From: Dave Airlie <airlied@gmail.com>
To: dri-devel@lists.freedesktop.org
Cc: nouveau@lists.freedesktop.org, dakr@kernel.org
Subject: [PATCH 2/2] nouveau: Membar before between semaphore writes and the interrupt
Date: Fri, 29 Aug 2025 12:16:33 +1000 [thread overview]
Message-ID: <20250829021633.1674524-2-airlied@gmail.com> (raw)
In-Reply-To: <20250829021633.1674524-1-airlied@gmail.com>
From: Faith Ekstrand <faith.ekstrand@collabora.com>
This ensures that the memory write and the interrupt are properly
ordered and we won't wake up the kernel before the semaphore write has
hit memory.
Fixes: b1ca384772b6 ("drm/nouveau/gv100-: switch to volta semaphore methods")
Cc: stable@vger.kernel.org
Signed-off-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
---
drivers/gpu/drm/nouveau/gv100_fence.c | 7 +-
.../drm/nouveau/include/nvhw/class/clc36f.h | 85 +++++++++++++++++++
2 files changed, 91 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/nouveau/gv100_fence.c b/drivers/gpu/drm/nouveau/gv100_fence.c
index cccdeca72002..317e516c4ec7 100644
--- a/drivers/gpu/drm/nouveau/gv100_fence.c
+++ b/drivers/gpu/drm/nouveau/gv100_fence.c
@@ -18,7 +18,7 @@ gv100_fence_emit32(struct nouveau_channel *chan, u64 virtual, u32 sequence)
struct nvif_push *push = &chan->chan.push;
int ret;
- ret = PUSH_WAIT(push, 8);
+ ret = PUSH_WAIT(push, 13);
if (ret)
return ret;
@@ -32,6 +32,11 @@ gv100_fence_emit32(struct nouveau_channel *chan, u64 virtual, u32 sequence)
NVDEF(NVC36F, SEM_EXECUTE, PAYLOAD_SIZE, 32BIT) |
NVDEF(NVC36F, SEM_EXECUTE, RELEASE_TIMESTAMP, DIS));
+ PUSH_MTHD(push, NVC36F, MEM_OP_A, 0,
+ MEM_OP_B, 0,
+ MEM_OP_C, NVDEF(NVC36F, MEM_OP_C, MEMBAR_TYPE, SYS_MEMBAR),
+ MEM_OP_D, NVDEF(NVC36F, MEM_OP_D, OPERATION, MEMBAR));
+
PUSH_MTHD(push, NVC36F, NON_STALL_INTERRUPT, 0);
PUSH_KICK(push);
diff --git a/drivers/gpu/drm/nouveau/include/nvhw/class/clc36f.h b/drivers/gpu/drm/nouveau/include/nvhw/class/clc36f.h
index 8735dda4c8a7..338f74b9f501 100644
--- a/drivers/gpu/drm/nouveau/include/nvhw/class/clc36f.h
+++ b/drivers/gpu/drm/nouveau/include/nvhw/class/clc36f.h
@@ -7,6 +7,91 @@
#define NVC36F_NON_STALL_INTERRUPT (0x00000020)
#define NVC36F_NON_STALL_INTERRUPT_HANDLE 31:0
+// NOTE - MEM_OP_A and MEM_OP_B have been replaced in gp100 with methods for
+// specifying the page address for a targeted TLB invalidate and the uTLB for
+// a targeted REPLAY_CANCEL for UVM.
+// The previous MEM_OP_A/B functionality is in MEM_OP_C/D, with slightly
+// rearranged fields.
+#define NVC36F_MEM_OP_A (0x00000028)
+#define NVC36F_MEM_OP_A_TLB_INVALIDATE_CANCEL_TARGET_CLIENT_UNIT_ID 5:0 // only relevant for REPLAY_CANCEL_TARGETED
+#define NVC36F_MEM_OP_A_TLB_INVALIDATE_INVALIDATION_SIZE 5:0 // Used to specify size of invalidate, used for invalidates which are not of the REPLAY_CANCEL_TARGETED type
+#define NVC36F_MEM_OP_A_TLB_INVALIDATE_CANCEL_TARGET_GPC_ID 10:6 // only relevant for REPLAY_CANCEL_TARGETED
+#define NVC36F_MEM_OP_A_TLB_INVALIDATE_CANCEL_MMU_ENGINE_ID 6:0 // only relevant for REPLAY_CANCEL_VA_GLOBAL
+#define NVC36F_MEM_OP_A_TLB_INVALIDATE_SYSMEMBAR 11:11
+#define NVC36F_MEM_OP_A_TLB_INVALIDATE_SYSMEMBAR_EN 0x00000001
+#define NVC36F_MEM_OP_A_TLB_INVALIDATE_SYSMEMBAR_DIS 0x00000000
+#define NVC36F_MEM_OP_A_TLB_INVALIDATE_TARGET_ADDR_LO 31:12
+#define NVC36F_MEM_OP_B (0x0000002c)
+#define NVC36F_MEM_OP_B_TLB_INVALIDATE_TARGET_ADDR_HI 31:0
+#define NVC36F_MEM_OP_C (0x00000030)
+#define NVC36F_MEM_OP_C_MEMBAR_TYPE 2:0
+#define NVC36F_MEM_OP_C_MEMBAR_TYPE_SYS_MEMBAR 0x00000000
+#define NVC36F_MEM_OP_C_MEMBAR_TYPE_MEMBAR 0x00000001
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB 0:0
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_ONE 0x00000000
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_ALL 0x00000001 // Probably nonsensical for MMU_TLB_INVALIDATE_TARGETED
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_GPC 1:1
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_GPC_ENABLE 0x00000000
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_GPC_DISABLE 0x00000001
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY 4:2 // only relevant if GPC ENABLE
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_NONE 0x00000000
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_START 0x00000001
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_START_ACK_ALL 0x00000002
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_CANCEL_TARGETED 0x00000003
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_CANCEL_GLOBAL 0x00000004
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_CANCEL_VA_GLOBAL 0x00000005
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACK_TYPE 6:5 // only relevant if GPC ENABLE
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACK_TYPE_NONE 0x00000000
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACK_TYPE_GLOBALLY 0x00000001
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACK_TYPE_INTRANODE 0x00000002
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE 9:7 //only relevant for REPLAY_CANCEL_VA_GLOBAL
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_READ 0
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_WRITE 1
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_ATOMIC_STRONG 2
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_RSVRVD 3
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_ATOMIC_WEAK 4
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_ATOMIC_ALL 5
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_WRITE_AND_ATOMIC 6
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_ALL 7
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL 9:7 // Invalidate affects this level and all below
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_ALL 0x00000000 // Invalidate tlb caches at all levels of the page table
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_PTE_ONLY 0x00000001
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE0 0x00000002
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE1 0x00000003
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE2 0x00000004
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE3 0x00000005
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE4 0x00000006
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE5 0x00000007
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_APERTURE 11:10 // only relevant if PDB_ONE
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_APERTURE_VID_MEM 0x00000000
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_APERTURE_SYS_MEM_COHERENT 0x00000002
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_APERTURE_SYS_MEM_NONCOHERENT 0x00000003
+#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_ADDR_LO 31:12 // only relevant if PDB_ONE
+#define NVC36F_MEM_OP_C_ACCESS_COUNTER_CLR_TARGETED_NOTIFY_TAG 19:0
+// MEM_OP_D MUST be preceded by MEM_OPs A-C.
+#define NVC36F_MEM_OP_D (0x00000034)
+#define NVC36F_MEM_OP_D_TLB_INVALIDATE_PDB_ADDR_HI 26:0 // only relevant if PDB_ONE
+#define NVC36F_MEM_OP_D_OPERATION 31:27
+#define NVC36F_MEM_OP_D_OPERATION_MEMBAR 0x00000005
+#define NVC36F_MEM_OP_D_OPERATION_MMU_TLB_INVALIDATE 0x00000009
+#define NVC36F_MEM_OP_D_OPERATION_MMU_TLB_INVALIDATE_TARGETED 0x0000000a
+#define NVC36F_MEM_OP_D_OPERATION_L2_PEERMEM_INVALIDATE 0x0000000d
+#define NVC36F_MEM_OP_D_OPERATION_L2_SYSMEM_INVALIDATE 0x0000000e
+// CLEAN_LINES is an alias for Tegra/GPU IP usage
+#define NVC36F_MEM_OP_B_OPERATION_L2_INVALIDATE_CLEAN_LINES 0x0000000e
+#define NVC36F_MEM_OP_D_OPERATION_L2_CLEAN_COMPTAGS 0x0000000f
+#define NVC36F_MEM_OP_D_OPERATION_L2_FLUSH_DIRTY 0x00000010
+#define NVC36F_MEM_OP_D_OPERATION_L2_WAIT_FOR_SYS_PENDING_READS 0x00000015
+#define NVC36F_MEM_OP_D_OPERATION_ACCESS_COUNTER_CLR 0x00000016
+#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TYPE 1:0
+#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TYPE_MIMC 0x00000000
+#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TYPE_MOMC 0x00000001
+#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TYPE_ALL 0x00000002
+#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TYPE_TARGETED 0x00000003
+#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TARGETED_TYPE 2:2
+#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TARGETED_TYPE_MIMC 0x00000000
+#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TARGETED_TYPE_MOMC 0x00000001
+#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TARGETED_BANK 6:3
#define NVC36F_SEM_ADDR_LO (0x0000005c)
#define NVC36F_SEM_ADDR_LO_OFFSET 31:2
#define NVC36F_SEM_ADDR_HI (0x00000060)
--
2.50.1
next prev parent reply other threads:[~2025-08-29 2:23 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-08-29 2:16 [PATCH 1/2] nouveau: fix disabling the nonstall irq due to storm code. (v2) Dave Airlie
2025-08-29 2:16 ` Dave Airlie [this message]
2025-08-29 18:21 ` [PATCH 2/2] nouveau: Membar before between semaphore writes and the interrupt Danilo Krummrich
2025-08-29 23:09 ` James Jones
2025-08-29 14:49 ` [PATCH 1/2] nouveau: fix disabling the nonstall irq due to storm code. (v2) Faith Ekstrand
2025-08-29 18:21 ` Danilo Krummrich
2025-09-01 18:17 ` M Henning
2025-09-01 20:52 ` Dave Airlie
-- strict thread matches above, loose matches on Subject: below --
2025-08-28 20:55 [PATCH 1/2] nouveau: fix disabling the nonstall irq due to storm code Dave Airlie
2025-08-28 20:55 ` [PATCH 2/2] nouveau: Membar before between semaphore writes and the interrupt Dave Airlie
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20250829021633.1674524-2-airlied@gmail.com \
--to=airlied@gmail.com \
--cc=dakr@kernel.org \
--cc=dri-devel@lists.freedesktop.org \
--cc=nouveau@lists.freedesktop.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).