[Qemu-devel] [PATCH 0/2] hw/sd: Fix medium change

[Qemu-devel] [PATCH 0/2] hw/sd: Fix medium change

[Max Reitz]

Commit de2c6c0536c5c5ebb6e0ce7dcfd8fa9476edab52 broke changing mediums
in SD drives; this series fixes that, and the fact that
blockdev-open-tray did not do anything on SD drives so far.


Max Reitz (2):
  hw/sd: Add medium insertion status
  hw/sd: Implement is_tray_open() BlockDevOp

 hw/sd/sd.c | 23 +++++++++++++++++------
 1 file changed, 17 insertions(+), 6 deletions(-)

-- 
2.6.4

[Qemu-devel] [PATCH 1/2] hw/sd: Add medium insertion status

[Max Reitz]

Right now, the change_media_cb (sd_cardchange()) completely ignores its
@load parameter. This means that issuing a blockdev-open-tray command
will actually not have any effect.

Fix this by keeping track of the medium insertion status in the SDState
and updating it in sd_init() and sd_cardchange().

Cc: qemu-stable <qemu-stable@nongnu.org>
Signed-off-by: Max Reitz <mreitz@redhat.com>
---
 hw/sd/sd.c | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/hw/sd/sd.c b/hw/sd/sd.c
index 1a9935c..0751ba2 100644
--- a/hw/sd/sd.c
+++ b/hw/sd/sd.c
@@ -114,6 +114,7 @@ struct SDState {
     uint8_t *buf;
 
     bool enable;
+    bool medium_inserted;
 };
 
 static void sd_set_mode(SDState *sd)
@@ -429,8 +430,10 @@ static void sd_cardchange(void *opaque, bool load)
 {
     SDState *sd = opaque;
 
-    qemu_set_irq(sd->inserted_cb, blk_is_inserted(sd->blk));
-    if (blk_is_inserted(sd->blk)) {
+    sd->medium_inserted = load && blk_is_inserted(sd->blk);
+
+    qemu_set_irq(sd->inserted_cb, sd->medium_inserted);
+    if (sd->medium_inserted) {
         sd_reset(sd);
         qemu_set_irq(sd->readonly_cb, sd->wp_switch);
     }
@@ -497,6 +500,7 @@ SDState *sd_init(BlockBackend *blk, bool is_spi)
         /* FIXME ignoring blk_attach_dev() failure is dangerously brittle */
         blk_attach_dev(sd->blk, sd);
         blk_set_dev_ops(sd->blk, &sd_block_ops, sd);
+        sd->medium_inserted = blk_is_inserted(sd->blk);
     }
     vmstate_register(NULL, -1, &sd_vmstate, sd);
     return sd;
@@ -507,7 +511,7 @@ void sd_set_cb(SDState *sd, qemu_irq readonly, qemu_irq insert)
     sd->readonly_cb = readonly;
     sd->inserted_cb = insert;
     qemu_set_irq(readonly, sd->blk ? blk_is_read_only(sd->blk) : 0);
-    qemu_set_irq(insert, sd->blk ? blk_is_inserted(sd->blk) : 0);
+    qemu_set_irq(insert, sd->blk ? sd->medium_inserted : 0);
 }
 
 static void sd_erase(SDState *sd)
@@ -1349,7 +1353,7 @@ int sd_do_command(SDState *sd, SDRequest *req,
     sd_rsp_type_t rtype;
     int rsplen;
 
-    if (!sd->blk || !blk_is_inserted(sd->blk) || !sd->enable) {
+    if (!sd->blk || !sd->medium_inserted || !sd->enable) {
         return 0;
     }
 
@@ -1517,7 +1521,7 @@ void sd_write_data(SDState *sd, uint8_t value)
 {
     int i;
 
-    if (!sd->blk || !blk_is_inserted(sd->blk) || !sd->enable)
+    if (!sd->blk || !sd->medium_inserted || !sd->enable)
         return;
 
     if (sd->state != sd_receivingdata_state) {
@@ -1643,7 +1647,7 @@ uint8_t sd_read_data(SDState *sd)
     uint8_t ret;
     int io_len;
 
-    if (!sd->blk || !blk_is_inserted(sd->blk) || !sd->enable)
+    if (!sd->blk || !sd->medium_inserted || !sd->enable)
         return 0x00;
 
     if (sd->state != sd_sendingdata_state) {
-- 
2.6.4

[Qemu-devel] [PATCH 2/2] hw/sd: Implement is_tray_open() BlockDevOp

[Max Reitz]

This makes the change HMP/QMP command on SD devices work again, after it
has been broken in commit de2c6c0536c5c5ebb6e0ce7dcfd8fa9476edab52.

Reported-by: Peter Maydell <peter.maydell@linaro.org>
Cc: qemu-stable <qemu-stable@nongnu.org>
Signed-off-by: Max Reitz <mreitz@redhat.com>
---
 hw/sd/sd.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/hw/sd/sd.c b/hw/sd/sd.c
index 0751ba2..7d1e1cb 100644
--- a/hw/sd/sd.c
+++ b/hw/sd/sd.c
@@ -439,8 +439,15 @@ static void sd_cardchange(void *opaque, bool load)
     }
 }
 
+static bool sd_is_tray_open(void *opaque)
+{
+    SDState *sd = opaque;
+    return !sd->medium_inserted;
+}
+
 static const BlockDevOps sd_block_ops = {
     .change_media_cb = sd_cardchange,
+    .is_tray_open = sd_is_tray_open,
 };
 
 static const VMStateDescription sd_vmstate = {
-- 
2.6.4

Re: [Qemu-devel] [PATCH 1/2] hw/sd: Add medium insertion status

[John Snow]

On 01/07/2016 04:03 PM, Max Reitz wrote:
> Right now, the change_media_cb (sd_cardchange()) completely ignores its
> @load parameter. This means that issuing a blockdev-open-tray command
> will actually not have any effect.
> 
> Fix this by keeping track of the medium insertion status in the SDState
> and updating it in sd_init() and sd_cardchange().
> 
> Cc: qemu-stable <qemu-stable@nongnu.org>
> Signed-off-by: Max Reitz <mreitz@redhat.com>
> ---
>  hw/sd/sd.c | 16 ++++++++++------
>  1 file changed, 10 insertions(+), 6 deletions(-)
> 
> diff --git a/hw/sd/sd.c b/hw/sd/sd.c
> index 1a9935c..0751ba2 100644
> --- a/hw/sd/sd.c
> +++ b/hw/sd/sd.c
> @@ -114,6 +114,7 @@ struct SDState {
>      uint8_t *buf;
>  
>      bool enable;
> +    bool medium_inserted;
>  };
>  
>  static void sd_set_mode(SDState *sd)
> @@ -429,8 +430,10 @@ static void sd_cardchange(void *opaque, bool load)
>  {
>      SDState *sd = opaque;
>  
> -    qemu_set_irq(sd->inserted_cb, blk_is_inserted(sd->blk));
> -    if (blk_is_inserted(sd->blk)) {
> +    sd->medium_inserted = load && blk_is_inserted(sd->blk);
> +
> +    qemu_set_irq(sd->inserted_cb, sd->medium_inserted);
> +    if (sd->medium_inserted) {
>          sd_reset(sd);
>          qemu_set_irq(sd->readonly_cb, sd->wp_switch);
>      }
> @@ -497,6 +500,7 @@ SDState *sd_init(BlockBackend *blk, bool is_spi)
>          /* FIXME ignoring blk_attach_dev() failure is dangerously brittle */
>          blk_attach_dev(sd->blk, sd);
>          blk_set_dev_ops(sd->blk, &sd_block_ops, sd);
> +        sd->medium_inserted = blk_is_inserted(sd->blk);
>      }
>      vmstate_register(NULL, -1, &sd_vmstate, sd);
>      return sd;
> @@ -507,7 +511,7 @@ void sd_set_cb(SDState *sd, qemu_irq readonly, qemu_irq insert)
>      sd->readonly_cb = readonly;
>      sd->inserted_cb = insert;
>      qemu_set_irq(readonly, sd->blk ? blk_is_read_only(sd->blk) : 0);
> -    qemu_set_irq(insert, sd->blk ? blk_is_inserted(sd->blk) : 0);
> +    qemu_set_irq(insert, sd->blk ? sd->medium_inserted : 0);
>  }
>  
>  static void sd_erase(SDState *sd)
> @@ -1349,7 +1353,7 @@ int sd_do_command(SDState *sd, SDRequest *req,
>      sd_rsp_type_t rtype;
>      int rsplen;
>  
> -    if (!sd->blk || !blk_is_inserted(sd->blk) || !sd->enable) {
> +    if (!sd->blk || !sd->medium_inserted || !sd->enable) {
>          return 0;
>      }
>  
> @@ -1517,7 +1521,7 @@ void sd_write_data(SDState *sd, uint8_t value)
>  {
>      int i;
>  
> -    if (!sd->blk || !blk_is_inserted(sd->blk) || !sd->enable)
> +    if (!sd->blk || !sd->medium_inserted || !sd->enable)
>          return;
>  
>      if (sd->state != sd_receivingdata_state) {
> @@ -1643,7 +1647,7 @@ uint8_t sd_read_data(SDState *sd)
>      uint8_t ret;
>      int io_len;
>  
> -    if (!sd->blk || !blk_is_inserted(sd->blk) || !sd->enable)
> +    if (!sd->blk || !sd->medium_inserted || !sd->enable)
>          return 0x00;
>  
>      if (sd->state != sd_sendingdata_state) {
> 

Same thing you did to the floppy. Looks OK.

Reviewed-by: John Snow <jsnow@redhat.com>

Re: [Qemu-devel] [PATCH 1/2] hw/sd: Add medium insertion status

[Peter Maydell]

On 7 January 2016 at 21:03, Max Reitz <mreitz@redhat.com> wrote:
> Right now, the change_media_cb (sd_cardchange()) completely ignores its
> @load parameter. This means that issuing a blockdev-open-tray command
> will actually not have any effect.
>
> Fix this by keeping track of the medium insertion status in the SDState
> and updating it in sd_init() and sd_cardchange().
>
> Cc: qemu-stable <qemu-stable@nongnu.org>
> Signed-off-by: Max Reitz <mreitz@redhat.com>
> ---
>  hw/sd/sd.c | 16 ++++++++++------
>  1 file changed, 10 insertions(+), 6 deletions(-)
>
> diff --git a/hw/sd/sd.c b/hw/sd/sd.c
> index 1a9935c..0751ba2 100644
> --- a/hw/sd/sd.c
> +++ b/hw/sd/sd.c
> @@ -114,6 +114,7 @@ struct SDState {
>      uint8_t *buf;
>
>      bool enable;
> +    bool medium_inserted;

When would this be different from blk_is_inserted(sd->blk) ?

I don't see why we need a separate flag here rather
than querying the block layer.

If we do need a flag, why doesn't it need to be migrated?

thanks
-- PMM

Re: [Qemu-devel] [PATCH 1/2] hw/sd: Add medium insertion status

[Max Reitz]

[-- Attachment #1: Type: text/plain, Size: 1770 bytes --]

On 07.01.2016 22:45, Peter Maydell wrote:
> On 7 January 2016 at 21:03, Max Reitz <mreitz@redhat.com> wrote:
>> Right now, the change_media_cb (sd_cardchange()) completely ignores its
>> @load parameter. This means that issuing a blockdev-open-tray command
>> will actually not have any effect.
>>
>> Fix this by keeping track of the medium insertion status in the SDState
>> and updating it in sd_init() and sd_cardchange().
>>
>> Cc: qemu-stable <qemu-stable@nongnu.org>
>> Signed-off-by: Max Reitz <mreitz@redhat.com>
>> ---
>>  hw/sd/sd.c | 16 ++++++++++------
>>  1 file changed, 10 insertions(+), 6 deletions(-)
>>
>> diff --git a/hw/sd/sd.c b/hw/sd/sd.c
>> index 1a9935c..0751ba2 100644
>> --- a/hw/sd/sd.c
>> +++ b/hw/sd/sd.c
>> @@ -114,6 +114,7 @@ struct SDState {
>>      uint8_t *buf;
>>
>>      bool enable;
>> +    bool medium_inserted;
> 
> When would this be different from blk_is_inserted(sd->blk) ?

First, whenever you use a host CD-ROM drive for an SD card. But I highly
doubt this worked before, because the device model is actually never
notified if the medium in the host drive is exchanged.

For any other medium, blk_is_inserted() should generally be always true.

Second, whenever the change_media_cb is called with @load set to false.
This means that the medium is to be unloaded, which was completely
ignored by sd.c so far (see commit message). It just interpreted a call
to sd_cardchange() as a notification that the medium might have changed,
but this is wrong.

> I don't see why we need a separate flag here rather
> than querying the block layer.

Because of the second case.

> If we do need a flag, why doesn't it need to be migrated?

Oh, yes, you're right, it probably should.

Max


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]