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 3902FC3065B for ; Mon, 24 Jun 2024 21:54:12 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id D29D810E281; Mon, 24 Jun 2024 21:54:11 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=intel.com header.i=@intel.com header.b="gWL4QJ2T"; dkim-atps=neutral Received: from mgamail.intel.com (mgamail.intel.com [198.175.65.14]) by gabe.freedesktop.org (Postfix) with ESMTPS id A9F6710E155 for ; Mon, 24 Jun 2024 21:54:09 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1719266050; x=1750802050; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=rxBvHf4YWFQtxKF2NKRFcsfGuRHCXc8T5BAlG4Ov2mE=; b=gWL4QJ2TAfkv9EDwMJXdD57yll6CMBE9xj6enrJ/BKfPSYUho71C58BU P8e3THRzWIgpams6VHH3nRCAgflvhgjq4nabY8+7OYmLbNNz9uqhGXqrR Dm9fnZbp2+uv1+/WJROsQX5F/iWscH+9JPxnzU/DSzdFwdeFdRkUK6Tif wb+L9XBBqaGc2ukpi4ELVUHdYVBZcz+OWlTy2y/VSuKYaGH76y+LmlIUL oAzkC8TLCJmyy4QYxKeOsZmF8tOEH/bn3a0RtS4jALnXW5Wq815MZwJCR TqcqjZqyzw9z4MG5/P+FpE3mYmWLrEvGrQqkrFfwUn07i/dyE6/uQHwN1 g==; X-CSE-ConnectionGUID: m7Wbqna6T7+0gZbv86D+cA== X-CSE-MsgGUID: DPyBg421SEi7RakviwzWOQ== X-IronPort-AV: E=McAfee;i="6700,10204,11113"; a="20073727" X-IronPort-AV: E=Sophos;i="6.08,262,1712646000"; d="scan'208";a="20073727" Received: from fmviesa010.fm.intel.com ([10.60.135.150]) by orvoesa106.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 24 Jun 2024 14:54:09 -0700 X-CSE-ConnectionGUID: 354XYHTzS7y+Fk50qznRsg== X-CSE-MsgGUID: IVZs5bnLRkWGueOFOnqWig== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="6.08,262,1712646000"; d="scan'208";a="43532251" Received: from guc-pnp-dev-box-1.fm.intel.com ([10.1.27.7]) by fmviesa010.fm.intel.com with ESMTP; 24 Jun 2024 14:54:07 -0700 From: Zhanjun Dong To: intel-xe@lists.freedesktop.org Cc: Zhanjun Dong Subject: [PATCH v11 4/5] drm/xe/guc: Extract GuC error capture lists Date: Mon, 24 Jun 2024 14:54:03 -0700 Message-Id: <20240624215404.3213075-5-zhanjun.dong@intel.com> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20240624215404.3213075-1-zhanjun.dong@intel.com> References: <20240624215404.3213075-1-zhanjun.dong@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" Upon the G2H Notify-Err-Capture event, parse through the GuC Log Buffer (error-capture-subregion) and generate one or more capture-nodes. A single node represents a single "engine- instance-capture-dump" and contains at least 3 register lists: global, engine-class and engine-instance. An internal link list is maintained to store one or more nodes. Because the link-list node generation happen before the call to devcoredump, duplicate global and engine-class register lists for each engine-instance register dump if we find dependent-engine resets in a engine-capture-group. To avoid dynamically allocate the output nodes during gt reset, pre-allocate a fixed number of empty nodes up front (at the time of ADS registration) that we can consume from or return to an internal cached list of nodes. Signed-off-by: Zhanjun Dong --- drivers/gpu/drm/xe/abi/guc_actions_abi.h | 8 + drivers/gpu/drm/xe/xe_guc_capture.c | 678 +++++++++++++++++++++- drivers/gpu/drm/xe/xe_guc_capture.h | 1 + drivers/gpu/drm/xe/xe_guc_capture_types.h | 45 ++ drivers/gpu/drm/xe/xe_guc_ct.c | 2 + drivers/gpu/drm/xe/xe_guc_submit.c | 65 ++- drivers/gpu/drm/xe/xe_guc_submit.h | 2 + 7 files changed, 783 insertions(+), 18 deletions(-) diff --git a/drivers/gpu/drm/xe/abi/guc_actions_abi.h b/drivers/gpu/drm/xe/abi/guc_actions_abi.h index 43ad4652c2b2..b54fe40fc5a9 100644 --- a/drivers/gpu/drm/xe/abi/guc_actions_abi.h +++ b/drivers/gpu/drm/xe/abi/guc_actions_abi.h @@ -176,6 +176,14 @@ enum xe_guc_sleep_state_status { #define GUC_LOG_CONTROL_VERBOSITY_MASK (0xF << GUC_LOG_CONTROL_VERBOSITY_SHIFT) #define GUC_LOG_CONTROL_DEFAULT_LOGGING (1 << 8) +enum xe_guc_state_capture_event_status { + XE_GUC_STATE_CAPTURE_EVENT_STATUS_SUCCESS = 0x0, + XE_GUC_STATE_CAPTURE_EVENT_STATUS_NOSPACE = 0x1, +}; + +#define XE_GUC_STATE_CAPTURE_EVENT_STATUS_MASK 0x000000FF +#define XE_GUC_ACTION_STATE_CAPTURE_NOTIFICATION_DATA_LEN 1 + #define XE_GUC_TLB_INVAL_TYPE_SHIFT 0 #define XE_GUC_TLB_INVAL_MODE_SHIFT 8 /* Flush PPC or SMRO caches along with TLB invalidation request */ diff --git a/drivers/gpu/drm/xe/xe_guc_capture.c b/drivers/gpu/drm/xe/xe_guc_capture.c index c2a08d4a6751..dc5421e950e5 100644 --- a/drivers/gpu/drm/xe/xe_guc_capture.c +++ b/drivers/gpu/drm/xe/xe_guc_capture.c @@ -140,6 +140,8 @@ static const char * const capture_engine_class_names[] = { "GSC-Other", }; +static void guc_capture_create_prealloc_nodes(struct xe_guc *guc); + static const struct __guc_mmio_reg_descr_group * guc_capture_get_one_list(const struct __guc_mmio_reg_descr_group *reglists, u32 owner, u32 type, u32 id) @@ -449,6 +451,12 @@ xe_guc_capture_getlist(struct xe_guc *guc, u32 owner, u32 type, u32 classid, voi return cache->status; } + /* + * ADS population of input registers is a good + * time to pre-allocate cachelist output nodes + */ + guc_capture_create_prealloc_nodes(guc); + ret = xe_guc_capture_getlistsize(guc, owner, type, classid, &size); if (ret) { cache->is_valid = true; @@ -638,6 +646,672 @@ static void check_guc_capture_size(struct xe_guc *guc) buffer_size, spare_size, capture_size); } +static void +guc_capture_add_node_to_list(struct __guc_capture_parsed_output *node, + struct list_head *list) +{ + list_add_tail(&node->link, list); +} + +static void +guc_capture_add_node_to_outlist(struct xe_guc_state_capture *gc, + struct __guc_capture_parsed_output *node) +{ + guc_capture_add_node_to_list(node, &gc->outlist); +} + +static void +guc_capture_add_node_to_cachelist(struct xe_guc_state_capture *gc, + struct __guc_capture_parsed_output *node) +{ + guc_capture_add_node_to_list(node, &gc->cachelist); +} + +static void +guc_capture_init_node(struct xe_guc *guc, struct __guc_capture_parsed_output *node) +{ + struct guc_mmio_reg *tmp[GUC_CAPTURE_LIST_TYPE_MAX]; + int i; + + for (i = 0; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) { + tmp[i] = node->reginfo[i].regs; + memset(tmp[i], 0, sizeof(struct guc_mmio_reg) * + guc->capture->max_mmio_per_node); + } + memset(node, 0, sizeof(*node)); + for (i = 0; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) + node->reginfo[i].regs = tmp[i]; + + INIT_LIST_HEAD(&node->link); +} + +/* + * KMD Init time flows: + * -------------------- + * --> alloc A: GuC input capture regs lists (registered to GuC via ADS). + * xe_guc_ads acquires the register lists by calling + * xe_guc_capture_list_size and xe_guc_capture_list_get 'n' times, + * where n = 1 for global-reg-list + + * num_engine_classes for class-reg-list + + * num_engine_classes for instance-reg-list + * (since all instances of the same engine-class type + * have an identical engine-instance register-list). + * ADS module also calls separately for PF vs VF. + * + * --> alloc B: GuC output capture buf (registered via guc_init_params(log_param)) + * Size = #define CAPTURE_BUFFER_SIZE (warns if on too-small) + * Note2: 'x 3' to hold multiple capture groups + * + * GUC Runtime notify capture: + * -------------------------- + * --> G2H STATE_CAPTURE_NOTIFICATION + * L--> xe_guc_capture_process + * L--> Loop through B (head..tail) and for each engine instance's + * err-state-captured register-list we find, we alloc 'C': + * --> alloc C: A capture-output-node structure that includes misc capture info along + * with 3 register list dumps (global, engine-class and engine-instance) + * This node is created from a pre-allocated list of blank nodes in + * guc->capture->cachelist and populated with the error-capture + * data from GuC and then it's added into guc->capture->outlist linked + * list. This list is used for matchup and printout by xe_devcoredump_read + * and xe_hw_engine_snapshot_print, (when user invokes the devcoredump sysfs). + * + * GUC --> notify context reset: + * ----------------------------- + * --> guc_exec_queue_timedout_job + * L--> xe_devcoredump + * L--> devcoredump_snapshot(..IS_GUC_CAPTURE) + * --> xe_hw_engine_snapshot_capture(..IS_GUC_CAPTURE) + * L--> xe_hw_engine_snapshot_from_capture is where + * detach C from internal linked list and add it into + * xe_hw_engine_snapshot struct (if the context and + * engine of the event notification matches a node + * in the link list). + * + * User Sysfs / Debugfs + * -------------------- + * --> xe_devcoredump_read-> + * L--> xxx_snapshot_print + * L--> xe_hw_engine_snapshot_print + * register lists values of the xe_hw_engine_snapshot + * saved from the error-engine-dump. + * + */ + +static int guc_capture_buf_cnt(struct __guc_capture_bufstate *buf) +{ + if (buf->wr >= buf->rd) + return (buf->wr - buf->rd); + return (buf->size - buf->rd) + buf->wr; +} + +static int guc_capture_buf_cnt_to_end(struct __guc_capture_bufstate *buf) +{ + if (buf->rd > buf->wr) + return (buf->size - buf->rd); + return (buf->wr - buf->rd); +} + +/* + * GuC's error-capture output is a ring buffer populated in a byte-stream fashion: + * + * The GuC Log buffer region for error-capture is managed like a ring buffer. + * The GuC firmware dumps error capture logs into this ring in a byte-stream flow. + * Additionally, as per the current and foreseeable future, all packed error- + * capture output structures are dword aligned. + * + * That said, if the GuC firmware is in the midst of writing a structure that is larger + * than one dword but the tail end of the err-capture buffer-region has lesser space left, + * we would need to extract that structure one dword at a time straddled across the end, + * onto the start of the ring. + * + * Below function, guc_capture_log_remove_dw is a helper for that. All callers of this + * function would typically do a straight-up memcpy from the ring contents and will only + * call this helper if their structure-extraction is straddling across the end of the + * ring. GuC firmware does not add any padding. The reason for the no-padding is to ease + * scalability for future expansion of output data types without requiring a redesign + * of the flow controls. + */ +static int +guc_capture_log_remove_dw(struct xe_guc *guc, struct __guc_capture_bufstate *buf, + u32 *dw) +{ + int tries = 2; + int avail = 0; + + if (!guc_capture_buf_cnt(buf)) + return 0; + + while (tries--) { + avail = guc_capture_buf_cnt_to_end(buf); + if (avail >= sizeof(u32)) { + *dw = xe_map_rd(guc_to_xe(guc), &guc->log.bo->vmap, + buf->data_offset + buf->rd, u32); + buf->rd += 4; + return 4; + } + if (avail) + xe_gt_dbg(guc_to_gt(guc), "Register capture log not dword aligned, skipping.\n"); + buf->rd = 0; + } + + return 0; +} + +static bool +guc_capture_data_extracted(struct xe_guc *guc, struct __guc_capture_bufstate *b, + int size, void *dest) +{ + if (guc_capture_buf_cnt_to_end(b) >= size) { + xe_map_memcpy_from(guc_to_xe(guc), dest, &guc->log.bo->vmap, + b->data_offset + b->rd, size); + b->rd += size; + return true; + } + return false; +} + +static int +guc_capture_log_get_group_hdr(struct xe_guc *guc, struct __guc_capture_bufstate *buf, + struct guc_state_capture_group_header_t *ghdr) +{ + int read = 0; + int fullsize = sizeof(struct guc_state_capture_group_header_t); + + if (fullsize > guc_capture_buf_cnt(buf)) + return -1; + + if (guc_capture_data_extracted(guc, buf, fullsize, (void *)ghdr)) + return 0; + + read += guc_capture_log_remove_dw(guc, buf, &ghdr->owner); + read += guc_capture_log_remove_dw(guc, buf, &ghdr->info); + if (read != fullsize) + return -1; + + return 0; +} + +static int +guc_capture_log_get_data_hdr(struct xe_guc *guc, struct __guc_capture_bufstate *buf, + struct guc_state_capture_header_t *hdr) +{ + int read = 0; + int fullsize = sizeof(struct guc_state_capture_header_t); + + if (fullsize > guc_capture_buf_cnt(buf)) + return -1; + + if (guc_capture_data_extracted(guc, buf, fullsize, (void *)hdr)) + return 0; + + read += guc_capture_log_remove_dw(guc, buf, &hdr->owner); + read += guc_capture_log_remove_dw(guc, buf, &hdr->info); + read += guc_capture_log_remove_dw(guc, buf, &hdr->lrca); + read += guc_capture_log_remove_dw(guc, buf, &hdr->guc_id); + read += guc_capture_log_remove_dw(guc, buf, &hdr->num_mmios); + if (read != fullsize) + return -1; + + return 0; +} + +static int +guc_capture_log_get_register(struct xe_guc *guc, struct __guc_capture_bufstate *buf, + struct guc_mmio_reg *reg) +{ + int read = 0; + int fullsize = sizeof(struct guc_mmio_reg); + + if (fullsize > guc_capture_buf_cnt(buf)) + return -1; + + if (guc_capture_data_extracted(guc, buf, fullsize, (void *)reg)) + return 0; + + read += guc_capture_log_remove_dw(guc, buf, ®->offset); + read += guc_capture_log_remove_dw(guc, buf, ®->value); + read += guc_capture_log_remove_dw(guc, buf, ®->flags); + read += guc_capture_log_remove_dw(guc, buf, ®->mask); + if (read != fullsize) + return -1; + + return 0; +} + +static struct __guc_capture_parsed_output * +guc_capture_get_prealloc_node(struct xe_guc *guc) +{ + struct __guc_capture_parsed_output *found = NULL; + + if (!list_empty(&guc->capture->cachelist)) { + struct __guc_capture_parsed_output *n, *ntmp; + + /* get first avail node from the cache list */ + list_for_each_entry_safe(n, ntmp, &guc->capture->cachelist, link) { + found = n; + break; + } + } else { + struct __guc_capture_parsed_output *n, *ntmp; + + /* traverse down and steal back the oldest node already allocated */ + list_for_each_entry_safe(n, ntmp, &guc->capture->outlist, link) { + found = n; + } + } + if (found) { + list_del(&found->link); + guc_capture_init_node(guc, found); + } + + return found; +} + +static struct __guc_capture_parsed_output * +guc_capture_clone_node(struct xe_guc *guc, struct __guc_capture_parsed_output *original, + u32 keep_reglist_mask) +{ + struct __guc_capture_parsed_output *new; + int i; + + new = guc_capture_get_prealloc_node(guc); + if (!new) + return NULL; + if (!original) + return new; + + new->is_partial = original->is_partial; + + /* copy reg-lists that we want to clone */ + for (i = 0; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) { + if (keep_reglist_mask & BIT(i)) { + XE_WARN_ON(original->reginfo[i].num_regs > + guc->capture->max_mmio_per_node); + + memcpy(new->reginfo[i].regs, original->reginfo[i].regs, + original->reginfo[i].num_regs * sizeof(struct guc_mmio_reg)); + + new->reginfo[i].num_regs = original->reginfo[i].num_regs; + new->reginfo[i].vfid = original->reginfo[i].vfid; + + if (i == GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS) { + new->eng_class = original->eng_class; + } else if (i == GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE) { + new->eng_inst = original->eng_inst; + new->guc_id = original->guc_id; + new->lrca = original->lrca; + } + } + } + + return new; +} + +static int +guc_capture_extract_reglists(struct xe_guc *guc, struct __guc_capture_bufstate *buf) +{ + struct xe_gt *gt = guc_to_gt(guc); + struct guc_state_capture_group_header_t ghdr = {0}; + struct guc_state_capture_header_t hdr = {0}; + struct __guc_capture_parsed_output *node = NULL; + struct guc_mmio_reg *regs = NULL; + int i, numlists, numregs, ret = 0; + enum guc_capture_type datatype; + struct guc_mmio_reg tmp; + bool is_partial = false; + + i = guc_capture_buf_cnt(buf); + if (!i) + return -ENODATA; + + if (i % sizeof(u32)) { + xe_gt_warn(gt, "Got mis-aligned register capture entries\n"); + ret = -EIO; + goto bailout; + } + + /* first get the capture group header */ + if (guc_capture_log_get_group_hdr(guc, buf, &ghdr)) { + ret = -EIO; + goto bailout; + } + /* + * we would typically expect a layout as below where n would be expected to be + * anywhere between 3 to n where n > 3 if we are seeing multiple dependent engine + * instances being reset together. + * ____________________________________________ + * | Capture Group | + * | ________________________________________ | + * | | Capture Group Header: | | + * | | - num_captures = 5 | | + * | |______________________________________| | + * | ________________________________________ | + * | | Capture1: | | + * | | Hdr: GLOBAL, numregs=a | | + * | | ____________________________________ | | + * | | | Reglist | | | + * | | | - reg1, reg2, ... rega | | | + * | | |__________________________________| | | + * | |______________________________________| | + * | ________________________________________ | + * | | Capture2: | | + * | | Hdr: CLASS=RENDER/COMPUTE, numregs=b| | + * | | ____________________________________ | | + * | | | Reglist | | | + * | | | - reg1, reg2, ... regb | | | + * | | |__________________________________| | | + * | |______________________________________| | + * | ________________________________________ | + * | | Capture3: | | + * | | Hdr: INSTANCE=RCS, numregs=c | | + * | | ____________________________________ | | + * | | | Reglist | | | + * | | | - reg1, reg2, ... regc | | | + * | | |__________________________________| | | + * | |______________________________________| | + * | ________________________________________ | + * | | Capture4: | | + * | | Hdr: CLASS=RENDER/COMPUTE, numregs=d| | + * | | ____________________________________ | | + * | | | Reglist | | | + * | | | - reg1, reg2, ... regd | | | + * | | |__________________________________| | | + * | |______________________________________| | + * | ________________________________________ | + * | | Capture5: | | + * | | Hdr: INSTANCE=CCS0, numregs=e | | + * | | ____________________________________ | | + * | | | Reglist | | | + * | | | - reg1, reg2, ... rege | | | + * | | |__________________________________| | | + * | |______________________________________| | + * |__________________________________________| + */ + is_partial = FIELD_GET(CAP_GRP_HDR_CAPTURE_TYPE, ghdr.info); + numlists = FIELD_GET(CAP_GRP_HDR_NUM_CAPTURES, ghdr.info); + + while (numlists--) { + if (guc_capture_log_get_data_hdr(guc, buf, &hdr)) { + ret = -EIO; + break; + } + + datatype = FIELD_GET(CAP_HDR_CAPTURE_TYPE, hdr.info); + if (datatype > GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE) { + /* unknown capture type - skip over to next capture set */ + numregs = FIELD_GET(CAP_HDR_NUM_MMIOS, hdr.num_mmios); + while (numregs--) { + if (guc_capture_log_get_register(guc, buf, &tmp)) { + ret = -EIO; + break; + } + } + continue; + } else if (node) { + /* + * Based on the current capture type and what we have so far, + * decide if we should add the current node into the internal + * linked list for match-up when xe_devcoredump calls later + * (and alloc a blank node for the next set of reglists) + * or continue with the same node or clone the current node + * but only retain the global or class registers (such as the + * case of dependent engine resets). + */ + if (datatype == GUC_CAPTURE_LIST_TYPE_GLOBAL) { + guc_capture_add_node_to_outlist(guc->capture, node); + node = NULL; + } else if (datatype == GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS && + node->reginfo[GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS].num_regs) { + /* Add to list, clone node and duplicate global list */ + guc_capture_add_node_to_outlist(guc->capture, node); + node = guc_capture_clone_node(guc, node, + GCAP_PARSED_REGLIST_INDEX_GLOBAL); + } else if (datatype == GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE && + node->reginfo[GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE].num_regs) { + /* Add to list, clone node and duplicate global + class lists */ + guc_capture_add_node_to_outlist(guc->capture, node); + node = guc_capture_clone_node(guc, node, + (GCAP_PARSED_REGLIST_INDEX_GLOBAL | + GCAP_PARSED_REGLIST_INDEX_ENGCLASS)); + } + } + + if (!node) { + node = guc_capture_get_prealloc_node(guc); + if (!node) { + ret = -ENOMEM; + break; + } + if (datatype != GUC_CAPTURE_LIST_TYPE_GLOBAL) + xe_gt_dbg(gt, "Register capture missing global dump: %08x!\n", + datatype); + } + node->is_partial = is_partial; + node->reginfo[datatype].vfid = FIELD_GET(CAP_HDR_CAPTURE_VFID, hdr.owner); + + switch (datatype) { + case GUC_CAPTURE_LIST_TYPE_ENGINE_INSTANCE: + node->eng_class = FIELD_GET(CAP_HDR_ENGINE_CLASS, hdr.info); + node->eng_inst = FIELD_GET(CAP_HDR_ENGINE_INSTANCE, hdr.info); + node->lrca = hdr.lrca; + node->guc_id = hdr.guc_id; + break; + case GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS: + node->eng_class = FIELD_GET(CAP_HDR_ENGINE_CLASS, hdr.info); + break; + default: + break; + } + + numregs = FIELD_GET(CAP_HDR_NUM_MMIOS, hdr.num_mmios); + if (numregs > guc->capture->max_mmio_per_node) { + xe_gt_dbg(gt, "Register capture list extraction clipped by prealloc!\n"); + numregs = guc->capture->max_mmio_per_node; + } + node->reginfo[datatype].num_regs = numregs; + regs = node->reginfo[datatype].regs; + i = 0; + while (numregs--) { + if (guc_capture_log_get_register(guc, buf, ®s[i++])) { + ret = -EIO; + break; + } + } + } + +bailout: + if (node) { + /* If we have data, add to linked list for match-up when xe_devcoredump calls */ + for (i = GUC_CAPTURE_LIST_TYPE_GLOBAL; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) { + if (node->reginfo[i].regs) { + guc_capture_add_node_to_outlist(guc->capture, node); + node = NULL; + break; + } + } + if (node) /* else return it back to cache list */ + guc_capture_add_node_to_cachelist(guc->capture, node); + } + return ret; +} + +static int __guc_capture_flushlog_complete(struct xe_guc *guc) +{ + u32 action[] = { + XE_GUC_ACTION_LOG_BUFFER_FILE_FLUSH_COMPLETE, + GUC_LOG_BUFFER_CAPTURE + }; + + return xe_guc_ct_send_g2h_handler(&guc->ct, action, ARRAY_SIZE(action)); +} + +static void __guc_capture_process_output(struct xe_guc *guc) +{ + unsigned int buffer_size, read_offset, write_offset, full_count; + struct xe_uc *uc = container_of(guc, typeof(*uc), guc); + struct guc_log_buffer_state log_buf_state_local; + struct __guc_capture_bufstate buf; + bool new_overflow; + int ret; + u32 log_buf_state_offset; + u32 src_data_offset; + + log_buf_state_offset = sizeof(struct guc_log_buffer_state) * GUC_LOG_BUFFER_CAPTURE; + src_data_offset = xe_guc_get_log_buffer_offset(&guc->log, GUC_LOG_BUFFER_CAPTURE); + + /* + * Make a copy of the state structure, inside GuC log buffer + * (which is uncached mapped), on the stack to avoid reading + * from it multiple times. + */ + xe_map_memcpy_from(guc_to_xe(guc), &log_buf_state_local, &guc->log.bo->vmap, + log_buf_state_offset, sizeof(struct guc_log_buffer_state)); + + buffer_size = xe_guc_get_log_buffer_size(&guc->log, GUC_LOG_BUFFER_CAPTURE); + read_offset = log_buf_state_local.read_ptr; + write_offset = log_buf_state_local.sampled_write_ptr; + full_count = log_buf_state_local.buffer_full_cnt; + + /* Bookkeeping stuff */ + guc->log.stats[GUC_LOG_BUFFER_CAPTURE].flush += log_buf_state_local.flush_to_file; + new_overflow = xe_guc_check_log_buf_overflow(&guc->log, GUC_LOG_BUFFER_CAPTURE, + full_count); + + /* Now copy the actual logs. */ + if (unlikely(new_overflow)) { + /* copy the whole buffer in case of overflow */ + read_offset = 0; + write_offset = buffer_size; + } else if (unlikely((read_offset > buffer_size) || + (write_offset > buffer_size))) { + xe_gt_err(guc_to_gt(guc), + "Register capture buffer in invalid state: read = 0x%X, size = 0x%X!\n", + read_offset, buffer_size); + /* copy whole buffer as offsets are unreliable */ + read_offset = 0; + write_offset = buffer_size; + } + + buf.size = buffer_size; + buf.rd = read_offset; + buf.wr = write_offset; + buf.data_offset = src_data_offset; + + if (!xe_guc_read_stopped(guc)) { + do { + ret = guc_capture_extract_reglists(guc, &buf); + } while (ret >= 0); + } + + /* Update the state of log buffer err-cap state */ + xe_map_wr(guc_to_xe(guc), &guc->log.bo->vmap, + log_buf_state_offset + offsetof(struct guc_log_buffer_state, read_ptr), u32, + write_offset); + /* Clear the flush_to_file from local first, the local was loaded by above + * xe_map_memcpy_from. + */ + log_buf_state_local.flush_to_file = 0; + /* Then write out the "updated local" through xe_map_wr() */ + xe_map_wr(guc_to_xe(guc), &guc->log.bo->vmap, + log_buf_state_offset + offsetof(struct guc_log_buffer_state, flags), u32, + log_buf_state_local.flags); + __guc_capture_flushlog_complete(guc); +} + +/* + * xe_guc_capture_process - Process GuC register captured data + * @guc: The GuC object + * + * When GuC captured data is ready, GuC will send message + * XE_GUC_ACTION_STATE_CAPTURE_NOTIFICATION to host, this function will be + * called to process the data comes with the message. + * + * Returns: None + */ +void xe_guc_capture_process(struct xe_guc *guc) +{ + if (guc->capture) + __guc_capture_process_output(guc); +} + +static struct __guc_capture_parsed_output * +guc_capture_alloc_one_node(struct xe_guc *guc) +{ + struct drm_device *drm = guc_to_drm(guc); + struct __guc_capture_parsed_output *new; + int i; + + new = drmm_kzalloc(drm, sizeof(*new), GFP_KERNEL); + if (!new) + return NULL; + + for (i = 0; i < GUC_CAPTURE_LIST_TYPE_MAX; ++i) { + new->reginfo[i].regs = drmm_kzalloc(drm, guc->capture->max_mmio_per_node * + sizeof(struct guc_mmio_reg), GFP_KERNEL); + if (!new->reginfo[i].regs) { + while (i) + drmm_kfree(drm, new->reginfo[--i].regs); + drmm_kfree(drm, new); + return NULL; + } + } + guc_capture_init_node(guc, new); + + return new; +} + +static void +__guc_capture_create_prealloc_nodes(struct xe_guc *guc) +{ + struct __guc_capture_parsed_output *node = NULL; + int i; + + for (i = 0; i < PREALLOC_NODES_MAX_COUNT; ++i) { + node = guc_capture_alloc_one_node(guc); + if (!node) { + xe_gt_warn(guc_to_gt(guc), "Register capture pre-alloc-cache failure\n"); + /* dont free the priors, use what we got and cleanup at shutdown */ + return; + } + guc_capture_add_node_to_cachelist(guc->capture, node); + } +} + +static int +guc_get_max_reglist_count(struct xe_guc *guc) +{ + int i, j, k, tmp, maxregcount = 0; + + for (i = 0; i < GUC_CAPTURE_LIST_INDEX_MAX; ++i) { + for (j = 0; j < GUC_CAPTURE_LIST_TYPE_MAX; ++j) { + for (k = 0; k < GUC_MAX_ENGINE_CLASSES; ++k) { + if (j == GUC_CAPTURE_LIST_TYPE_GLOBAL && k > 0) + continue; + + tmp = guc_cap_list_num_regs(guc->capture, i, j, k); + if (tmp > maxregcount) + maxregcount = tmp; + } + } + } + if (!maxregcount) + maxregcount = PREALLOC_NODES_DEFAULT_NUMREGS; + + return maxregcount; +} + +static void +guc_capture_create_prealloc_nodes(struct xe_guc *guc) +{ + /* skip if we've already done the pre-alloc */ + if (guc->capture->max_mmio_per_node) + return; + + guc->capture->max_mmio_per_node = guc_get_max_reglist_count(guc); + __guc_capture_create_prealloc_nodes(guc); +} + /* * xe_guc_capture_init - Init for GuC register capture * @guc: The GuC object @@ -654,7 +1328,9 @@ int xe_guc_capture_init(struct xe_guc *guc) return -ENOMEM; guc->capture->reglists = guc_capture_get_device_reglist(guc); - check_guc_capture_size(guc); + INIT_LIST_HEAD(&guc->capture->outlist); + INIT_LIST_HEAD(&guc->capture->cachelist); + return 0; } diff --git a/drivers/gpu/drm/xe/xe_guc_capture.h b/drivers/gpu/drm/xe/xe_guc_capture.h index ead5a1d2b5ad..3d54e71cfba8 100644 --- a/drivers/gpu/drm/xe/xe_guc_capture.h +++ b/drivers/gpu/drm/xe/xe_guc_capture.h @@ -36,6 +36,7 @@ struct __guc_mmio_reg_descr_group { struct __guc_mmio_reg_descr *extlist; /* only used for steered registers */ }; +void xe_guc_capture_process(struct xe_guc *guc); int xe_guc_capture_getlist(struct xe_guc *guc, u32 owner, u32 type, u32 classid, void **outptr); int xe_guc_capture_getlistsize(struct xe_guc *guc, u32 owner, u32 type, u32 classid, size_t *size); int xe_guc_capture_getnullheader(struct xe_guc *guc, void **outptr, size_t *size); diff --git a/drivers/gpu/drm/xe/xe_guc_capture_types.h b/drivers/gpu/drm/xe/xe_guc_capture_types.h index 0eedb539348a..14890e51bad4 100644 --- a/drivers/gpu/drm/xe/xe_guc_capture_types.h +++ b/drivers/gpu/drm/xe/xe_guc_capture_types.h @@ -10,6 +10,51 @@ #include "xe_guc_fwif.h" +/* + * struct __guc_capture_bufstate + * + * Book-keeping structure used to track read and write pointers + * as we extract error capture data from the GuC-log-buffer's + * error-capture region as a stream of dwords. + */ +struct __guc_capture_bufstate { + u32 size; + u32 data_offset; + u32 rd; + u32 wr; +}; + +/* + * struct __guc_capture_parsed_output - extracted error capture node + * + * A single unit of extracted error-capture output data grouped together + * at an engine-instance level. We keep these nodes in a linked list. + * See cachelist and outlist below. + */ +struct __guc_capture_parsed_output { + /* + * A single set of 3 capture lists: a global-list + * an engine-class-list and an engine-instance list. + * outlist in __guc_capture_parsed_output will keep + * a linked list of these nodes that will eventually + * be detached from outlist and attached into to + * xe_codedump in response to a context reset + */ + struct list_head link; + bool is_partial; + u32 eng_class; + u32 eng_inst; + u32 guc_id; + u32 lrca; + struct gcap_reg_list_info { + u32 vfid; + u32 num_regs; + struct guc_mmio_reg *regs; + } reginfo[GUC_CAPTURE_LIST_TYPE_MAX]; +#define GCAP_PARSED_REGLIST_INDEX_GLOBAL BIT(GUC_CAPTURE_LIST_TYPE_GLOBAL) +#define GCAP_PARSED_REGLIST_INDEX_ENGCLASS BIT(GUC_CAPTURE_LIST_TYPE_ENGINE_CLASS) +}; + /* * struct __guc_capture_ads_cache * diff --git a/drivers/gpu/drm/xe/xe_guc_ct.c b/drivers/gpu/drm/xe/xe_guc_ct.c index b4137fe195a4..2d5b56f6fb63 100644 --- a/drivers/gpu/drm/xe/xe_guc_ct.c +++ b/drivers/gpu/drm/xe/xe_guc_ct.c @@ -1089,6 +1089,8 @@ static int process_g2h_msg(struct xe_guc_ct *ct, u32 *msg, u32 len) /* Selftest only at the moment */ break; case XE_GUC_ACTION_STATE_CAPTURE_NOTIFICATION: + ret = xe_guc_error_capture_handler(guc, payload, adj_len); + break; case XE_GUC_ACTION_NOTIFY_FLUSH_LOG_BUFFER_TO_FILE: /* FIXME: Handle this */ break; diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c index 74552391dc5a..3319e7993cc2 100644 --- a/drivers/gpu/drm/xe/xe_guc_submit.c +++ b/drivers/gpu/drm/xe/xe_guc_submit.c @@ -27,6 +27,7 @@ #include "xe_gt_clock.h" #include "xe_gt_printk.h" #include "xe_guc.h" +#include "xe_guc_capture.h" #include "xe_guc_ct.h" #include "xe_guc_exec_queue_types.h" #include "xe_guc_id_mgr.h" @@ -800,7 +801,7 @@ static void guc_exec_queue_free_job(struct drm_sched_job *drm_job) xe_sched_job_put(job); } -static int guc_read_stopped(struct xe_guc *guc) +int xe_guc_read_stopped(struct xe_guc *guc) { return atomic_read(&guc->submission_state.stopped); } @@ -822,7 +823,7 @@ static void disable_scheduling_deregister(struct xe_guc *guc, set_min_preemption_timeout(guc, q); smp_rmb(); ret = wait_event_timeout(guc->ct.wq, !exec_queue_pending_enable(q) || - guc_read_stopped(guc), HZ * 5); + xe_guc_read_stopped(guc), HZ * 5); if (!ret) { struct xe_gpu_scheduler *sched = &q->guc->sched; @@ -937,7 +938,7 @@ static void xe_guc_exec_queue_lr_cleanup(struct work_struct *w) */ ret = wait_event_timeout(guc->ct.wq, !exec_queue_pending_disable(q) || - guc_read_stopped(guc), HZ * 5); + xe_guc_read_stopped(guc), HZ * 5); if (!ret) { drm_warn(&xe->drm, "Schedule disable failed to respond"); xe_sched_submission_start(sched); @@ -1005,8 +1006,8 @@ static void enable_scheduling(struct xe_exec_queue *q) ret = wait_event_timeout(guc->ct.wq, !exec_queue_pending_enable(q) || - guc_read_stopped(guc), HZ * 5); - if (!ret || guc_read_stopped(guc)) { + xe_guc_read_stopped(guc), HZ * 5); + if (!ret || xe_guc_read_stopped(guc)) { xe_gt_warn(guc_to_gt(guc), "Schedule enable failed to respond"); set_exec_queue_banned(q); xe_gt_reset_async(q->gt); @@ -1109,8 +1110,8 @@ guc_exec_queue_timedout_job(struct drm_sched_job *drm_job) */ ret = wait_event_timeout(guc->ct.wq, !exec_queue_pending_enable(q) || - guc_read_stopped(guc), HZ * 5); - if (!ret || guc_read_stopped(guc)) + xe_guc_read_stopped(guc), HZ * 5); + if (!ret || xe_guc_read_stopped(guc)) goto trigger_reset; /* @@ -1134,8 +1135,8 @@ guc_exec_queue_timedout_job(struct drm_sched_job *drm_job) smp_rmb(); ret = wait_event_timeout(guc->ct.wq, !exec_queue_pending_disable(q) || - guc_read_stopped(guc), HZ * 5); - if (!ret || guc_read_stopped(guc)) { + xe_guc_read_stopped(guc), HZ * 5); + if (!ret || xe_guc_read_stopped(guc)) { trigger_reset: if (!ret) xe_gt_warn(guc_to_gt(guc), "Schedule disable failed to respond"); @@ -1307,7 +1308,7 @@ static void suspend_fence_signal(struct xe_exec_queue *q) struct xe_device *xe = guc_to_xe(guc); xe_assert(xe, exec_queue_suspended(q) || exec_queue_killed(q) || - guc_read_stopped(guc)); + xe_guc_read_stopped(guc)); xe_assert(xe, q->guc->suspend_pending); q->guc->suspend_pending = false; @@ -1323,9 +1324,9 @@ static void __guc_exec_queue_process_msg_suspend(struct xe_sched_msg *msg) if (guc_exec_queue_allowed_to_change_state(q) && !exec_queue_suspended(q) && exec_queue_enabled(q)) { wait_event(guc->ct.wq, q->guc->resume_time != RESUME_PENDING || - guc_read_stopped(guc)); + xe_guc_read_stopped(guc)); - if (!guc_read_stopped(guc)) { + if (!xe_guc_read_stopped(guc)) { s64 since_resume_ms = ktime_ms_delta(ktime_get(), q->guc->resume_time); @@ -1439,7 +1440,7 @@ static int guc_exec_queue_init(struct xe_exec_queue *q) q->entity = &ge->entity; - if (guc_read_stopped(guc)) + if (xe_guc_read_stopped(guc)) xe_sched_stop(sched); mutex_unlock(&guc->submission_state.lock); @@ -1566,7 +1567,7 @@ static void guc_exec_queue_suspend_wait(struct xe_exec_queue *q) struct xe_guc *guc = exec_queue_to_guc(q); wait_event(q->guc->suspend_wait, !q->guc->suspend_pending || - guc_read_stopped(guc)); + xe_guc_read_stopped(guc)); } static void guc_exec_queue_resume(struct xe_exec_queue *q) @@ -1677,7 +1678,7 @@ int xe_guc_submit_reset_prepare(struct xe_guc *guc) void xe_guc_submit_reset_wait(struct xe_guc *guc) { - wait_event(guc->ct.wq, !guc_read_stopped(guc)); + wait_event(guc->ct.wq, !xe_guc_read_stopped(guc)); } void xe_guc_submit_stop(struct xe_guc *guc) @@ -1686,7 +1687,7 @@ void xe_guc_submit_stop(struct xe_guc *guc) unsigned long index; struct xe_device *xe = guc_to_xe(guc); - xe_assert(xe, guc_read_stopped(guc) == 1); + xe_assert(xe, xe_guc_read_stopped(guc) == 1); mutex_lock(&guc->submission_state.lock); @@ -1724,7 +1725,7 @@ int xe_guc_submit_start(struct xe_guc *guc) unsigned long index; struct xe_device *xe = guc_to_xe(guc); - xe_assert(xe, guc_read_stopped(guc) == 1); + xe_assert(xe, xe_guc_read_stopped(guc) == 1); mutex_lock(&guc->submission_state.lock); atomic_dec(&guc->submission_state.stopped); @@ -1915,6 +1916,36 @@ int xe_guc_exec_queue_reset_handler(struct xe_guc *guc, u32 *msg, u32 len) return 0; } +/* + * xe_guc_error_capture_handler - Handler of GuC captured message + * @guc: The GuC object + * @msg: Point to the message + * @len: The message length + * + * When GuC captured data is ready, GuC will send message + * XE_GUC_ACTION_STATE_CAPTURE_NOTIFICATION to host, this function will be + * called 1st to check status before process the data comes with the message. + * + * Returns: None + */ +int xe_guc_error_capture_handler(struct xe_guc *guc, u32 *msg, u32 len) +{ + u32 status; + + if (unlikely(len != XE_GUC_ACTION_STATE_CAPTURE_NOTIFICATION_DATA_LEN)) { + xe_gt_dbg(guc_to_gt(guc), "Invalid length %u", len); + return -EPROTO; + } + + status = msg[0] & XE_GUC_STATE_CAPTURE_EVENT_STATUS_MASK; + if (status == XE_GUC_STATE_CAPTURE_EVENT_STATUS_NOSPACE) + xe_gt_warn(guc_to_gt(guc), "G2H-Error capture no space"); + + xe_guc_capture_process(guc); + + return 0; +} + int xe_guc_exec_queue_memory_cat_error_handler(struct xe_guc *guc, u32 *msg, u32 len) { diff --git a/drivers/gpu/drm/xe/xe_guc_submit.h b/drivers/gpu/drm/xe/xe_guc_submit.h index 4ad5f4c1b084..d92256de473e 100644 --- a/drivers/gpu/drm/xe/xe_guc_submit.h +++ b/drivers/gpu/drm/xe/xe_guc_submit.h @@ -19,12 +19,14 @@ void xe_guc_submit_reset_wait(struct xe_guc *guc); void xe_guc_submit_stop(struct xe_guc *guc); int xe_guc_submit_start(struct xe_guc *guc); +int xe_guc_read_stopped(struct xe_guc *guc); int xe_guc_sched_done_handler(struct xe_guc *guc, u32 *msg, u32 len); int xe_guc_deregister_done_handler(struct xe_guc *guc, u32 *msg, u32 len); int xe_guc_exec_queue_reset_handler(struct xe_guc *guc, u32 *msg, u32 len); int xe_guc_exec_queue_memory_cat_error_handler(struct xe_guc *guc, u32 *msg, u32 len); int xe_guc_exec_queue_reset_failure_handler(struct xe_guc *guc, u32 *msg, u32 len); +int xe_guc_error_capture_handler(struct xe_guc *guc, u32 *msg, u32 len); struct xe_guc_submit_exec_queue_snapshot * xe_guc_exec_queue_snapshot_capture(struct xe_exec_queue *q); -- 2.34.1