From: Matthew Brost <matthew.brost@intel.com>
To: <fei.yang@intel.com>
Cc: <intel-xe@lists.freedesktop.org>,
Stuart Summers <stuart.summers@intel.com>
Subject: Re: [PATCH] drm/xe: Wait for HW clearance before issuing the next TLB inval.
Date: Mon, 6 Apr 2026 11:59:18 -0700 [thread overview]
Message-ID: <adQChk1p9hRnvQxl@gsse-cloud1.jf.intel.com> (raw)
In-Reply-To: <20260404062250.3435419-1-fei.yang@intel.com>
On Fri, Apr 03, 2026 at 11:22:49PM -0700, fei.yang@intel.com wrote:
> From: Fei Yang <fei.yang@intel.com>
>
> Hardware requires the software to poll the valid bit and make sure it's
> cleared before issuing a new TLB invalidation request.
> We also need to avoid racing against GuC on TLB invalidations by only
> allowing KMD to issue TLB invalidation request during GuC reset.
>
> v2: separate ggtt inval to xe_gam_port and call in guc_reset only (Matt)
>
I guess CI doesn't like this... Unsure why, clearly missed something in
my suggestion.
Maybe it is best move this new code into xe_guc_tlb_inval.c for TLB now
unless someone really has time to dig in.
Matt
> Signed-off-by: Fei Yang <fei.yang@intel.com>
> Suggested-by: Matthew Brost <matthew.brost@intel.com>
> Cc: Stuart Summers <stuart.summers@intel.com>
> ---
> drivers/gpu/drm/xe/Makefile | 1 +
> drivers/gpu/drm/xe/xe_gam_port.c | 72 +++++++++++++++++++++++++++
> drivers/gpu/drm/xe/xe_gam_port.h | 15 ++++++
> drivers/gpu/drm/xe/xe_gt.c | 1 +
> drivers/gpu/drm/xe/xe_gt_types.h | 7 +++
> drivers/gpu/drm/xe/xe_guc.c | 4 ++
> drivers/gpu/drm/xe/xe_guc_tlb_inval.c | 18 -------
> 7 files changed, 100 insertions(+), 18 deletions(-)
> create mode 100644 drivers/gpu/drm/xe/xe_gam_port.c
> create mode 100644 drivers/gpu/drm/xe/xe_gam_port.h
>
> diff --git a/drivers/gpu/drm/xe/Makefile b/drivers/gpu/drm/xe/Makefile
> index f9abaf687d46..96127ad60e9b 100644
> --- a/drivers/gpu/drm/xe/Makefile
> +++ b/drivers/gpu/drm/xe/Makefile
> @@ -78,6 +78,7 @@ xe-y += xe_bb.o \
> xe_guc_rc.o \
> xe_guc_submit.o \
> xe_guc_tlb_inval.o \
> + xe_gam_port.o \
> xe_heci_gsc.o \
> xe_huc.o \
> xe_hw_engine.o \
> diff --git a/drivers/gpu/drm/xe/xe_gam_port.c b/drivers/gpu/drm/xe/xe_gam_port.c
> new file mode 100644
> index 000000000000..137887cf93b8
> --- /dev/null
> +++ b/drivers/gpu/drm/xe/xe_gam_port.c
> @@ -0,0 +1,72 @@
> +// SPDX-License-Identifier: MIT
> +/*
> + * Copyright © 2025 Intel Corporation
> + */
> +
> +#include "xe_gt_stats.h"
> +#include "xe_gt_types.h"
> +#include "xe_gt_printk.h"
> +#include "xe_mmio.h"
> +#include "xe_gam_port.h"
> +
> +#include "regs/xe_guc_regs.h"
> +
> +/*
> + * GGTT TLB invalidation as part of GuC reset flow while the communication
> + * between host and GuC is disabled.
> + */
> +
> +int xe_gam_port_tlb_inval_ggtt(struct xe_gt *gt)
> +{
> + struct xe_device *xe = gt_to_xe(gt);
> + struct xe_mmio *mmio = >->mmio;
> + int ret = 0;
> +
> + spin_lock(>->mmio_tlb_invl_lock);
> +
> + if (xe->info.platform == XE_PVC || GRAPHICS_VER(xe) >= 20) {
> + /* Wait 1-second for the valid bit to be cleared */
> + ret = xe_mmio_wait32(mmio, PVC_GUC_TLB_INV_DESC0, PVC_GUC_TLB_INV_DESC0_VALID,
> + 0, 1000 * USEC_PER_MSEC, NULL, true);
> + if (ret) {
> + ret = -ECANCELED;
> + goto out;
> + }
> +
> + xe_mmio_write32(mmio, PVC_GUC_TLB_INV_DESC1, PVC_GUC_TLB_INV_DESC1_INVALIDATE);
> + xe_mmio_write32(mmio, PVC_GUC_TLB_INV_DESC0, PVC_GUC_TLB_INV_DESC0_VALID);
> +
> + /* Wait 1-second for the valid bit to be cleared */
> + ret = xe_mmio_wait32(mmio, PVC_GUC_TLB_INV_DESC0, PVC_GUC_TLB_INV_DESC0_VALID,
> + 0, 1000 * USEC_PER_MSEC, NULL, true);
> + if (ret) {
> + ret = -ECANCELED;
> + goto out;
> + }
> + } else {
> + /* Wait 1-second for the valid bit to be cleared */
> + ret = xe_mmio_wait32(mmio, GUC_TLB_INV_CR, GUC_TLB_INV_CR_INVALIDATE,
> + 0, 1000 * USEC_PER_MSEC, NULL, true);
> + if (ret) {
> + ret = -ECANCELED;
> + goto out;
> + }
> +
> + xe_mmio_write32(mmio, GUC_TLB_INV_CR, GUC_TLB_INV_CR_INVALIDATE);
> +
> + /* Wait 1-second for the valid bit to be cleared */
> + ret = xe_mmio_wait32(mmio, GUC_TLB_INV_CR, GUC_TLB_INV_CR_INVALIDATE,
> + 0, 1000 * USEC_PER_MSEC, NULL, true);
> + if (ret) {
> + ret = -ECANCELED;
> + goto out;
> + }
> + }
> +
> +out:
> + spin_unlock(>->mmio_tlb_invl_lock);
> +
> + if (ret)
> + xe_gt_warn(gt, "TLB INVAL cancelled due to uncleared valid bit\n");
> + return ret;
> +}
> diff --git a/drivers/gpu/drm/xe/xe_gam_port.h b/drivers/gpu/drm/xe/xe_gam_port.h
> new file mode 100644
> index 000000000000..b5c383d635f1
> --- /dev/null
> +++ b/drivers/gpu/drm/xe/xe_gam_port.h
> @@ -0,0 +1,15 @@
> +/* SPDX-License-Identifier: MIT */
> +/*
> + * Copyright © 2025 Intel Corporation
> + */
> +
> +#ifndef _XE_GAM_PORT_H_
> +#define _XE_GAM_PORT_H_
> +
> +#include <linux/types.h>
> +
> +struct xe_gt;
> +
> +int xe_gam_port_tlb_inval_ggtt(struct xe_gt *gt);
> +
> +#endif
> diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
> index 8a31c963c372..6bb990069eff 100644
> --- a/drivers/gpu/drm/xe/xe_gt.c
> +++ b/drivers/gpu/drm/xe/xe_gt.c
> @@ -514,6 +514,7 @@ int xe_gt_init_early(struct xe_gt *gt)
>
> xe_force_wake_init_gt(gt, gt_to_fw(gt));
> spin_lock_init(>->global_invl_lock);
> + spin_lock_init(>->mmio_tlb_invl_lock);
>
> err = xe_gt_tlb_inval_init_early(gt);
> if (err)
> diff --git a/drivers/gpu/drm/xe/xe_gt_types.h b/drivers/gpu/drm/xe/xe_gt_types.h
> index 8b55cf25a75f..2418a0b2f19c 100644
> --- a/drivers/gpu/drm/xe/xe_gt_types.h
> +++ b/drivers/gpu/drm/xe/xe_gt_types.h
> @@ -324,6 +324,13 @@ struct xe_gt {
> */
> spinlock_t global_invl_lock;
>
> + /**
> + * @mmio_tlb_invl_lock: prevents back to back TLB invalidation
> + * without polling for hardware clearance for the previous
> + * invalidation
> + */
> + spinlock_t mmio_tlb_invl_lock;
> +
> /** @wa_active: keep track of active workarounds */
> struct {
> /** @wa_active.gt: bitmap with active GT workarounds */
> diff --git a/drivers/gpu/drm/xe/xe_guc.c b/drivers/gpu/drm/xe/xe_guc.c
> index e762eada21db..70eff5b22fbb 100644
> --- a/drivers/gpu/drm/xe/xe_guc.c
> +++ b/drivers/gpu/drm/xe/xe_guc.c
> @@ -48,6 +48,7 @@
> #include "xe_uc_fw.h"
> #include "xe_wa.h"
> #include "xe_wopcm.h"
> +#include "xe_gam_port.h"
>
> static u32 guc_bo_ggtt_addr(struct xe_guc *guc,
> struct xe_bo *bo)
> @@ -991,6 +992,9 @@ int xe_guc_reset(struct xe_guc *guc)
> goto err_out;
> }
>
> + if (xe_gam_port_tlb_inval_ggtt(gt))
> + xe_gt_warn(gt, "GGTT TLB invalidation timed out\n");
> +
> return 0;
>
> err_out:
> diff --git a/drivers/gpu/drm/xe/xe_guc_tlb_inval.c b/drivers/gpu/drm/xe/xe_guc_tlb_inval.c
> index ced58f46f846..7c3e63e49760 100644
> --- a/drivers/gpu/drm/xe/xe_guc_tlb_inval.c
> +++ b/drivers/gpu/drm/xe/xe_guc_tlb_inval.c
> @@ -61,8 +61,6 @@ static int send_tlb_inval_all(struct xe_tlb_inval *tlb_inval, u32 seqno)
> static int send_tlb_inval_ggtt(struct xe_tlb_inval *tlb_inval, u32 seqno)
> {
> struct xe_guc *guc = tlb_inval->private;
> - struct xe_gt *gt = guc_to_gt(guc);
> - struct xe_device *xe = guc_to_xe(guc);
>
> /*
> * Returning -ECANCELED in this function is squashed at the caller and
> @@ -77,22 +75,6 @@ static int send_tlb_inval_ggtt(struct xe_tlb_inval *tlb_inval, u32 seqno)
> };
>
> return send_tlb_inval(guc, action, ARRAY_SIZE(action));
> - } else if (xe_device_uc_enabled(xe) && !xe_device_wedged(xe)) {
> - struct xe_mmio *mmio = >->mmio;
> -
> - if (IS_SRIOV_VF(xe))
> - return -ECANCELED;
> -
> - CLASS(xe_force_wake, fw_ref)(gt_to_fw(gt), XE_FW_GT);
> - if (xe->info.platform == XE_PVC || GRAPHICS_VER(xe) >= 20) {
> - xe_mmio_write32(mmio, PVC_GUC_TLB_INV_DESC1,
> - PVC_GUC_TLB_INV_DESC1_INVALIDATE);
> - xe_mmio_write32(mmio, PVC_GUC_TLB_INV_DESC0,
> - PVC_GUC_TLB_INV_DESC0_VALID);
> - } else {
> - xe_mmio_write32(mmio, GUC_TLB_INV_CR,
> - GUC_TLB_INV_CR_INVALIDATE);
> - }
> }
>
> return -ECANCELED;
> --
> 2.43.0
>
next prev parent reply other threads:[~2026-04-06 18:59 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-04-04 6:22 [PATCH] drm/xe: Wait for HW clearance before issuing the next TLB inval fei.yang
2026-04-04 6:23 ` ✗ CI.checkpatch: warning for drm/xe: Wait for HW clearance before issuing the next TLB inval. (rev2) Patchwork
2026-04-04 6:24 ` ✓ CI.KUnit: success " Patchwork
2026-04-04 7:12 ` ✗ Xe.CI.BAT: failure " Patchwork
2026-04-04 7:31 ` ✗ Xe.CI.FULL: " Patchwork
2026-04-06 18:59 ` Matthew Brost [this message]
2026-04-06 21:01 ` [PATCH] drm/xe: Wait for HW clearance before issuing the next TLB inval Matt Roper
-- strict thread matches above, loose matches on Subject: below --
2026-03-17 23:21 fei.yang
2026-03-17 23:28 ` Summers, Stuart
2026-03-22 5:35 ` Matthew Brost
2026-03-24 20:39 ` Yang, Fei
2026-03-24 20:53 ` Matthew Brost
2026-03-24 20:58 ` Matthew Brost
2026-03-24 21:10 ` Summers, Stuart
2026-03-24 23:36 ` Matthew Brost
2026-03-25 18:37 ` Summers, Stuart
2026-03-25 22:00 ` Matthew Brost
2026-03-25 22:25 ` Summers, Stuart
2026-03-25 22:38 ` Matthew Brost
2026-03-25 22:43 ` Summers, Stuart
2026-03-26 0:54 ` Matthew Brost
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=adQChk1p9hRnvQxl@gsse-cloud1.jf.intel.com \
--to=matthew.brost@intel.com \
--cc=fei.yang@intel.com \
--cc=intel-xe@lists.freedesktop.org \
--cc=stuart.summers@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