Re: [Qemu-devel] [PATCH v2 2/3] hw/sd: model a power-up delay, as a workaround for an EDK2 bug

[Andrew Baumann]

> From: Peter Crosthwaite [mailto:address@hidden
> Sent: Monday, 21 December 2015 14:57
> On Mon, Dec 21, 2015 at 2:25 PM, Andrew Baumann
> <address@hidden> wrote:
> >> From: qemu-devel-
> address@hidden
> >> [mailto:qemu-devel-
> >> address@hidden On Behalf Of
> >> Peter Crosthwaite
> >> Sent: Monday, 21 December 2015 13:46
> >> On Wed, Dec 16, 2015 at 11:02 AM, Andrew Baumann
> >> <address@hidden> wrote:
> >> > The SD spec for ACMD41 says that a zero argument is an "inquiry"
> >> > ACMD41, which does not start initialisation and is used only for
> >> > retrieving the OCR. However, Tianocore EDK2 (UEFI) has a bug [1]: it
> >> > first sends an inquiry (zero) ACMD41. If that first request returns an
> >> > OCR value with the power up bit (0x80000000) set, it assumes the card
> >> > is ready and continues, leaving the card in the wrong state. (My
> >> > assumption is that this works on hardware, because no real card is
> >> > immediately powered up upon reset.)
> >> >
> >> > This change models a delay of 0.5ms from the first ACMD41 to the
> power
> >> > being up. However, it also immediately sets the power on upon seeing a
> >> > non-zero (non-enquiry) ACMD41. This speeds up UEFI boot; it should
> >> > also account for guests that simply delay after card reset and then
> >> > issue an ACMD41 that they expect will succeed.
> > [...]
> >> > @@ -449,6 +462,8 @@ static const VMStateDescription sd_vmstate = {
> >> >      .fields = (VMStateField[]) {
> >> >          VMSTATE_UINT32(mode, SDState),
> >> >          VMSTATE_INT32(state, SDState),
> >> > +        VMSTATE_UINT32(ocr, SDState),
> >> > +        VMSTATE_TIMER_PTR(ocr_power_timer, SDState),
> >>
> >> If you change the VMSTATE layout, you need to bump the version. As
> >> this is very common code, it may have stricter version bump
> >> requirements. Last I knew however, there was a way to add new fields
> >> at the end of VMSD without breaking backwards compatibility. Peter or
> >> Juan may know more.
> >
> > I'll admit that I didn't think about these issues when adding the fields, 
> > but
> after a (quick) look at vmstate_save_state() and vmstate_load_state(), they
> seem to be using named fields in a json format, so I don't think the order of
> fields should matter if we are adding new ones. However, if we want to be
> able to migrate sd instances across across this change, then we'll need to
> arrange for the OCR to appear already powered-on if we're coming from a
> previous version. Does VMState have a way to do that? Essentially I just
> need to specify a default value for the ocr field if coming from an old 
> vmstate
> version <= 1 that differs from the value set in sd_reset().
> >
> 
> You can open code post_load logic as a callback, and I think you have
> access to the image version from there.

So, how about something like this:

+static int sd_vmstate_post_load(void *opaque, int version_id)
+{
+    SDState *sd = opaque;
+
+    if (version_id < 2) {
+        /* prior versions did not save the OCR or model a power up
+         * delay, so we need to mirror this state */
+        sd_ocr_powerup(sd);
+    }
+
+    return 0;
+}
+
 static const VMStateDescription sd_vmstate = {
     .name = "sd-card",
-    .version_id = 1,
+    .version_id = 2,
     .minimum_version_id = 1,
+    .post_load = sd_vmstate_post_load,
     .fields = (VMStateField[]) {

Any idea how I can easily test this? Sorry, I am not at all familiar with the 
vmstate/migration stuff. Is there a good/known working device model to test 
that uses SD? I guess I just do savevm on the old one and loadvm on the new one?

Thanks,
Andrew