From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 96400CF45D7 for ; Mon, 12 Jan 2026 23:27:40 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 5450710E439; Mon, 12 Jan 2026 23:27:40 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=intel.com header.i=@intel.com header.b="H+yUTlWQ"; dkim-atps=neutral Received: from mgamail.intel.com (mgamail.intel.com [192.198.163.14]) by gabe.freedesktop.org (Postfix) with ESMTPS id 2058110E438 for ; Mon, 12 Jan 2026 23:27:37 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1768260457; x=1799796457; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=cr+1NOWIpShnJYE2YaaKJSzUD8pk1VCpz2avesgqeIw=; b=H+yUTlWQOUXcnS+eA2tNYDWMhNvUixquvf4vHzRqaQB16W/DQN0EEn/T zEJOiufCtJyZv6hYs+ba9VTS2O3NbYahPovZoEK2jrNK0n+K+qkenGB/N 8AV80lL83iQvMOqKimF4wIdyndFsfkJjdF1TlLFp8aXumgMv8bd7W52nV LS3xxxjMPJrthNi5G3QmARhZI4eJvLJqqP29IXN57aPtYxXON9jxhiyRf CoF+GWQ9n41AHFXZ1IbwOEEp1P0Ttv6Y95LMC63X2fGJCQ4tls10MpoCr IH4BfdPxx+rtieqlJykya/nVw1CF3eXzG1fdXbfKcMBorEtj7vjzq8nh5 w==; X-CSE-ConnectionGUID: hUeoHsNDRNWzFKKO53fYOA== X-CSE-MsgGUID: 0BLp+6psRmOYU50Z6mGDFw== X-IronPort-AV: E=McAfee;i="6800,10657,11669"; a="69594773" X-IronPort-AV: E=Sophos;i="6.21,222,1763452800"; d="scan'208";a="69594773" Received: from fmviesa003.fm.intel.com ([10.60.135.143]) by fmvoesa108.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 12 Jan 2026 15:27:37 -0800 X-CSE-ConnectionGUID: JGi+AbP9TAihz3A0JE6uRw== X-CSE-MsgGUID: LXIAEuENSP6wNZJLGjnwTg== X-ExtLoop1: 1 Received: from lstrano-desk.jf.intel.com ([10.54.39.91]) by fmviesa003-auth.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 12 Jan 2026 15:27:36 -0800 From: Matthew Brost To: intel-xe@lists.freedesktop.org Cc: stuart.summers@intel.com Subject: [PATCH v3 10/11] drm/xe: Add context-based invalidation to GuC TLB invalidation backend Date: Mon, 12 Jan 2026 15:27:29 -0800 Message-Id: <20260112232730.3347414-11-matthew.brost@intel.com> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20260112232730.3347414-1-matthew.brost@intel.com> References: <20260112232730.3347414-1-matthew.brost@intel.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-BeenThere: intel-xe@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Intel Xe graphics driver List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: intel-xe-bounces@lists.freedesktop.org Sender: "Intel-xe" Introduce context-based invalidation support to the GuC TLB invalidation backend. This is implemented by iterating over each exec queue per GT within a VM, skipping inactive queues, and issuing a context-based (GuC ID) H2G TLB invalidation. All H2G messages, except the final one, are sent with an invalid seqno, which the G2H handler drops to ensure the TLB invalidation fence is only signaled once all H2G messages are completed. A watermark mechanism is also added to switch between context-based TLB invalidations and full device-wide invalidations, as the return on investment for context-based invalidation diminishes when many exec queues are mapped. v2: - Fix checkpatch warnings v3: - Rebase on PRL - Use ref counting to avoid racing with deregisters Signed-off-by: Matthew Brost --- drivers/gpu/drm/xe/xe_device_types.h | 2 + drivers/gpu/drm/xe/xe_guc_tlb_inval.c | 145 +++++++++++++++++++++++++- drivers/gpu/drm/xe/xe_pci.c | 1 + drivers/gpu/drm/xe/xe_pci_types.h | 1 + 4 files changed, 145 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h index 8db870aaa382..b51acff4edcd 100644 --- a/drivers/gpu/drm/xe/xe_device_types.h +++ b/drivers/gpu/drm/xe/xe_device_types.h @@ -358,6 +358,8 @@ struct xe_device { u8 has_pre_prod_wa:1; /** @info.has_pxp: Device has PXP support */ u8 has_pxp:1; + /** @info.has_ctx_tlb_inval: Has context based TLB invalidations */ + u8 has_ctx_tlb_inval:1; /** @info.has_range_tlb_inval: Has range based TLB invalidations */ u8 has_range_tlb_inval:1; /** @info.has_soc_remapper_sysctrl: Has SoC remapper system controller */ diff --git a/drivers/gpu/drm/xe/xe_guc_tlb_inval.c b/drivers/gpu/drm/xe/xe_guc_tlb_inval.c index 070d2e2cb7c9..328eced5f692 100644 --- a/drivers/gpu/drm/xe/xe_guc_tlb_inval.c +++ b/drivers/gpu/drm/xe/xe_guc_tlb_inval.c @@ -6,15 +6,19 @@ #include "abi/guc_actions_abi.h" #include "xe_device.h" +#include "xe_exec_queue.h" +#include "xe_exec_queue_types.h" #include "xe_gt_stats.h" #include "xe_gt_types.h" #include "xe_guc.h" #include "xe_guc_ct.h" +#include "xe_guc_exec_queue_types.h" #include "xe_guc_tlb_inval.h" #include "xe_force_wake.h" #include "xe_mmio.h" #include "xe_sa.h" #include "xe_tlb_inval.h" +#include "xe_vm.h" #include "regs/xe_guc_regs.h" @@ -156,10 +160,16 @@ static int send_tlb_inval_ppgtt(struct xe_guc *guc, u32 seqno, u64 start, { #define MAX_TLB_INVALIDATION_LEN 7 struct xe_gt *gt = guc_to_gt(guc); + struct xe_device *xe = guc_to_xe(guc); u32 action[MAX_TLB_INVALIDATION_LEN]; u64 length = end - start; int len = 0, err; + xe_gt_assert(gt, (type == XE_GUC_TLB_INVAL_PAGE_SELECTIVE && + !xe->info.has_ctx_tlb_inval) || + (type == XE_GUC_TLB_INVAL_PAGE_SELECTIVE_CTX && + xe->info.has_ctx_tlb_inval)); + action[len++] = XE_GUC_ACTION_TLB_INVALIDATION; action[len++] = !prl_sa ? seqno : TLB_INVALIDATION_SEQNO_INVALID; if (!gt_to_xe(gt)->info.has_range_tlb_inval || @@ -168,9 +178,11 @@ static int send_tlb_inval_ppgtt(struct xe_guc *guc, u32 seqno, u64 start, } else { u64 normalize_len = normalize_invalidation_range(gt, &start, &end); + bool need_flush = !prl_sa && + seqno != TLB_INVALIDATION_SEQNO_INVALID; /* Flush on NULL case, Media is not required to modify flush due to no PPC so NOP */ - action[len++] = MAKE_INVAL_OP_FLUSH(type, !prl_sa); + action[len++] = MAKE_INVAL_OP_FLUSH(type, need_flush); action[len++] = id; action[len++] = lower_32_bits(start); action[len++] = upper_32_bits(start); @@ -181,8 +193,10 @@ static int send_tlb_inval_ppgtt(struct xe_guc *guc, u32 seqno, u64 start, #undef MAX_TLB_INVALIDATION_LEN err = send_tlb_inval(guc, action, len); - if (!err && prl_sa) + if (!err && prl_sa) { + xe_gt_assert(gt, seqno != TLB_INVALIDATION_SEQNO_INVALID); err = send_page_reclaim(guc, seqno, xe_sa_bo_gpu_addr(prl_sa)); + } return err; } @@ -201,6 +215,114 @@ static int send_tlb_inval_asid_ppgtt(struct xe_tlb_inval *tlb_inval, u32 seqno, XE_GUC_TLB_INVAL_PAGE_SELECTIVE, prl_sa); } +static bool queue_mapped_in_guc(struct xe_guc *guc, struct xe_exec_queue *q) +{ + return q->gt == guc_to_gt(guc); +} + +static int send_tlb_inval_ctx_ppgtt(struct xe_tlb_inval *tlb_inval, u32 seqno, + u64 start, u64 end, u32 asid, + struct drm_suballoc *prl_sa) +{ + struct xe_guc *guc = tlb_inval->private; + struct xe_device *xe = guc_to_xe(guc); + struct xe_exec_queue *q, *next, *last_q = NULL; + struct xe_vm *vm; + LIST_HEAD(tlb_inval_list); + int err = 0; + + lockdep_assert_held(&tlb_inval->seqno_lock); + + if (xe->info.force_execlist) + return -ECANCELED; + + vm = xe_device_asid_to_vm(xe, asid); + if (IS_ERR(vm)) + return PTR_ERR(vm); + + down_read(&vm->exec_queues.lock); + + /* + * XXX: Randomly picking a threshold for now. This will need to be + * tuned based on expected UMD queue counts and performance profiling. + */ +#define EXEC_QUEUE_COUNT_FULL_THRESHOLD 8 + if (vm->exec_queues.count[guc_to_gt(guc)->info.id] >= + EXEC_QUEUE_COUNT_FULL_THRESHOLD) { + u32 action[] = { + XE_GUC_ACTION_TLB_INVALIDATION, + seqno, + MAKE_INVAL_OP(XE_GUC_TLB_INVAL_FULL), + }; + + err = send_tlb_inval(guc, action, ARRAY_SIZE(action)); + goto err_unlock; + } +#undef EXEC_QUEUE_COUNT_FULL_THRESHOLD + + /* + * Move exec queues to a temporary list to issue invalidations. The exec + * queue must be mapped in the current GuC, active, and a reference must + * be taken to prevent concurrent deregistrations. + */ + list_for_each_entry_safe(q, next, &vm->exec_queues.list, + vm_exec_queue_link) + if (queue_mapped_in_guc(guc, q) && q->ops->active(q) && + xe_exec_queue_get_unless_zero(q)) { + last_q = q; + list_move_tail(&q->vm_exec_queue_link, &tlb_inval_list); + } + + if (!last_q) { + /* + * We can't break fence ordering for TLB invalidation jobs, if + * TLB invalidations are inflight issue a dummy invalidation to + * maintain ordering. Nor can we move safely the seqno_recv when + * returning -ECANCELED if TLB invalidations are in flight. Use + * GGTT invalidation as dummy invalidation given ASID + * invalidations are unsupported here. + */ + if (xe_tlb_inval_idle(tlb_inval)) + err = -ECANCELED; + else + err = send_tlb_inval_ggtt(tlb_inval, seqno); + goto err_unlock; + } + + list_for_each_entry_safe(q, next, &tlb_inval_list, vm_exec_queue_link) { + struct drm_suballoc *__prl_sa = NULL; + int __seqno = TLB_INVALIDATION_SEQNO_INVALID; + u32 type = XE_GUC_TLB_INVAL_PAGE_SELECTIVE_CTX; + + xe_assert(xe, q->vm == vm); + + if (err) + goto unref; + + if (last_q == q) { + __prl_sa = prl_sa; + __seqno = seqno; + } + + err = send_tlb_inval_ppgtt(guc, __seqno, start, end, + q->guc->id, type, __prl_sa); + +unref: + /* + * Must always return exec queue to original list / drop + * reference + */ + xe_exec_queue_put(q); + list_move_tail(&q->vm_exec_queue_link, &vm->exec_queues.list); + } + +err_unlock: + up_read(&vm->exec_queues.lock); + xe_vm_put(vm); + + return err; +} + static bool tlb_inval_initialized(struct xe_tlb_inval *tlb_inval) { struct xe_guc *guc = tlb_inval->private; @@ -228,7 +350,7 @@ static long tlb_inval_timeout_delay(struct xe_tlb_inval *tlb_inval) return hw_tlb_timeout + 2 * delay; } -static const struct xe_tlb_inval_ops guc_tlb_inval_ops = { +static const struct xe_tlb_inval_ops guc_tlb_inval_asid_ops = { .all = send_tlb_inval_all, .ggtt = send_tlb_inval_ggtt, .ppgtt = send_tlb_inval_asid_ppgtt, @@ -237,6 +359,15 @@ static const struct xe_tlb_inval_ops guc_tlb_inval_ops = { .timeout_delay = tlb_inval_timeout_delay, }; +static const struct xe_tlb_inval_ops guc_tlb_inval_ctx_ops = { + .ggtt = send_tlb_inval_ggtt, + .all = send_tlb_inval_all, + .ppgtt = send_tlb_inval_ctx_ppgtt, + .initialized = tlb_inval_initialized, + .flush = tlb_inval_flush, + .timeout_delay = tlb_inval_timeout_delay, +}; + /** * xe_guc_tlb_inval_init_early() - Init GuC TLB invalidation early * @guc: GuC object @@ -248,8 +379,14 @@ static const struct xe_tlb_inval_ops guc_tlb_inval_ops = { void xe_guc_tlb_inval_init_early(struct xe_guc *guc, struct xe_tlb_inval *tlb_inval) { + struct xe_device *xe = guc_to_xe(guc); + tlb_inval->private = guc; - tlb_inval->ops = &guc_tlb_inval_ops; + + if (xe->info.has_ctx_tlb_inval) + tlb_inval->ops = &guc_tlb_inval_ctx_ops; + else + tlb_inval->ops = &guc_tlb_inval_asid_ops; } /** diff --git a/drivers/gpu/drm/xe/xe_pci.c b/drivers/gpu/drm/xe/xe_pci.c index 91e0553a8163..6ea1199f703e 100644 --- a/drivers/gpu/drm/xe/xe_pci.c +++ b/drivers/gpu/drm/xe/xe_pci.c @@ -889,6 +889,7 @@ static int xe_info_init(struct xe_device *xe, xe->info.has_device_atomics_on_smem = 1; xe->info.has_range_tlb_inval = graphics_desc->has_range_tlb_inval; + xe->info.has_ctx_tlb_inval = graphics_desc->has_ctx_tlb_inval; xe->info.has_usm = graphics_desc->has_usm; xe->info.has_64bit_timestamp = graphics_desc->has_64bit_timestamp; diff --git a/drivers/gpu/drm/xe/xe_pci_types.h b/drivers/gpu/drm/xe/xe_pci_types.h index 5f20f56571d1..000b54cbcd0e 100644 --- a/drivers/gpu/drm/xe/xe_pci_types.h +++ b/drivers/gpu/drm/xe/xe_pci_types.h @@ -71,6 +71,7 @@ struct xe_graphics_desc { u8 has_atomic_enable_pte_bit:1; u8 has_indirect_ring_state:1; u8 has_range_tlb_inval:1; + u8 has_ctx_tlb_inval:1; u8 has_usm:1; u8 has_64bit_timestamp:1; }; -- 2.34.1