linux-kernel - [PATCH 7/9] nvme-apple: Serialize command issue

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 21 Mar 2022 17:50:47 +0100
From:   Sven Peter <sven@...npeter.dev>
To:     Keith Busch <kbusch@...nel.org>, Jens Axboe <axboe@...com>,
        Christoph Hellwig <hch@....de>,
        Sagi Grimberg <sagi@...mberg.me>
Cc:     Sven Peter <sven@...npeter.dev>, Hector Martin <marcan@...can.st>,
        Alyssa Rosenzweig <alyssa@...enzweig.io>,
        Rob Herring <robh+dt@...nel.org>,
        Arnd Bergmann <arnd@...db.de>, Marc Zyngier <maz@...nel.org>,
        devicetree@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
        linux-kernel@...r.kernel.org, linux-nvme@...ts.infradead.org,
        Jens Axboe <axboe@...nel.dk>
Subject: [PATCH 7/9] nvme-apple: Serialize command issue

From: Jens Axboe <axboe@...nel.dk>

This controller shouldn't need serialization of command issue since
the SQ is replaced by a simple array and commands are issued by writing
the array index to a MMIO register.
Without serialization however sometimes commands are still executed
correctly and appear in the CQ but never trigger an interrupt.

Signed-off-by: Jens Axboe <axboe@...nel.dk>
[sven: added our best guess why this needs to be done]
Signed-off-by: Sven Peter <sven@...npeter.dev>
---
 drivers/nvme/host/apple.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/drivers/nvme/host/apple.c b/drivers/nvme/host/apple.c
index 587d6c7014a0..a4193429564e 100644
--- a/drivers/nvme/host/apple.c
+++ b/drivers/nvme/host/apple.c
@@ -292,6 +292,7 @@ static void apple_nvmmu_inval(struct apple_nvme_queue *q, unsigned int tag)
 static void apple_nvme_submit_cmd(struct apple_nvme_queue *q,
 				  struct nvme_command *cmd)
 {
+	struct apple_nvme *anv = queue_to_apple_nvme(q);
 	u32 tag = nvme_tag_from_cid(cmd->common.command_id);
 	struct apple_nvmmu_tcb *tcb = &q->tcbs[tag];
 
@@ -308,7 +309,18 @@ static void apple_nvme_submit_cmd(struct apple_nvme_queue *q,
 		tcb->dma_flags |= APPLE_ANS_TCB_DMA_FROM_DEVICE;
 
 	memcpy(&q->sqes[tag], cmd, sizeof(*cmd));
+
+	/*
+	 * This lock here doesn't make much sense at a first glace but
+	 * removing it will result in occasional missed completetion
+	 * interrupts even though the commands still appear on the CQ.
+	 * It's unclear why this happens but our best guess is that
+	 * there is a bug in the firmware triggered when a write to the
+	 * CQ and the SQ happen simultaneously.
+	 */
+	spin_lock_irq(&anv->lock);
 	writel(tag, q->sq_db);
+	spin_unlock_irq(&anv->lock);
 }
 
 /*
-- 
2.25.1