Skip to contentSkip to conversation

drm/amdkfd: Avoid integer underflow in EOP ring size calculation.

PATCH In review David Francis 2 participants 2 messages Aug 5 – Sep 15

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/amdkfd: Avoid integer underflow in EOP ring size calculation.

The low 6 bits of cp_hqd_eop_control store the base-2 logarithm of the EOP ring size. This was calculated as

order_base_2(q->eop_ring_buffer_size / 4) - 1

But order_base_2 can in theory return 0, so this could underflow (although in practice the ring buffer size cannot be less than 4096).

Change this to

order_base_2(q->eop_ring_buffer_size / 8)

using properties of logarithms.

Also add to the above comment to make the mathematics more clear.

---
 drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c | 6 +++++-
 drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c | 5 ++++-
 2 files changed, 9 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c +5 -1
index 75e5a9f67d50..2c0d501bf527 100644
@@ -285,6 +285,10 @@ static void update_mqd(struct mqd_manager *mm, void *mqd,
 		1 << CP_HQD_IB_CONTROL__IB_EXE_DISABLE__SHIFT;
 
 	/*
+	 * The lowest 6 bits of eop_control store the EOP ring size. If
+	 * their value is X, the ring size is 2^(X + 1) dwords, or
+	 * 2^(X + 3) bytes.
+	 *
 	 * HW does not clamp this field correctly. Maximum EOP queue size
 	 * is constrained by per-SE EOP done signal count, which is 8-bit.
 	 * Limit is 0xFF EOP entries (= 0x7F8 dwords). CP will not submit
@@ -296,7 +300,7 @@ static void update_mqd(struct mqd_manager *mm, void *mqd,
 	 *
 	 */
 	m->cp_hqd_eop_control = q->eop_ring_buffer_size ?
-		min(0xA, order_base_2(q->eop_ring_buffer_size / 4) - 1) : 0;
+		min(0xA, order_base_2(q->eop_ring_buffer_size / 8)) : 0;
 
 	m->cp_hqd_eop_base_addr_lo =
 			lower_32_bits(q->eop_ring_buffer_address >> 8);
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c +4 -1
index 60b87a500698..029572548c14 100644
@@ -208,6 +208,9 @@ static void __update_mqd(struct mqd_manager *mm, void *mqd,
 			mtype << CP_HQD_IB_CONTROL__MTYPE__SHIFT;
 
 	/*
+	 * The lowest 6 bits of eop_control store the EOP ring size. If
+	 * their value is X, the ring size is 2^(X + 1) dwords, or
+	 * 2^(X + 3) bytes.
 	 * HW does not clamp this field correctly. Maximum EOP queue size
 	 * is constrained by per-SE EOP done signal count, which is 8-bit.
 	 * Limit is 0xFF EOP entries (= 0x7F8 dwords). CP will not submit
@@ -215,7 +218,7 @@ static void __update_mqd(struct mqd_manager *mm, void *mqd,
 	 * is safe, giving a maximum field value of 0xA.
 	 */
 	m->cp_hqd_eop_control |= q->eop_ring_buffer_size ? min(0xA,
-		order_base_2(q->eop_ring_buffer_size / 4) - 1) : 0;
+		order_base_2(q->eop_ring_buffer_size / 8)) : 0;
 	m->cp_hqd_eop_base_addr_lo =
 			lower_32_bits(q->eop_ring_buffer_address >> 8);
 	m->cp_hqd_eop_base_addr_hi =
Signature · 1 line
2.34.1
Reviewed-by on v1 1/1
#1 lore raw

AMD General

Quoted 87 lines

-----Original Message-----
From: amd-gfx <[email protected]> On Behalf Of David Francis
Sent: August 5, 2026 9:32 AM
To: [email protected]
Cc: Deucher, Alexander <[email protected]>; Francis, David <[email protected]>
Subject: [PATCH] drm/amdkfd: Avoid integer underflow in EOP ring size calculation.

The low 6 bits of cp_hqd_eop_control store the base-2 logarithm of the EOP ring size. This was calculated as

order_base_2(q->eop_ring_buffer_size / 4) - 1

But order_base_2 can in theory return 0, so this could underflow (although in practice the ring buffer size cannot be less than 4096).

Change this to

order_base_2(q->eop_ring_buffer_size / 8)

using properties of logarithms.

Also add to the above comment to make the mathematics more clear.

---
 drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c | 6 +++++-
 drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c | 5 ++++-
 2 files changed, 9 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c +4 -0
index 75e5a9f67d50..2c0d501bf527 100644
@@ -285,6 +285,10 @@ static void update_mqd(struct mqd_manager *mm, void
 *mqd,
              1 << CP_HQD_IB_CONTROL__IB_EXE_DISABLE__SHIFT;
 
      /*
+      * The lowest 6 bits of eop_control store the EOP ring size. If
+      * their value is X, the ring size is 2^(X + 1) dwords, or
+      * 2^(X + 3) bytes.
+      *
       * HW does not clamp this field correctly. Maximum EOP queue size
       * is constrained by per-SE EOP done signal count, which is 8-bit.

* Limit is 0xFF EOP entries (= 0x7F8 dwords). CP will not submit

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c +1 -1
@@ -296,7 +300,7 @@ static void update_mqd(struct mqd_manager *mm, void
 *mqd,
       *
       */
      m->cp_hqd_eop_control = q->eop_ring_buffer_size ?
-             min(0xA, order_base_2(q->eop_ring_buffer_size / 4) - 1) : 0;
+             min(0xA, order_base_2(q->eop_ring_buffer_size / 8)) : 0;
 
      m->cp_hqd_eop_base_addr_lo =
                      lower_32_bits(q->eop_ring_buffer_address >> 8);
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c +3 -0
index 60b87a500698..029572548c14 100644
@@ -208,6 +208,9 @@ static void __update_mqd(struct mqd_manager *mm,
 void *mqd,
                      mtype << CP_HQD_IB_CONTROL__MTYPE__SHIFT;
 
      /*
+      * The lowest 6 bits of eop_control store the EOP ring size. If
+      * their value is X, the ring size is 2^(X + 1) dwords, or
+      * 2^(X + 3) bytes.
       * HW does not clamp this field correctly. Maximum EOP queue size
       * is constrained by per-SE EOP done signal count, which is 8-bit.

* Limit is 0xFF EOP entries (= 0x7F8 dwords). CP will not submit

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c +1 -1
@@ -215,7 +218,7 @@ static void __update_mqd(struct mqd_manager *mm,
 void *mqd,
       * is safe, giving a maximum field value of 0xA.
       */
      m->cp_hqd_eop_control |= q->eop_ring_buffer_size ? min(0xA,
-             order_base_2(q->eop_ring_buffer_size / 4) - 1) : 0;
+             order_base_2(q->eop_ring_buffer_size / 8)) : 0;
      m->cp_hqd_eop_base_addr_lo =
                      lower_32_bits(q->eop_ring_buffer_address >> 8);
      m->cp_hqd_eop_base_addr_hi =

--
2.34.1