iommu/amd: Do not re-fetch iommu->cmd_buf_tail

The compiler is not smart enough to realize that iommu->cmd_buf_tail
can't be modified across memcpy:

41 8b 45 74          mov    0x74(%r13),%eax   # iommu->cmd_buf_tail
44 8d 78 10          lea    0x10(%rax),%r15d  # += sizeof(*cmd)
41 81 e7 ff 1f 00 00 and    $0x1fff,%r15d     # %= CMD_BUFFER_SIZE
49 03 45 68          add    0x68(%r13),%rax   # target = iommu->cmd_buf + iommu->cmd_buf_tail
45 89 7d 74          mov    %r15d,0x74(%r13)  # store to iommu->cmd_buf_tail
49 8b 34 24          mov    (%r12),%rsi       # memcpy
49 8b 7c 24 08       mov    0x8(%r12),%rdi    # memcpy
48 89 30             mov    %rsi,(%rax)       # memcpy
48 89 78 08          mov    %rdi,0x8(%rax)    # memcpy
49 8b 55 38          mov    0x38(%r13),%rdx   # iommu->mmio_base
41 8b 45 74          mov    0x74(%r13),%eax   # redundant load of iommu->cmd_buf_tail
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
89 82 08 20 00 00    mov    %eax,0x2008(%rdx) # writel

CC: Tom Lendacky <thomas.lendacky@amd.com>
CC: Joerg Roedel <jroedel@suse.de>
CC: linux-kernel@vger.kernel.org
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>

drivers/iommu/amd_iommu.c

@@ -852,17 +852,18 @@ static void copy_cmd_to_buffer(struct amd_iommu *iommu,
 			       struct iommu_cmd *cmd)
 {
 	u8 *target;
-
-	target = iommu->cmd_buf + iommu->cmd_buf_tail;
-
-	iommu->cmd_buf_tail += sizeof(*cmd);
-	iommu->cmd_buf_tail %= CMD_BUFFER_SIZE;
+	u32 tail;
 
 	/* Copy command to buffer */
+	tail = iommu->cmd_buf_tail;
+	target = iommu->cmd_buf + tail;
 	memcpy(target, cmd, sizeof(*cmd));
 
+	tail = (tail + sizeof(*cmd)) % CMD_BUFFER_SIZE;
+	iommu->cmd_buf_tail = tail;
+
 	/* Tell the IOMMU about it */
-	writel(iommu->cmd_buf_tail, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
+	writel(tail, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
 }
 
 static void build_completion_wait(struct iommu_cmd *cmd, u64 address)