From patchwork Fri Jun 30 05:37:08 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Cyril Bur X-Patchwork-Id: 782660 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from lists.ozlabs.org (lists.ozlabs.org [103.22.144.68]) (using TLSv1.2 with cipher ADH-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 3wzQM63Vftz9s82 for ; Fri, 30 Jun 2017 15:39:14 +1000 (AEST) Received: from lists.ozlabs.org (lists.ozlabs.org [IPv6:2401:3900:2:1::3]) by lists.ozlabs.org (Postfix) with ESMTP id 3wzQM62rbjzDr6S for ; Fri, 30 Jun 2017 15:39:14 +1000 (AEST) X-Original-To: skiboot@lists.ozlabs.org Delivered-To: skiboot@lists.ozlabs.org Received: from mx0a-001b2d01.pphosted.com (mx0a-001b2d01.pphosted.com [148.163.156.1]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by lists.ozlabs.org (Postfix) with ESMTPS id 3wzQLr4CmYzDr60 for ; Fri, 30 Jun 2017 15:39:00 +1000 (AEST) Received: from pps.filterd (m0098409.ppops.net [127.0.0.1]) by mx0a-001b2d01.pphosted.com (8.16.0.20/8.16.0.20) with SMTP id v5U5cfd3102002 for ; Fri, 30 Jun 2017 01:38:57 -0400 Received: from e23smtp08.au.ibm.com (e23smtp08.au.ibm.com [202.81.31.141]) by mx0a-001b2d01.pphosted.com with ESMTP id 2bd0yuppyn-1 (version=TLSv1.2 cipher=AES256-SHA bits=256 verify=NOT) for ; Fri, 30 Jun 2017 01:38:56 -0400 Received: from localhost by e23smtp08.au.ibm.com with IBM ESMTP SMTP Gateway: Authorized Use Only! Violators will be prosecuted for from ; Fri, 30 Jun 2017 15:38:54 +1000 Received: from d23relay09.au.ibm.com (202.81.31.228) by e23smtp08.au.ibm.com (202.81.31.205) with IBM ESMTP SMTP Gateway: Authorized Use Only! Violators will be prosecuted; Fri, 30 Jun 2017 15:38:51 +1000 Received: from d23av01.au.ibm.com (d23av01.au.ibm.com [9.190.234.96]) by d23relay09.au.ibm.com (8.14.9/8.14.9/NCO v10.0) with ESMTP id v5U5cpCo64749622 for ; Fri, 30 Jun 2017 15:38:51 +1000 Received: from d23av01.au.ibm.com (localhost [127.0.0.1]) by d23av01.au.ibm.com (8.14.4/8.14.4/NCO v10.0 AVout) with ESMTP id v5U5coHe014539 for ; Fri, 30 Jun 2017 15:38:50 +1000 Received: from ozlabs.au.ibm.com (ozlabs.au.ibm.com [9.192.253.14]) by d23av01.au.ibm.com (8.14.4/8.14.4/NCO v10.0 AVin) with ESMTP id v5U5coci014536; Fri, 30 Jun 2017 15:38:50 +1000 Received: from camb691.ozlabs.ibm.com (haven.au.ibm.com [9.192.254.114]) (using TLSv1.2 with cipher DHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by ozlabs.au.ibm.com (Postfix) with ESMTPSA id C8A8AA0124; Fri, 30 Jun 2017 15:38:50 +1000 (AEST) From: Cyril Bur To: skiboot@lists.ozlabs.org Date: Fri, 30 Jun 2017 15:37:08 +1000 X-Mailer: git-send-email 2.13.2 In-Reply-To: <1824973.XXWCfeLztN@new-mexico> References: <1824973.XXWCfeLztN@new-mexico> X-TM-AS-MML: disable x-cbid: 17063005-0048-0000-0000-0000024D9ADA X-IBM-AV-DETECTION: SAVI=unused REMOTE=unused XFE=unused x-cbparentid: 17063005-0049-0000-0000-000047FE9EDD Message-Id: <20170630053708.23917-1-cyril.bur@au1.ibm.com> X-Proofpoint-Virus-Version: vendor=fsecure engine=2.50.10432:, , definitions=2017-06-30_03:, , signatures=0 X-Proofpoint-Spam-Details: rule=outbound_notspam policy=outbound score=0 spamscore=0 suspectscore=1 malwarescore=0 phishscore=0 adultscore=0 bulkscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1703280000 definitions=main-1706300090 Subject: [Skiboot] [PATCH] core/flash: Make opal_flash_op() actually asynchronous X-BeenThere: skiboot@lists.ozlabs.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: Mailing list for skiboot development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: alistair@popple.id.au, sam@mendozajonas.com, rlippert@google.com MIME-Version: 1.0 Errors-To: skiboot-bounces+incoming=patchwork.ozlabs.org@lists.ozlabs.org Sender: "Skiboot" This patch provides a simple (although not particularly efficient) asynchronous capability to the opal_flash interface. The advantage of this approach is that it doesn't require any changing of blocklevel or its backends to provide an asynchronous implementation. This is also the disadvantage of this implementation as all it actually does is break the work up in chunks that it can performed quickly, but still synchronously. Only a backend could provide a more asynchronous implementation. This solves a problem we have right now with the opal_flash_erase call where it can block in Skiboot for around three minutes. This causes a variety of problems in Linux due to a processor being gone for a long time. For example: [ 98.610043] INFO: rcu_sched detected stalls on CPUs/tasks: [ 98.610050] 113-...: (1 GPs behind) idle=96f/140000000000000/0 softirq=527/528 fqs=1044 [ 98.610051] (detected by 112, t=2102 jiffies, g=223, c=222, q=123) [ 98.610060] Task dump for CPU 113: [ 98.610062] pflash R running task 0 3335 3333 0x00040004 [ 98.610066] Call Trace: [ 98.610070] [c000001fdd847730] [0000000000000001] 0x1 (unreliable) [ 98.610076] [c000001fdd847900] [c000000000013854] __switch_to+0x1e8/0x1f4 [ 98.610081] [c000001fdd847960] [c0000000006122c4] __schedule+0x32c/0x874 [ 98.610083] [c000001fdd847a30] [c000001fdd847b40] 0xc000001fdd847b40 It is for this reason that breaking the work up in smaller chunks solves this problem as Skiboot can return the CPU to Linux between chunks to avoid Linux getting upset. Reported-By: Samuel Mendoza-Jonas Signed-off-by: Cyril Bur --- Like this Alistair? Something something you're correct. Lets go with this one. core/flash.c | 116 +++++++++++++++++++++++++++----- doc/opal-api/opal-flash-110-111-112.rst | 4 ++ 2 files changed, 103 insertions(+), 17 deletions(-) diff --git a/core/flash.c b/core/flash.c index 177f7ae1..71cc3c87 100644 --- a/core/flash.c +++ b/core/flash.c @@ -28,6 +28,23 @@ #include #include #include +#include +#include + +enum flash_op { + FLASH_OP_READ, + FLASH_OP_WRITE, + FLASH_OP_ERASE, +}; + +struct flash_async_info { + enum flash_op op; + struct timer poller; + uint64_t token; + uint64_t pos; + uint64_t len; + uint64_t buf; +}; struct flash { struct list_node list; @@ -36,6 +53,7 @@ struct flash { uint64_t size; uint32_t block_size; int id; + struct flash_async_info async; }; static LIST_HEAD(flashes); @@ -200,6 +218,53 @@ static int flash_nvram_probe(struct flash *flash, struct ffs_handle *ffs) /* core flash support */ +/* + * Called with flash lock held, drop it on async completion + */ +static void flash_poll(struct timer *t __unused, void *data, uint64_t now __unused) +{ + struct flash *flash = data; + uint64_t offset, buf, len; + int rc; + + offset = flash->async.pos; + buf = flash->async.buf; + len = MIN(flash->async.len, flash->block_size); + + switch (flash->async.op) { + case FLASH_OP_READ: + rc = blocklevel_raw_read(flash->bl, offset, (void *)buf, len); + break; + case FLASH_OP_WRITE: + rc = blocklevel_raw_write(flash->bl, offset, (void *)buf, len); + break; + case FLASH_OP_ERASE: + rc = blocklevel_erase(flash->bl, offset, len); + break; + default: + assert(0); + } + + if (rc) + rc = OPAL_HARDWARE; + + flash->async.pos += len; + flash->async.buf += len; + flash->async.len -= len; + if (!rc && flash->async.len) { + /* + * We want to get called pretty much straight away, just have + * to be sure that we jump back out to Linux so that if this + * very long we don't cause RCU or the scheduler to freak + */ + schedule_timer(&flash->async.poller, msecs_to_tb(1)); + return; + } + + opal_queue_msg(OPAL_MSG_ASYNC_COMP, NULL, NULL, flash->async.token, rc); + flash_release(flash); +} + static struct dt_node *flash_add_dt_node(struct flash *flash, int id) { struct dt_node *flash_node; @@ -295,6 +360,7 @@ int flash_register(struct blocklevel_device *bl) flash->size = size; flash->block_size = block_size; flash->id = num_flashes(); + init_timer(&flash->async.poller, flash_poll, flash); list_add(&flashes, &flash->list); @@ -323,16 +389,11 @@ int flash_register(struct blocklevel_device *bl) return OPAL_SUCCESS; } -enum flash_op { - FLASH_OP_READ, - FLASH_OP_WRITE, - FLASH_OP_ERASE, -}; - static int64_t opal_flash_op(enum flash_op op, uint64_t id, uint64_t offset, uint64_t buf, uint64_t size, uint64_t token) { struct flash *flash = NULL; + uint64_t len; int rc; list_for_each(&flashes, flash, list) @@ -351,9 +412,16 @@ static int64_t opal_flash_op(enum flash_op op, uint64_t id, uint64_t offset, prlog(PR_DEBUG, "Requested flash op %d beyond flash size %" PRIu64 "\n", op, flash->size); rc = OPAL_PARAMETER; - goto err; + goto out; } + len = MIN(size, flash->block_size); + flash->async.op = op; + flash->async.token = token; + flash->async.buf = buf + len; + flash->async.len = size - len; + flash->async.pos = offset + len; + /* * These ops intentionally have no smarts (ecc correction or erase * before write) to them. @@ -363,29 +431,43 @@ static int64_t opal_flash_op(enum flash_op op, uint64_t id, uint64_t offset, */ switch (op) { case FLASH_OP_READ: - rc = blocklevel_raw_read(flash->bl, offset, (void *)buf, size); + rc = blocklevel_raw_read(flash->bl, offset, (void *)buf, len); break; case FLASH_OP_WRITE: - rc = blocklevel_raw_write(flash->bl, offset, (void *)buf, size); + rc = blocklevel_raw_write(flash->bl, offset, (void *)buf, len); break; case FLASH_OP_ERASE: - rc = blocklevel_erase(flash->bl, offset, size); + rc = blocklevel_erase(flash->bl, offset, len); break; default: assert(0); } if (rc) { + prlog(PR_ERR, "%s: Op %d failed with rc %d\n", __func__, op, rc); rc = OPAL_HARDWARE; - goto err; + goto out; } - flash_release(flash); - - opal_queue_msg(OPAL_MSG_ASYNC_COMP, NULL, NULL, token, rc); - return OPAL_ASYNC_COMPLETION; - -err: + if (size - len) { + /* Work remains */ + schedule_timer(&flash->async.poller, msecs_to_tb(1)); + /* Don't release the flash */ + return OPAL_ASYNC_COMPLETION; + } else { + /* + * As tempting as it might be here to return OPAL_SUCCESS + * here, don't! As of 1/07/2017 the powernv_flash driver in + * Linux will handle OPAL_SUCCESS as an error, the only thing + * that makes it handle things as though they're working is + * receiving OPAL_ASYNC_COMPLETION. + * + * XXX TODO: Revisit this in a few years *sigh* + */ + opal_queue_msg(OPAL_MSG_ASYNC_COMP, NULL, NULL, flash->async.token, rc); + } + rc = OPAL_ASYNC_COMPLETION; +out: flash_release(flash); return rc; } diff --git a/doc/opal-api/opal-flash-110-111-112.rst b/doc/opal-api/opal-flash-110-111-112.rst index 71ba866d..086c4095 100644 --- a/doc/opal-api/opal-flash-110-111-112.rst +++ b/doc/opal-api/opal-flash-110-111-112.rst @@ -20,6 +20,10 @@ success, the calls will return ``OPAL_ASYNC_COMPLETION``, and an opal_async_completion message will be sent (with the appropriate token argument) when the operation completes. +Due to an error in the powernv_flash driver in Linux these three OPAL +calls should never return ``OPAL_SUCCESS`` as the driver is likely to +treat this return value as an error. + All calls share the same return values: ``OPAL_ASYNC_COMPLETION``