On 2021/3/3 15:54, Viresh Kumar wrote:
On 01-03-21, 14:41, Jie Deng wrote:
diff --git a/drivers/i2c/busses/i2c-virtio.c b/drivers/i2c/busses/i2c-virtio.c
+static int virtio_i2c_send_reqs(struct virtqueue *vq,
+ struct virtio_i2c_req *reqs,
+ struct i2c_msg *msgs, int nr)
+{
+ struct scatterlist *sgs[3], out_hdr, msg_buf, in_hdr;
+ int i, outcnt, incnt, err = 0;
+ u8 *buf;
+
+ for (i = 0; i < nr; i++) {
+ if (!msgs[i].len)
+ break;
+
+ reqs[i].out_hdr.addr = cpu_to_le16(msgs[i].addr << 1);
+
+ if (i != nr - 1)
+ reqs[i].out_hdr.flags |= VIRTIO_I2C_FLAGS_FAIL_NEXT;
+
+ outcnt = incnt = 0;
+ sg_init_one(&out_hdr, &reqs[i].out_hdr, sizeof(reqs[i].out_hdr));
+ sgs[outcnt++] = &out_hdr;
+
+ buf = kzalloc(msgs[i].len, GFP_KERNEL);
+ if (!buf)
+ break;
+
+ if (msgs[i].flags & I2C_M_RD) {
+ reqs[i].read_buf = buf;
+ sg_init_one(&msg_buf, reqs[i].read_buf, msgs[i].len);
+ sgs[outcnt + incnt++] = &msg_buf;
+ } else {
+ reqs[i].write_buf = buf;
+ memcpy(reqs[i].write_buf, msgs[i].buf, msgs[i].len);
+ sg_init_one(&msg_buf, reqs[i].write_buf, msgs[i].len);
+ sgs[outcnt++] = &msg_buf;
+ }
+
+ sg_init_one(&in_hdr, &reqs[i].in_hdr, sizeof(reqs[i].in_hdr));
+ sgs[outcnt + incnt++] = &in_hdr;
+
+ err = virtqueue_add_sgs(vq, sgs, outcnt, incnt, &reqs[i], GFP_KERNEL);
+ if (err < 0) {
+ pr_err("failed to add msg[%d] to virtqueue.\n", i);
+ if (msgs[i].flags & I2C_M_RD) {
+ kfree(reqs[i].read_buf);
+ reqs[i].read_buf = NULL;
+ } else {
+ kfree(reqs[i].write_buf);
+ reqs[i].write_buf = NULL;
+ }
+ break;
+ }
+ }
+
+ return i;
+}
diff --git a/include/uapi/linux/virtio_i2c.h b/include/uapi/linux/virtio_i2c.h
+/**
+ * struct virtio_i2c_out_hdr - the virtio I2C message OUT header
+ * @addr: the controlled device address
+ * @padding: used to pad to full dword
+ * @flags: used for feature extensibility
+ */
+struct virtio_i2c_out_hdr {
+ __le16 addr;
+ __le16 padding;
+ __le32 flags;
+};
Both this code and the virtio spec (which is already merged) are
missing msgs[i].flags and they are never sent to backend. The only
flags available here are the ones defined by virtio spec and these are
not i2c flags.
I also looked at your i2c backend for acrn and it mistakenly copies
the hdr.flag, which is the virtio flag and not i2c flag.
https://github.com/projectacrn/acrn-hypervisor/blob/master/devicemodel/hw/pci/virtio/virtio_i2c.c#L539
I will send a fix for the specs if you agree that there is a problem
here.
what am I missing here ? This should have been caught in your testing
and so I feel I must be missing something.
This is not a problem. My original proposal was to mirror the struct
i2c_msg.
The code you looked at was based on that.
However, the virtio TC prefer not to mirror it. They have some concerns.
For example, there is a bit I2C_M_RD in i2c_msg.flag which has the same
meaning with
the R/W in virtio descriptor. This is a repetition which may cause problems.
So the virtio_i2c_out_hdr.flags is used to instead of i2c_msg.flags for
extension.
You can check this link
https://github.com/oasis-tcs/virtio-spec/issues/88 to learn the whole story.
There are discussions about the spec (v1 ~ v7).
I'm updating these drivers step by step to make sure they finally follow
the spec.
