Intel-XE Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: Lucas De Marchi <lucas.demarchi@intel.com>
To: intel-xe@lists.freedesktop.org
Cc: Lucas De Marchi <lucas.demarchi@intel.com>,
	Shekhar Chauhan <shekhar.chauhan@intel.com>,
	Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>,
	Matt Roper <matthew.d.roper@intel.com>,
	Tejas Upadhyay <tejas.upadhyay@intel.com>
Subject: [PATCH 12/23] drm/xe/xe3p: Flush userptr/shrinker bo cachelines manually
Date: Mon, 13 Oct 2025 20:24:44 -0700	[thread overview]
Message-ID: <20251013-xe3p-v1-12-bfb74f038215@intel.com> (raw)
In-Reply-To: <20251013-xe3p-v1-0-bfb74f038215@intel.com>

From: Tejas Upadhyay <tejas.upadhyay@intel.com>

Starting with Xe3p, HW will flush cachelines marked with XA only when
media is off. We have few cases where kernel will have non-XA cachelines
which needs manual flush as we postpone the invalidation.

Flush asap from correctness POV to ensure non accelerated CPU copy to
swap/shmem file will see coherent view of memory, but also from security
POV where later flush can't corrupt the next user of those pages.

Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
[ TODO: xe_device_needs_cache_flush() seems a bad name that doesn't
  really review the context - it may need to be renamed/localized ]
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
---
 drivers/gpu/drm/xe/xe_bo.c      |  3 ++-
 drivers/gpu/drm/xe/xe_device.c  | 20 ++++++++++++++++++++
 drivers/gpu/drm/xe/xe_device.h  |  1 +
 drivers/gpu/drm/xe/xe_userptr.c |  3 ++-
 4 files changed, 25 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index 7b65020818738..05bc61d9e37cf 100644
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -673,7 +673,8 @@ static int xe_bo_trigger_rebind(struct xe_device *xe, struct xe_bo *bo,
 
 		if (!xe_vm_in_fault_mode(vm)) {
 			drm_gpuvm_bo_evict(vm_bo, true);
-			continue;
+			if (!xe_device_needs_cache_flush(xe))
+				continue;
 		}
 
 		if (!idle) {
diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
index 7efa8da9e1069..168a45fe36838 100644
--- a/drivers/gpu/drm/xe/xe_device.c
+++ b/drivers/gpu/drm/xe/xe_device.c
@@ -1081,6 +1081,26 @@ void xe_device_l2_flush(struct xe_device *xe)
 	xe_force_wake_put(gt_to_fw(gt), fw_ref);
 }
 
+/**
+ * xe_device_needs_cache_flush - Whether the cache needs to be flushed
+ * @xe: The device to check.
+ *
+ * Return: true if the device needs cache flush, false otherwise.
+ */
+bool xe_device_needs_cache_flush(struct xe_device *xe)
+{
+	/*
+	 * Xe3p will flush cachelines marked with XA only when media is off. We
+	 * have few cases where kernel will have non-XA cachelines which needs
+	 * manual flush and this is one of them as we postpone the
+	 * invalidation. Flush asap from correctness POV to ensure non
+	 * accelerated CPU copy to swap/shmem file will see coherent view of
+	 * memory, but also from security POV where later flush can't corrupt
+	 * the next user of those pages.
+	 */
+	return GRAPHICS_VER(xe) >= 35 && !IS_DGFX(xe);
+}
+
 /**
  * xe_device_td_flush() - Flush transient L3 cache entries
  * @xe: The device
diff --git a/drivers/gpu/drm/xe/xe_device.h b/drivers/gpu/drm/xe/xe_device.h
index 32cc6323b7f64..15e67db44b56c 100644
--- a/drivers/gpu/drm/xe/xe_device.h
+++ b/drivers/gpu/drm/xe/xe_device.h
@@ -179,6 +179,7 @@ void xe_device_snapshot_print(struct xe_device *xe, struct drm_printer *p);
 u64 xe_device_canonicalize_addr(struct xe_device *xe, u64 address);
 u64 xe_device_uncanonicalize_addr(struct xe_device *xe, u64 address);
 
+bool xe_device_needs_cache_flush(struct xe_device *xe);
 void xe_device_td_flush(struct xe_device *xe);
 void xe_device_l2_flush(struct xe_device *xe);
 
diff --git a/drivers/gpu/drm/xe/xe_userptr.c b/drivers/gpu/drm/xe/xe_userptr.c
index f16e92cd80904..86ce1c3ef41aa 100644
--- a/drivers/gpu/drm/xe/xe_userptr.c
+++ b/drivers/gpu/drm/xe/xe_userptr.c
@@ -112,7 +112,8 @@ static void __vma_userptr_invalidate(struct xe_vm *vm, struct xe_userptr_vma *uv
 				    false, MAX_SCHEDULE_TIMEOUT);
 	XE_WARN_ON(err <= 0);
 
-	if (xe_vm_in_fault_mode(vm) && userptr->initial_bind) {
+	if ((xe_vm_in_fault_mode(vm) || xe_device_needs_cache_flush(vm->xe)) &&
+	    userptr->initial_bind) {
 		err = xe_vm_invalidate_vma(vma);
 		XE_WARN_ON(err);
 	}

-- 
2.51.0


  parent reply	other threads:[~2025-10-14  3:25 UTC|newest]

Thread overview: 78+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-10-14  3:24 [PATCH 00/23] drm/xe: Add Xe3p support Lucas De Marchi
2025-10-14  3:24 ` [PATCH 01/23] drm/xe/xe3: Add support for graphics IP versions 30.04 & 30.05 Lucas De Marchi
2025-10-14  6:17   ` Shekhar Chauhan
2025-10-14 16:11   ` Matt Roper
2025-10-14  3:24 ` [PATCH 02/23] drm/xe/xe3p_lpm: Add support for media IP versions 35.00 & 35.03 Lucas De Marchi
2025-10-14  6:22   ` Shekhar Chauhan
2025-10-14 16:14   ` Matt Roper
2025-10-14  3:24 ` [PATCH 03/23] drm/xe/xe3p_lpm: Configure MAIN_GAMCTRL_QUEUE_SELECT Lucas De Marchi
2025-10-14 16:34   ` Matt Roper
2025-10-15  2:28     ` Lucas De Marchi
2025-10-15  6:36       ` Vivekanandan, Balasubramani
2025-10-15 14:03         ` Lucas De Marchi
2025-10-16 14:20           ` Vivekanandan, Balasubramani
2025-10-15 14:59       ` Matt Roper
2025-10-14  3:24 ` [PATCH 04/23] drm/xe/xe3p_lpm: Add special check in Media GT for Main GAMCTRL Lucas De Marchi
2025-10-14 16:36   ` Matt Roper
2025-10-14  3:24 ` [PATCH 05/23] drm/xe/xe3p_lpm: Stop reading the CTC_MODE register Lucas De Marchi
2025-10-14 11:58   ` Shekhar Chauhan
2025-10-14 16:52     ` Matt Roper
2025-10-15  3:41       ` Lucas De Marchi
2025-10-15  9:19         ` Vivekanandan, Balasubramani
2025-10-15 15:04           ` Matt Roper
2025-10-14 16:40   ` Matt Roper
2025-10-14 16:53     ` Matt Roper
2025-10-14  3:24 ` [PATCH 06/23] drm/xe/xe3p_lpm: Skip disabling NOA on unsupported IPs Lucas De Marchi
2025-10-14 17:04   ` Matt Roper
2025-10-14  3:24 ` [PATCH 07/23] drm/xe/xe3p_lpm: Handle MCR steering Lucas De Marchi
2025-10-15  9:56   ` Vivekanandan, Balasubramani
2025-10-14  3:24 ` [PATCH 08/23] drm/xe/xe3p: Stop programming RCU_MODE's fixed slice mode setting Lucas De Marchi
2025-10-15 12:24   ` Vivekanandan, Balasubramani
2025-10-14  3:24 ` [PATCH 09/23] drm/xe/xe3p: Determine service copy availability from fuse Lucas De Marchi
2025-10-15 20:14   ` Gustavo Sousa
2025-10-14  3:24 ` [PATCH 10/23] drm/xe/xe3p: Skip TD flush Lucas De Marchi
2025-10-14 19:35   ` Matt Roper
2025-10-14  3:24 ` [PATCH 11/23] drm/xe/xe3p: Enable L2 flush optimization feature Lucas De Marchi
2025-10-14 19:43   ` Matt Roper
2025-10-15  4:02     ` Lucas De Marchi
2025-10-14  3:24 ` Lucas De Marchi [this message]
2025-10-14 12:58   ` [PATCH 12/23] drm/xe/xe3p: Flush userptr/shrinker bo cachelines manually Thomas Hellström
2025-10-15 18:42     ` Lucas De Marchi
2025-10-14  3:24 ` [PATCH 13/23] drm/xe: Dump CURRENT_LRCA and CSMQDEBUG registers Lucas De Marchi
2025-10-14 17:24   ` Matt Roper
2025-10-15  4:07     ` Lucas De Marchi
2025-10-14  3:24 ` [PATCH 14/23] drm/xe/nvl: Define NVL-S platform Lucas De Marchi
2025-10-14  7:34   ` Shekhar Chauhan
2025-10-14  3:24 ` [PATCH 15/23] drm/xe/nvls: Define GuC firmware for NVL-S Lucas De Marchi
2025-10-15 18:49   ` Lucas De Marchi
2025-10-14  3:24 ` [PATCH 16/23] drm/xe/nvls: Attach MOCS table " Lucas De Marchi
2025-10-14  7:45   ` Shekhar Chauhan
2025-10-14  3:24 ` [PATCH 17/23] drm/xe/xe3p_xpc: Add Xe3p_XPC IP definition Lucas De Marchi
2025-10-14  8:04   ` Shekhar Chauhan
2025-10-14  8:12     ` Shekhar Chauhan
2025-10-14 17:33     ` Matt Roper
2025-10-15  2:47       ` Shekhar Chauhan
2025-10-14 17:36   ` Matt Roper
2025-10-14  3:24 ` [PATCH 18/23] drm/xe/xe3p_xpc: Add L3 bank mask Lucas De Marchi
2025-10-14 17:46   ` Matt Roper
2025-10-14  3:24 ` [PATCH 19/23] drm/xe/xe3p_xpc: Add MCR steering Lucas De Marchi
2025-10-14  3:24 ` [PATCH 20/23] drm/xe/xe3p_xpc: Add support for compute walker for non-MSIx Lucas De Marchi
2025-10-14 18:07   ` Matt Roper
2025-10-15 17:07     ` Lucas De Marchi
2025-10-15 17:12       ` Matt Roper
2025-10-14  3:24 ` [PATCH 21/23] drm/xe/xe3p_xpc: Skip compression tuning on platforms without flatccs Lucas De Marchi
2025-10-14  8:09   ` Shekhar Chauhan
2025-10-14  8:13     ` Shekhar Chauhan
2025-10-14 18:14     ` Matt Roper
2025-10-15  2:52       ` Shekhar Chauhan
2025-10-15 14:38         ` Lucas De Marchi
2025-10-15 15:34           ` Shekhar Chauhan
2025-10-14  3:24 ` [PATCH 22/23] drm/xe/xe3p_xpc: Setup PAT table Lucas De Marchi
2025-10-16 14:07   ` Vivekanandan, Balasubramani
2025-10-14  3:24 ` [PATCH 23/23] drm/xe/xe3p: Add xe3p EU stall data format Lucas De Marchi
2025-10-14 19:11   ` Matt Roper
2025-10-15 23:05     ` Harish Chegondi
2025-10-14  3:45 ` ✗ CI.checkpatch: warning for drm/xe: Add Xe3p support Patchwork
2025-10-14  3:47 ` ✓ CI.KUnit: success " Patchwork
2025-10-14  4:22 ` ✓ Xe.CI.BAT: " Patchwork
2025-10-14 12:06 ` ✓ Xe.CI.Full: " Patchwork

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=20251013-xe3p-v1-12-bfb74f038215@intel.com \
    --to=lucas.demarchi@intel.com \
    --cc=balasubramani.vivekanandan@intel.com \
    --cc=intel-xe@lists.freedesktop.org \
    --cc=matthew.d.roper@intel.com \
    --cc=shekhar.chauhan@intel.com \
    --cc=tejas.upadhyay@intel.com \
    /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