tpm_crb: fix crb_req_canceled behavior
authorJarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Fri, 2 Sep 2016 19:34:17 +0000 (22:34 +0300)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 16 Oct 2016 15:36:15 +0000 (17:36 +0200)
commit 72fd50e14e46dc0edf360631bdece87c2f066a97 upstream.

The req_canceled() callback is used by tpm_transmit() periodically to
check whether the request has been canceled while it is receiving a
response from the TPM.

The TPM_CRB_CTRL_CANCEL register was cleared already in the crb_cancel
callback, which has two consequences:

* Cancel might not happen.
* req_canceled() always returns zero.

A better place to clear the register is when starting to send a new
command. The behavior of TPM_CRB_CTRL_CANCEL is described in the
section 5.5.3.6 of the PTP specification.

Fixes: 30fc8d138e91 ("tpm: TPM 2.0 CRB Interface")
Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/char/tpm/tpm_crb.c

index 61e64293b7654b36d5db6b320d1bc78100379eb3..2b21398c3adcfc4bd5aaa81f7fcc6a980b75834e 100644 (file)
@@ -149,6 +149,11 @@ static int crb_send(struct tpm_chip *chip, u8 *buf, size_t len)
        struct crb_priv *priv = chip->vendor.priv;
        int rc = 0;
 
+       /* Zero the cancel register so that the next command will not get
+        * canceled.
+        */
+       iowrite32(0, &priv->cca->cancel);
+
        if (len > le32_to_cpu(ioread32(&priv->cca->cmd_size))) {
                dev_err(&chip->dev,
                        "invalid command count value %x %zx\n",
@@ -182,8 +187,6 @@ static void crb_cancel(struct tpm_chip *chip)
 
        if ((priv->flags & CRB_FL_ACPI_START) && crb_do_acpi_start(chip))
                dev_err(&chip->dev, "ACPI Start failed\n");
-
-       iowrite32(0, &priv->cca->cancel);
 }
 
 static bool crb_req_canceled(struct tpm_chip *chip, u8 status)