Linux-Crypto Archive on lore.kernel.org
 help / color / Atom feed
From: Leonard Crestez <leonard.crestez@nxp.com>
To: Andrey Smirnov <andrew.smirnov@gmail.com>,
	Horia Geanta <horia.geanta@nxp.com>
Cc: "linux-crypto@vger.kernel.org" <linux-crypto@vger.kernel.org>,
	Chris Healy <cphealy@gmail.com>,
	Lucas Stach <l.stach@pengutronix.de>,
	Herbert Xu <herbert@gondor.apana.org.au>,
	Iuliana Prodan <iuliana.prodan@nxp.com>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Abel Vesa <abel.vesa@nxp.com>
Subject: Re: [PATCH 12/12] crypto: caam - change JR device ownership scheme
Date: Fri, 13 Sep 2019 19:16:03 +0000
Message-ID: <VI1PR04MB7023A7EC91599A537CB6A487EEB30@VI1PR04MB7023.eurprd04.prod.outlook.com> (raw)
In-Reply-To: <20190904023515.7107-13-andrew.smirnov@gmail.com>

On 04.09.2019 05:35, Andrey Smirnov wrote:
> Returning -EBUSY from platform device's .remove() callback won't stop
> the removal process, so the code in caam_jr_remove() is not going to
> have the desired effect of preventing JR from being removed.
> 
> In order to be able to deal with removal of the JR device, change the
> code as follows:
> 
>    1. To make sure that underlying struct device remains valid for as
>       long as we have a reference to it, add appropriate device
>       refcount management to caam_jr_alloc() and caam_jr_free()
> 
>    2. To make sure that device removal doesn't happen in parallel to
>        use using the device in caam_jr_enqueue() augment the latter to
>        acquire/release device lock for the duration of the subroutine
> 
>    3. In order to handle the case when caam_jr_enqueue() is executed
>       right after corresponding caam_jr_remove(), add code to check
>       that driver data has not been set to NULL.

What happens if a driver is unbound while a transform is executing 
asynchronously?

Doing get_device and put_device in caam_jr_alloc and caam_jr_free 
doesn't protect you against an explicit unbind of caam_jr, that path 
doesn't care about device reference counts. For example on imx6qp:

# echo 2102000.jr1 > /sys/bus/platform/drivers/caam_jr/unbind

CPU: 2 PID: 561 Comm: bash Not tainted 5.3.0-rc7-next-20190904
Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
Backtrace:
[<c010f8a4>] (dump_backtrace) from [<c010fc5c>] (show_stack+0x20/0x24)
[<c010fc3c>] (show_stack) from [<c0d21600>] (dump_stack+0xdc/0x114)
[<c0d21524>] (dump_stack) from [<c0955774>] (caam_jr_remove+0x24/0xf8)
[<c0955750>] (caam_jr_remove) from [<c06e2d98>] 
(platform_drv_remove+0x34/0x4c)
[<c06e2d64>] (platform_drv_remove) from [<c06e0b74>] 
(device_release_driver_internal+0xf8/0x1b0)
[<c06e0a7c>] (device_release_driver_internal) from [<c06e0c70>] 
(device_driver_detach+0x20/0x24)
[<c06e0c50>] (device_driver_detach) from [<c06de5a4>] 
(unbind_store+0x6c/0xe0)
[<c06de538>] (unbind_store) from [<c06dd950>] (drv_attr_store+0x30/0x3c)
[<c06dd920>] (drv_attr_store) from [<c0364c18>] (sysfs_kf_write+0x50/0x5c)
[<c0364bc8>] (sysfs_kf_write) from [<c0363e0c>] 
(kernfs_fop_write+0x10c/0x1f8)
[<c0363d00>] (kernfs_fop_write) from [<c02c3a30>] (__vfs_write+0x44/0x1d0)
[<c02c39ec>] (__vfs_write) from [<c02c68ec>] (vfs_write+0xb0/0x194)
[<c02c683c>] (vfs_write) from [<c02c6b7c>] (ksys_write+0x64/0xd8)
[<c02c6b18>] (ksys_write) from [<c02c6c08>] (sys_write+0x18/0x1c)
[<c02c6bf0>] (sys_write) from [<c0101000>] (ret_fast_syscall+0x0/0x28)

I think you need to do some form of slow wait loop in jrpriv until 
jrpriv->tfm_count reaches zero. Preventing new operations from starting 
would help but isn't strictly necessary for correctness.

> diff --git a/drivers/crypto/caam/jr.c b/drivers/crypto/caam/jr.c
> index 47b389cb1c62..8a30bbd7f2aa 100644
> --- a/drivers/crypto/caam/jr.c
> +++ b/drivers/crypto/caam/jr.c
> @@ -124,14 +124,6 @@ static int caam_jr_remove(struct platform_device *pdev)
>   	jrdev = &pdev->dev;
>   	jrpriv = dev_get_drvdata(jrdev);
>   
> -	/*
> -	 * Return EBUSY if job ring already allocated.
> -	 */
> -	if (atomic_read(&jrpriv->tfm_count)) {
> -		dev_err(jrdev, "Device is busy\n");
> -		return -EBUSY;
> -	}
> -
>   	/* Unregister JR-based RNG & crypto algorithms */
>   	unregister_algs();
>   
> @@ -300,7 +292,7 @@ struct device *caam_jr_alloc(void)
>   
>   	if (min_jrpriv) {
>   		atomic_inc(&min_jrpriv->tfm_count);
> -		dev = min_jrpriv->dev;
> +		dev = get_device(min_jrpriv->dev);
>   	}
>   	spin_unlock(&driver_data.jr_alloc_lock);
>   
> @@ -318,13 +310,16 @@ void caam_jr_free(struct device *rdev)
>   	struct caam_drv_private_jr *jrpriv = dev_get_drvdata(rdev);
>   
>   	atomic_dec(&jrpriv->tfm_count);
> +	put_device(rdev);
>   }
>   EXPORT_SYMBOL(caam_jr_free);
>   
>   /**
>    * caam_jr_enqueue() - Enqueue a job descriptor head. Returns 0 if OK,
>    * -EBUSY if the queue is full, -EIO if it cannot map the caller's
> - * descriptor.
> + * descriptor, -ENODEV if given device was removed and is no longer
> + * valid
> + *
>    * @dev:  device of the job ring to be used. This device should have
>    *        been assigned prior by caam_jr_register().
>    * @desc: points to a job descriptor that execute our request. All
> @@ -354,15 +349,32 @@ int caam_jr_enqueue(struct device *dev, u32 *desc,
>   				u32 status, void *areq),
>   		    void *areq)
>   {
> -	struct caam_drv_private_jr *jrp = dev_get_drvdata(dev);
> +	struct caam_drv_private_jr *jrp;
>   	struct caam_jrentry_info *head_entry;
>   	int head, tail, desc_size;
>   	dma_addr_t desc_dma;
>   
> +	/*
> +	 * Lock the device to prevent it from being removed while we
> +	 * are using it
> +	 */
> +	device_lock(dev);
> +
> +	/*
> +	 * If driver data is NULL, it is very likely that this device
> +	 * was removed already. Nothing we can do here but bail out.
> +	 */
> +	jrp = dev_get_drvdata(dev);
> +	if (!jrp) {
> +		device_unlock(dev);
> +		return -ENODEV;
> +	}
> +
>   	desc_size = (caam32_to_cpu(*desc) & HDR_JD_LENGTH_MASK) * sizeof(u32);
>   	desc_dma = dma_map_single(dev, desc, desc_size, DMA_TO_DEVICE);
>   	if (dma_mapping_error(dev, desc_dma)) {
>   		dev_err(dev, "caam_jr_enqueue(): can't map jobdesc\n");
> +		device_unlock(dev);
>   		return -EIO;
>   	}
>   
> @@ -375,6 +387,7 @@ int caam_jr_enqueue(struct device *dev, u32 *desc,
>   	    CIRC_SPACE(head, tail, JOBR_DEPTH) <= 0) {
>   		spin_unlock_bh(&jrp->inplock);
>   		dma_unmap_single(dev, desc_dma, desc_size, DMA_TO_DEVICE);
> +		device_unlock(dev);
>   		return -EBUSY;
>   	}
>   
> @@ -411,6 +424,7 @@ int caam_jr_enqueue(struct device *dev, u32 *desc,
>   		jrp->inpring_avail = rd_reg32(&jrp->rregs->inpring_avail);
>   
>   	spin_unlock_bh(&jrp->inplock);
> +	device_unlock(dev);
>   
>   	return 0;
>   }

  reply index

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-09-04  2:35 [PATCH 00/12] CAAM bugfixes, small improvements Andrey Smirnov
2019-09-04  2:35 ` [PATCH 01/12] crypto: caam - make sure clocks are enabled first Andrey Smirnov
2019-09-06 11:18   ` Horia Geanta
2019-09-09  7:21     ` Herbert Xu
2019-09-09  7:22       ` Herbert Xu
2019-09-04  2:35 ` [PATCH 02/12] crypto: caam - use devres to unmap JR's registers Andrey Smirnov
2019-09-04  2:43   ` Fabio Estevam
2019-09-04  2:55     ` Andrey Smirnov
2019-09-09 13:01   ` Horia Geanta
2019-09-04  2:35 ` [PATCH 03/12] crypto: caam - check irq_of_parse_and_map for errors Andrey Smirnov
2019-09-06 12:29   ` Horia Geanta
2019-09-04  2:35 ` [PATCH 04/12] crypto: caam - dispose of IRQ mapping only after IRQ is freed Andrey Smirnov
2019-09-06 12:26   ` Horia Geanta
2019-09-09  7:46   ` crypto: caam - Cast to long first before pointer conversion Herbert Xu
2019-09-09 11:06     ` Horia Geanta
2019-09-09 13:55     ` [v2 PATCH] " Herbert Xu
2019-09-04  2:35 ` [PATCH 05/12] crypto: caam - use devres to unmap memory Andrey Smirnov
2019-09-09 13:20   ` Horia Geanta
2019-09-04  2:35 ` [PATCH 06/12] crypto: caam - use devres to remove debugfs Andrey Smirnov
2019-09-09 13:25   ` Horia Geanta
2019-09-04  2:35 ` [PATCH 07/12] crypto: caam - use devres to de-initialize the RNG Andrey Smirnov
2019-09-09 15:39   ` Horia Geanta
2019-09-18  6:06     ` Andrey Smirnov
2019-09-04  2:35 ` [PATCH 08/12] crypto: caam - use devres to de-initialize QI Andrey Smirnov
2019-09-04  2:35 ` [PATCH 09/12] crypto: caam - user devres to populate platform devices Andrey Smirnov
2019-09-04  2:35 ` [PATCH 10/12] crypto: caam - populate platform devices last Andrey Smirnov
2019-09-04  2:35 ` [PATCH 11/12] crypto: caam - convert caamrng to platform device Andrey Smirnov
2019-09-11  9:35   ` [PATCH] crypto: caam - use the same jr for rng init/exit Horia Geanta
2019-09-18  6:01     ` Andrey Smirnov
2019-09-04  2:35 ` [PATCH 12/12] crypto: caam - change JR device ownership scheme Andrey Smirnov
2019-09-13 19:16   ` Leonard Crestez [this message]
2019-09-18  3:13     ` Andrey Smirnov
2019-09-09  7:53 ` [PATCH 00/12] CAAM bugfixes, small improvements Herbert Xu
2019-09-09 12:07   ` Horia Geanta
2019-09-09 12:52     ` Herbert Xu
2019-09-09 13:26       ` Horia Geanta
2019-09-09 13:52         ` Herbert Xu

Reply instructions:

You may reply publically 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=VI1PR04MB7023A7EC91599A537CB6A487EEB30@VI1PR04MB7023.eurprd04.prod.outlook.com \
    --to=leonard.crestez@nxp.com \
    --cc=abel.vesa@nxp.com \
    --cc=andrew.smirnov@gmail.com \
    --cc=cphealy@gmail.com \
    --cc=herbert@gondor.apana.org.au \
    --cc=horia.geanta@nxp.com \
    --cc=iuliana.prodan@nxp.com \
    --cc=l.stach@pengutronix.de \
    --cc=linux-crypto@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    /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

Linux-Crypto Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-crypto/0 linux-crypto/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-crypto linux-crypto/ https://lore.kernel.org/linux-crypto \
		linux-crypto@vger.kernel.org linux-crypto@archiver.kernel.org
	public-inbox-index linux-crypto


Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-crypto


AGPL code for this site: git clone https://public-inbox.org/ public-inbox