Re: [PATCH] rumpkernel dependencies

[Samuel Thibault]

> diff --git a/libmachdevrump/block.c b/libmachdevrump/block.c
> new file mode 100644
> index 000000000..e4b519bb8
> --- /dev/null
> +++ b/libmachdevrump/block.c
> +static io_return_t
> +device_open (mach_port_t reply_port, mach_msg_type_name_t reply_port_type,
> +          dev_mode_t mode, char *name, device_t *devp,
> +          mach_msg_type_name_t *devicePoly)
> +{
> +  io_return_t err = D_SUCCESS;
> +  struct block_data *bd = NULL;
> +  char *dev_name;
> +  off_t media_size;
> +  uint32_t block_size;
> +
> +  mach_print("device open\n");
> +  dev_name = translate_name (name);

Are these mach_print calls still needed for your debugging?  I'm fine
with committing some of them for now, but perhaps they are now useless
already?

> diff --git a/libmachdevrump/dev_hdr.h b/libmachdevrump/dev_hdr.h
> new file mode 100644
> index 000000000..79edc43a3
> --- /dev/null
> +++ b/libmachdevrump/dev_hdr.h

Most of this doesn't seem to be used? Only this:

+/* This structure is associated with each open device port.
+ * The port representing the device points to this structure.  */
+struct emul_device
+{
+    struct device_emulation_ops *emul_ops;
+    void *emul_data;
+};
+
+typedef struct emul_device *emul_device_t;
+
+#define DEVICE_NULL     ((device_t) 0)
+
+/*
+ * Generic device header.  May be allocated with the device,
+ * or built when the device is opened.
+ */
+struct mach_device {
+       struct port_info port;
+       struct emul_device      dev;            /* the real device structure */
+};
+typedef        struct mach_device *mach_device_t;
+#define        MACH_DEVICE_NULL ((mach_device_t)0)




I see that most of the patch is actually coming from the incubator's
dde-based libmachdev. I'd rather avoid having two copies of that code in
different Hurd repos :)

I'd say what we want to do is:

- put the files common to dde-based and rump-based libmachdev to
  libmachdev/.  That includes basically all your files except disk.c,
  machdevrump.c and Makefile. The only real discrepancy I see is in
  ds_routines.c' call to l4dde26_process_from_ddekit, but we can make
  libmachdevdde provide a function that does that call then call
  ds_server(). AIUI the call to is_master_device() should indeed be made
  like in the incubator?  I don't think we should deallocate it at the
  very least, was it really needed in your tests?

- move the dde-based files (for network) to a libmachdevdde/

- put your rump-based disk.c to a libmachdevrump/

and the rumpdisk daemon will just need to link in both libmachdev and
libmachdevrump.

Samuel