powerpc/powernv: Fix reading of OPAL msglog

memory_return_from_buffer returns a signed value, so ret should be
ssize_t.

Fixes the following issue reported by David Binderman:

  [linux-3.15/arch/powerpc/platforms/powernv/opal-msglog.c:65]: (style)
  Checking if unsigned variable 'ret' is less than zero.
  [linux-3.15/arch/powerpc/platforms/powernv/opal-msglog.c:82]: (style)
  Checking if unsigned variable 'ret' is less than zero.

  Local variable "ret" is of type size_t. This is always unsigned,
  so it is pointless to check if it is less than zero.

  https://bugzilla.kernel.org/show_bug.cgi?id=77551

Fixing this exposes a real bug for the case where the entire count
bytes is successfully read from the POS_WRAP case. The second
memory_read_from_buffer will return EINVAL, causing the entire read to
return EINVAL to userspace, despite the data being copied correctly. The
fix is to test for the case where the data has been read and return
early.

Signed-off-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

diff --git a/arch/powerpc/platforms/powernv/opal-msglog.c b/arch/powerpc/platforms/powernv/opal-msglog.c
index 1bb25b9525046f14abfa2e38c2db188b0686ff94..44ed78af1a0dd53bb9e33f16b3ce465ffc5c6f4f 100644 (file)
--- a/arch/powerpc/platforms/powernv/opal-msglog.c
+++ b/arch/powerpc/platforms/powernv/opal-msglog.c
@@ -37,7 +37,8 @@ static ssize_t opal_msglog_read(struct file *file, struct kobject *kobj,
 {
        struct memcons *mc = bin_attr->private;
        const char *conbuf;
-       size_t ret, first_read = 0;
+       ssize_t ret;
+       size_t first_read = 0;
        uint32_t out_pos, avail;
 
        if (!mc)
@@ -69,6 +70,9 @@ static ssize_t opal_msglog_read(struct file *file, struct kobject *kobj,
                to += first_read;
                count -= first_read;
                pos -= avail;
+
+               if (count <= 0)
+                       goto out;
        }
 
        /* Sanity check. The firmware should not do this to us. */