void amdgpu_pasid_mgr_cleanup(void)
 {
-	spin_lock(&amdgpu_pasid_idr_lock);
-	idr_destroy(&amdgpu_pasid_idr);
-	spin_unlock(&amdgpu_pasid_idr_lock);
+	xa_destroy(&amdgpu_pasid_xa);
 }
-- 
2.53.0

Re: [PATCH v7] drm/amdgpu: replace PASID IDR with XArray

[Christian König]

On 3/31/26 16:21, Mikhail Gavrilov wrote:
> Replace the PASID IDR + spinlock with XArray as noted in the TODO
> left by commit dccd79bb1c7f ("drm/amdgpu: fix the idr allocation
> flags").
> 
> The IDR conversion still has an IRQ safety issue:
> amdgpu_pasid_free() can be called from hardirq context via the fence
> signal path, but amdgpu_pasid_idr_lock is taken with plain spin_lock()
> in process context, creating a potential deadlock:
> 
>      CPU0
>      ----
>      spin_lock(&amdgpu_pasid_idr_lock)   // process context, IRQs on
>      <Interrupt>
>        spin_lock(&amdgpu_pasid_idr_lock) // deadlock
> 
>    The hardirq call chain is:
> 
>      sdma_v6_0_process_trap_irq
>       -> amdgpu_fence_process
>        -> dma_fence_signal
>         -> drm_sched_job_done
>          -> dma_fence_signal
>           -> amdgpu_pasid_free_cb
>            -> amdgpu_pasid_free
> 
> Use XArray with XA_FLAGS_LOCK_IRQ (all xa operations use IRQ-safe
> locking internally) and XA_FLAGS_ALLOC1 (zero is not a valid PASID).
> Both xa_alloc_cyclic() and xa_erase() then handle locking
> consistently, fixing the IRQ safety issue and removing the need for
> an explicit spinlock.
> 
> Suggested-by: Lijo Lazar <lijo.lazar@amd.com>
> Fixes: e6d765de3d6b ("drm/amdgpu: prevent immediate PASID reuse case")
> Signed-off-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>

Reviewed-by: Christian König <christian.koenig@amd.com>

> ---
> 
> v7: Rebased on amd-staging-drm-next which already includes
>     dccd79bb1c7f ("drm/amdgpu: fix the idr allocation flags").
>     Updated commit message to reflect that sleeping-under-spinlock
>     is already fixed and the xarray conversion now addresses the
>     remaining IRQ safety issue.  Inverted error check to
>     if (r < 0) return r; per Christian König.
> v6: Use DEFINE_XARRAY_FLAGS with XA_FLAGS_LOCK_IRQ | XA_FLAGS_ALLOC1
>     so all xa operations use IRQ-safe locking internally.  Drop
>     Cc: stable since the regression was never released to any stable
>     kernel. (Christian König)
>     https://lore.kernel.org/all/20260331111733.118553-1-mikhail.v.gavrilov@gmail.com/
> v5: Use explicit xa_lock_irqsave/__xa_erase for amdgpu_pasid_free()
>     since xa_erase() only uses plain xa_lock() which is not safe from
>     hardirq context.
>     https://lore.kernel.org/all/20260330191120.105065-1-mikhail.v.gavrilov@gmail.com/
> v4: Use xa_alloc_cyclic/xa_erase directly instead of explicit
>     xa_lock_irqsave, as suggested by Lijo Lazar.
>     https://lore.kernel.org/all/20260330162038.25073-1-mikhail.v.gavrilov@gmail.com/
> v3: Replace IDR with XArray instead of fixing the spinlock, as
>     suggested by Lijo Lazar.
>     https://lore.kernel.org/all/20260330110346.16548-1-mikhail.v.gavrilov@gmail.com/
> v2: Added second patch fixing the {HARDIRQ-ON-W} -> {IN-HARDIRQ-W}
>     lock inconsistency (spin_lock -> spin_lock_irqsave).
>     https://lore.kernel.org/all/20260330053025.19203-1-mikhail.v.gavrilov@gmail.com/
> v1: Fixed sleeping-under-spinlock (idr_alloc_cyclic with GFP_KERNEL)
>     using idr_preload/GFP_NOWAIT.
>     https://lore.kernel.org/all/20260328213900.19255-1-mikhail.v.gavrilov@gmail.com/
> 
>  drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c | 34 ++++++++++---------------
>  1 file changed, 13 insertions(+), 21 deletions(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
> index e495a8fa13fd..a6ac3b4ce0df 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
> @@ -22,7 +22,7 @@
>   */
>  #include "amdgpu_ids.h"
>  
> -#include <linux/idr.h>
> +#include <linux/xarray.h>
>  #include <linux/dma-fence-array.h>
>  
>  
> @@ -40,8 +40,8 @@
>   * VMs are looked up from the PASID per amdgpu_device.
>   */
>  
> -static DEFINE_IDR(amdgpu_pasid_idr);
> -static DEFINE_SPINLOCK(amdgpu_pasid_idr_lock);
> +static DEFINE_XARRAY_FLAGS(amdgpu_pasid_xa, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ALLOC1);
> +static u32 amdgpu_pasid_xa_next;
>  
>  /* Helper to free pasid from a fence callback */
>  struct amdgpu_pasid_cb {
> @@ -62,22 +62,19 @@ struct amdgpu_pasid_cb {
>   */
>  int amdgpu_pasid_alloc(unsigned int bits)
>  {
> -	int pasid;
> +	u32 pasid;
> +	int r;
>  
>  	if (bits == 0)
>  		return -EINVAL;
>  
> -	spin_lock(&amdgpu_pasid_idr_lock);
> -	/* TODO: Need to replace the idr with an xarry, and then
> -	 * handle the internal locking with ATOMIC safe paths.
> -	 */
> -	pasid = idr_alloc_cyclic(&amdgpu_pasid_idr, NULL, 1,
> -				 1U << bits, GFP_ATOMIC);
> -	spin_unlock(&amdgpu_pasid_idr_lock);
> -
> -	if (pasid >= 0)
> -		trace_amdgpu_pasid_allocated(pasid);
> +	r = xa_alloc_cyclic(&amdgpu_pasid_xa, &pasid, xa_mk_value(0),
> +			    XA_LIMIT(1, (1U << bits) - 1),
> +			    &amdgpu_pasid_xa_next, GFP_KERNEL);
> +	if (r < 0)
> +		return r;
>  
> +	trace_amdgpu_pasid_allocated(pasid);
>  	return pasid;
>  }
>  
> @@ -88,10 +85,7 @@ int amdgpu_pasid_alloc(unsigned int bits)
>  void amdgpu_pasid_free(u32 pasid)
>  {
>  	trace_amdgpu_pasid_freed(pasid);
> -
> -	spin_lock(&amdgpu_pasid_idr_lock);
> -	idr_remove(&amdgpu_pasid_idr, pasid);
> -	spin_unlock(&amdgpu_pasid_idr_lock);
> +	xa_erase(&amdgpu_pasid_xa, pasid);
>  }
>  
>  static void amdgpu_pasid_free_cb(struct dma_fence *fence,
> @@ -634,7 +628,5 @@ void amdgpu_vmid_mgr_fini(struct amdgpu_device *adev)
>   */
>  void amdgpu_pasid_mgr_cleanup(void)
>  {
> -	spin_lock(&amdgpu_pasid_idr_lock);
> -	idr_destroy(&amdgpu_pasid_idr);
> -	spin_unlock(&amdgpu_pasid_idr_lock);
> +	xa_destroy(&amdgpu_pasid_xa);
>  }

Re: [PATCH v7] drm/amdgpu: replace PASID IDR with XArray

[Alex Deucher]

Applied.  Thanks!

Alex

On Tue, Mar 31, 2026 at 10:29 AM Christian König
<christian.koenig@amd.com> wrote:
>
>
>
> On 3/31/26 16:21, Mikhail Gavrilov wrote:
> > Replace the PASID IDR + spinlock with XArray as noted in the TODO
> > left by commit dccd79bb1c7f ("drm/amdgpu: fix the idr allocation
> > flags").
> >
> > The IDR conversion still has an IRQ safety issue:
> > amdgpu_pasid_free() can be called from hardirq context via the fence
> > signal path, but amdgpu_pasid_idr_lock is taken with plain spin_lock()
> > in process context, creating a potential deadlock:
> >
> >      CPU0
> >      ----
> >      spin_lock(&amdgpu_pasid_idr_lock)   // process context, IRQs on
> >      <Interrupt>
> >        spin_lock(&amdgpu_pasid_idr_lock) // deadlock
> >
> >    The hardirq call chain is:
> >
> >      sdma_v6_0_process_trap_irq
> >       -> amdgpu_fence_process
> >        -> dma_fence_signal
> >         -> drm_sched_job_done
> >          -> dma_fence_signal
> >           -> amdgpu_pasid_free_cb
> >            -> amdgpu_pasid_free
> >
> > Use XArray with XA_FLAGS_LOCK_IRQ (all xa operations use IRQ-safe
> > locking internally) and XA_FLAGS_ALLOC1 (zero is not a valid PASID).
> > Both xa_alloc_cyclic() and xa_erase() then handle locking
> > consistently, fixing the IRQ safety issue and removing the need for
> > an explicit spinlock.
> >
> > Suggested-by: Lijo Lazar <lijo.lazar@amd.com>
> > Fixes: e6d765de3d6b ("drm/amdgpu: prevent immediate PASID reuse case")
> > Signed-off-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
>
> Reviewed-by: Christian König <christian.koenig@amd.com>
>
> > ---
> >
> > v7: Rebased on amd-staging-drm-next which already includes
> >     dccd79bb1c7f ("drm/amdgpu: fix the idr allocation flags").
> >     Updated commit message to reflect that sleeping-under-spinlock
> >     is already fixed and the xarray conversion now addresses the
> >     remaining IRQ safety issue.  Inverted error check to
> >     if (r < 0) return r; per Christian König.
> > v6: Use DEFINE_XARRAY_FLAGS with XA_FLAGS_LOCK_IRQ | XA_FLAGS_ALLOC1
> >     so all xa operations use IRQ-safe locking internally.  Drop
> >     Cc: stable since the regression was never released to any stable
> >     kernel. (Christian König)
> >     https://lore.kernel.org/all/20260331111733.118553-1-mikhail.v.gavrilov@gmail.com/
> > v5: Use explicit xa_lock_irqsave/__xa_erase for amdgpu_pasid_free()
> >     since xa_erase() only uses plain xa_lock() which is not safe from
> >     hardirq context.
> >     https://lore.kernel.org/all/20260330191120.105065-1-mikhail.v.gavrilov@gmail.com/
> > v4: Use xa_alloc_cyclic/xa_erase directly instead of explicit
> >     xa_lock_irqsave, as suggested by Lijo Lazar.
> >     https://lore.kernel.org/all/20260330162038.25073-1-mikhail.v.gavrilov@gmail.com/
> > v3: Replace IDR with XArray instead of fixing the spinlock, as
> >     suggested by Lijo Lazar.
> >     https://lore.kernel.org/all/20260330110346.16548-1-mikhail.v.gavrilov@gmail.com/
> > v2: Added second patch fixing the {HARDIRQ-ON-W} -> {IN-HARDIRQ-W}
> >     lock inconsistency (spin_lock -> spin_lock_irqsave).
> >     https://lore.kernel.org/all/20260330053025.19203-1-mikhail.v.gavrilov@gmail.com/
> > v1: Fixed sleeping-under-spinlock (idr_alloc_cyclic with GFP_KERNEL)
> >     using idr_preload/GFP_NOWAIT.
> >     https://lore.kernel.org/all/20260328213900.19255-1-mikhail.v.gavrilov@gmail.com/
> >
> >  drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c | 34 ++++++++++---------------
> >  1 file changed, 13 insertions(+), 21 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
> > index e495a8fa13fd..a6ac3b4ce0df 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
> > @@ -22,7 +22,7 @@
> >   */
> >  #include "amdgpu_ids.h"
> >
> > -#include <linux/idr.h>
> > +#include <linux/xarray.h>
> >  #include <linux/dma-fence-array.h>
> >
> >
> > @@ -40,8 +40,8 @@
> >   * VMs are looked up from the PASID per amdgpu_device.
> >   */
> >
> > -static DEFINE_IDR(amdgpu_pasid_idr);
> > -static DEFINE_SPINLOCK(amdgpu_pasid_idr_lock);
> > +static DEFINE_XARRAY_FLAGS(amdgpu_pasid_xa, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ALLOC1);
> > +static u32 amdgpu_pasid_xa_next;
> >
> >  /* Helper to free pasid from a fence callback */
> >  struct amdgpu_pasid_cb {
> > @@ -62,22 +62,19 @@ struct amdgpu_pasid_cb {
> >   */
> >  int amdgpu_pasid_alloc(unsigned int bits)
> >  {
> > -     int pasid;
> > +     u32 pasid;
> > +     int r;
> >
> >       if (bits == 0)
> >               return -EINVAL;
> >
> > -     spin_lock(&amdgpu_pasid_idr_lock);
> > -     /* TODO: Need to replace the idr with an xarry, and then
> > -      * handle the internal locking with ATOMIC safe paths.
> > -      */
> > -     pasid = idr_alloc_cyclic(&amdgpu_pasid_idr, NULL, 1,
> > -                              1U << bits, GFP_ATOMIC);
> > -     spin_unlock(&amdgpu_pasid_idr_lock);
> > -
> > -     if (pasid >= 0)
> > -             trace_amdgpu_pasid_allocated(pasid);
> > +     r = xa_alloc_cyclic(&amdgpu_pasid_xa, &pasid, xa_mk_value(0),
> > +                         XA_LIMIT(1, (1U << bits) - 1),
> > +                         &amdgpu_pasid_xa_next, GFP_KERNEL);
> > +     if (r < 0)
> > +             return r;
> >
> > +     trace_amdgpu_pasid_allocated(pasid);
> >       return pasid;
> >  }
> >
> > @@ -88,10 +85,7 @@ int amdgpu_pasid_alloc(unsigned int bits)
> >  void amdgpu_pasid_free(u32 pasid)
> >  {
> >       trace_amdgpu_pasid_freed(pasid);
> > -
> > -     spin_lock(&amdgpu_pasid_idr_lock);
> > -     idr_remove(&amdgpu_pasid_idr, pasid);
> > -     spin_unlock(&amdgpu_pasid_idr_lock);
> > +     xa_erase(&amdgpu_pasid_xa, pasid);
> >  }
> >
> >  static void amdgpu_pasid_free_cb(struct dma_fence *fence,
> > @@ -634,7 +628,5 @@ void amdgpu_vmid_mgr_fini(struct amdgpu_device *adev)
> >   */
> >  void amdgpu_pasid_mgr_cleanup(void)
> >  {
> > -     spin_lock(&amdgpu_pasid_idr_lock);
> > -     idr_destroy(&amdgpu_pasid_idr);
> > -     spin_unlock(&amdgpu_pasid_idr_lock);
> > +     xa_destroy(&amdgpu_pasid_xa);
> >  }
>

Re: [PATCH v7] drm/amdgpu: replace PASID IDR with XArray

[Mikhail Gavrilov]

On Tue, Mar 31, 2026 at 7:38 PM Alex Deucher <alexdeucher@gmail.com> wrote:
>
> Applied.  Thanks!
>

Hi Christian, Alex,

While testing v7, I noticed that xa_erase() and xa_alloc_cyclic()
use plain xa_lock()/xa_unlock() regardless of XA_FLAGS_LOCK_IRQ —
the flag only affects lockdep annotations, not runtime locking.

The XArray API provides separate _irq variants for this:
xa_alloc_cyclic_irq() and xa_erase_irq(), both defined as inlines
in include/linux/xarray.h using xa_lock_irq/xa_unlock_irq.

I confirmed this by hitting the same lockdep WARNING with v6
(which has the same locking), and by reading lib/xarray.c:

  void *xa_erase(struct xarray *xa, unsigned long index)
  {
      xa_lock(xa);              // plain spin_lock, not _irq
      entry = __xa_erase(xa, index);
      xa_unlock(xa);
      return entry;
  }

Should I send a v8 using xa_alloc_cyclic_irq() and xa_erase_irq()?

Thanks,
Mikhail

Claude review: drm/amdgpu: replace PASID IDR with XArray

[Claude Code Review Bot]

Overall Series Review

Subject: drm/amdgpu: replace PASID IDR with XArray
Author: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
Patches: 4
Reviewed: 2026-04-01T07:43:13.815301

---

This is a single patch (v7) replacing the PASID IDR with XArray. The conversion is well-motivated — it addresses a real IRQ safety issue in `amdgpu_pasid_free()` which can be called from hardirq context. The code is clean and the approach is correct in principle. However, **the patch has a critical bug: it uses the wrong XArray API functions**.

The commit message claims that `XA_FLAGS_LOCK_IRQ` makes "all xa operations use IRQ-safe locking internally", but this is **incorrect**. The `XA_FLAGS_LOCK_IRQ` flag only affects the lock type stored in the xarray (so that `xa_lock()`/`xa_unlock()` resolve to the right primitive internally), but the convenience wrappers `xa_alloc_cyclic()` and `xa_erase()` use plain `xa_lock()`/`xa_unlock()` — **not** `xa_lock_irq()`/`xa_unlock_irq()`. The IRQ-safe convenience wrappers are separate functions: `xa_alloc_cyclic_irq()` and `xa_erase_irq()`.

This means the patch **does not actually fix the IRQ safety issue** it claims to fix. `amdgpu_pasid_free()` calling `xa_erase()` from hardirq context is still unsafe, exactly as the original `spin_lock()` was.

---
Generated by Claude Code Patch Reviewer

Claude review: drm/amdgpu: replace PASID IDR with XArray

[Claude Code Review Bot]

Patch Review

**Critical bug — wrong locking wrappers:**

```c
+	r = xa_alloc_cyclic(&amdgpu_pasid_xa, &pasid, xa_mk_value(0),
+			    XA_LIMIT(1, (1U << bits) - 1),
+			    &amdgpu_pasid_xa_next, GFP_KERNEL);
```

This must be `xa_alloc_cyclic_irq()` to disable interrupts while holding the xa_lock. The plain `xa_alloc_cyclic()` uses `xa_lock()`/`xa_unlock()` (see `include/linux/xarray.h:976-987`), which does not disable IRQs.

```c
+	xa_erase(&amdgpu_pasid_xa, pasid);
```

This must be `xa_erase_irq()`. The plain `xa_erase()` uses `xa_lock()`/`xa_unlock()` (see `lib/xarray.c:1665-1674`), which does not disable IRQs. Since `amdgpu_pasid_free()` can be called from hardirq context (via the fence signal path documented in the commit message), and `xa_alloc_cyclic()` in process context takes the same lock without disabling IRQs, the exact same deadlock scenario described in the commit message still applies.

**The fix:** Replace `xa_alloc_cyclic()` with `xa_alloc_cyclic_irq()` and `xa_erase()` with `xa_erase_irq()`. Alternatively, use explicit `xa_lock_irqsave()`/`__xa_erase()`/`xa_unlock_irqrestore()` in `amdgpu_pasid_free()` (since it's called from hardirq, `_irqsave` is the safest choice there), and `xa_alloc_cyclic_irq()` for the allocation path.

**Note:** The v5 changelog mentions that the author previously used explicit `xa_lock_irqsave`/`__xa_erase` for this exact reason, but was told to use the convenience wrappers in v6. The v6/v7 approach unfortunately reintroduces the problem.

**Minor nits (non-blocking):**

- The `XA_FLAGS_LOCK_IRQ` flag in `DEFINE_XARRAY_FLAGS` is not harmful, but it also isn't doing what the author thinks. It sets the lock class for lockdep — it doesn't make convenience wrappers use IRQ-safe locking. It's still correct to set it (it tells lockdep the lock is used in IRQ context), but the actual API calls must be the `_irq` variants.

- The `xa_mk_value(0)` store is fine — it stores a non-NULL value so the slot is occupied, preventing double-allocation of the same PASID.

- The `XA_FLAGS_ALLOC1` usage is correct (zero is not a valid PASID), though the `XA_LIMIT(1, ...)` already enforces the minimum of 1.

- The cleanup function change to `xa_destroy()` is correct and straightforward.

---
Generated by Claude Code Patch Reviewer