Re: [Qemu-devel] [PATCH v2 5/6] atapi: GESN: implement 'media' subcommand

[Kevin Wolf]

Am 12.04.2011 11:27, schrieb Amit Shah:
> Implement the 'media' sub-command of the GET_EVENT_STATUS_NOTIFICATION
> command.  This helps us report tray open, tray closed, no media, media
> present states to the guest.
> 
> Newer Linux kernels (2.6.38+) rely on this command to revalidate discs
> after media change.
> 
> This patch also sends out tray open/closed status to the guest driver
> when requested e.g. via the CDROM_DRIVE_STATUS ioctl (thanks Markus).
> Without such notification, the guest and qemu's tray open/close status
> was frequently out of sync, causing installers like Anaconda detecting
> no disc instead of tray open, confusing them terribly.
> 
> Signed-off-by: Amit Shah <address@hidden>
> ---
>  hw/ide/core.c     |   92 +++++++++++++++++++++++++++++++++++++++++++++++++++-
>  hw/ide/internal.h |    6 +++
>  2 files changed, 96 insertions(+), 2 deletions(-)
> 
> diff --git a/hw/ide/core.c b/hw/ide/core.c
> index fe50d8a..2683070 100644
> --- a/hw/ide/core.c
> +++ b/hw/ide/core.c
> @@ -1084,6 +1084,47 @@ static int ide_dvd_read_structure(IDEState *s, int 
> format,
>      }
>  }
>  
> +static unsigned int event_status_media(IDEState *s,
> +                                       uint8_t *buf,
> +                                       unsigned int max_len,
> +                                       unsigned int event_class,
> +                                       unsigned int supported_events)
> +{
> +    enum media_event_code {
> +        no_change = 0,       /* Status unchanged */
> +        eject_requested,     /* received a request from user to eject */
> +        new_media,           /* new media inserted and ready for access */
> +        media_removal,       /* only for media changers */
> +        media_changed,       /* only for media changers */
> +        bg_format_completed, /* MRW or DVD+RW b/g format completed */
> +        bg_format_restarted, /* MRW or DVD+RW b/g format restarted */
> +    };
> +    enum media_status {
> +        tray_open = 1,
> +        media_present = 2,
> +    };
> +    uint8_t event_code, media_status;
> +
> +    media_status = 0;
> +    if (s->bs->tray_open) {
> +        media_status = tray_open;
> +    } else if (bdrv_is_inserted(s->bs)) {
> +        media_status = media_present;
> +    }
> +
> +    /* Event notification descriptor */
> +    event_code = no_change;
> +    if (media_status != tray_open && s->events.new_media) {
> +        event_code = new_media;
> +        s->events.new_media = false;
> +    }
> +
> +    buf[4] = event_code;
> +    buf[5] = media_status;
> +
> +    return 6; /* We wrote to just 2 extra bytes from the header */

I must admit that I don't understand your answer you gave on v1 here.
Let me quote:

>> > After media_state, there are two more fields for start/end slot (even
>> > though they are reserved because we don't have a multiple slot device)
> Yes, they're reserved, so we shouldn't change them.  Any change might
> trigger bad response from guests.

I'm not sure what we would be _changing_. We're building a response
structure here, not modifying existing data.

So what your code does is to receive a short response that leaves out
the reserved fields. I don't think this is how it's supposed to work. We
should include the reserved fields in the descriptor length and zero
them (MMC-5, section 3.6.8):

"“Reserved” is a keyword referring to bits, bytes, words, fields and
code values that are set aside for future standardization. A reserved
bit, byte, word or field shall be set to zero, or in accordance with a
future extension to this standard."

> +}
> +
>  static unsigned int event_status_nea(uint8_t *buf, unsigned int max_len)
>  {
>      cpu_to_ube16(buf, 0x00); /* No event descriptor returned */
> @@ -1107,7 +1148,28 @@ static void 
> handle_get_event_status_notification(IDEState *s,
>          uint8_t len_lsb;
>          uint8_t control;
>      } __attribute__((packed)) *gesn_cdb;
> +    enum notification_class_request_type {
> +        reserved1 = 1 << 0,
> +        operational_change = 1 << 1,
> +        power_management = 1 << 2,
> +        external_request = 1 << 3,
> +        media = 1 << 4,
> +        multi_host = 1 << 5,
> +        device_busy = 1 << 6,
> +        reserved2 = 1 << 7,
> +    };
> +    enum event_notification_class_field {
> +        enc_no_events = 0,
> +        enc_operational_change,
> +        enc_power_management,
> +        enc_external_request,
> +        enc_media,
> +        enc_multiple_hosts,
> +        enc_device_busy,
> +        enc_reserved,
> +    };
>      unsigned int max_len, used_len;
> +    unsigned int supported_events;
>  
>      gesn_cdb = (void *)packet;
>      max_len = ube16_to_cpu(&gesn_cdb->len_msb);
> @@ -1122,8 +1184,33 @@ static void 
> handle_get_event_status_notification(IDEState *s,
>  
>      /* polling mode operation */
>  
> -    /* We don't support any event class (yet). */
> -    used_len = event_status_nea(buf, max_len);
> +    /*
> +     * These are the supported events.
> +     *
> +     * We currently only support requests of the 'media' type.
> +     */
> +    supported_events = media;
> +
> +    /*
> +     * Event notification header; will be overwritten by the
> +     * NO_EVENT_AVAILABLE code if we don't have events: according to
> +     * MMC-5 6.7.2.2, if nea = 1, event class field should be 0.
> +     */
> +    cpu_to_ube16(buf, max_len);

That doesn't look right. I think it needs to be used_len - sizeof(header).

> +    buf[2] = event_class;
> +    buf[3] = supported_events;

The spec doesn't talk about "event class" for buf[2], but about
"notification class" (same applies for the comment above).  We should
stick to that as buf[3] is called "supported event class", so they might
be confused.

> +
> +    /*
> +     * Responses to requests are to be based on request priority.  The
> +     * notification_class_request_type enum above specifies the
> +     * priority: upper elements are higher prio than lower ones.
> +     */
> +    if (gesn_cdb->request & media) {
> +        used_len = event_status_media(s, buf, max_len, enc_media,
> +                                      supported_events);
> +    } else {
> +        used_len = event_status_nea(buf, max_len);

This expands to:

    cpu_to_ube16(buf, 0x00); /* No event descriptor returned */
    buf[2] = 0x80;           /* No Event Available (NEA) */
    buf[3] = 0x00;           /* Empty supported event classes */

The first one is covered if you fix what I mentioned above (used_len -
sizeof(header)). The third one is wrong, we still support the media
class. So what is left is one line:

    buf[2] = 0x80;           /* No Event Available (NEA) */

I'm still not convinced that having event_status_nea() as a separate
function makes a lot of sense.

> +    }
>      ide_atapi_cmd_reply(s, used_len, max_len);
>  }
>  
> @@ -1650,6 +1737,7 @@ static void cdrom_change_cb(void *opaque, int reason)
>      s->sense_key = SENSE_UNIT_ATTENTION;
>      s->asc = ASC_MEDIUM_MAY_HAVE_CHANGED;
>      s->cdrom_changed = 1;
> +    s->events.new_media = true;
>      ide_set_irq(s->bus);
>  }
>  
> diff --git a/hw/ide/internal.h b/hw/ide/internal.h
> index d533fb6..ba7e9a8 100644
> --- a/hw/ide/internal.h
> +++ b/hw/ide/internal.h
> @@ -373,6 +373,11 @@ typedef int DMAFunc(IDEDMA *);
>  typedef int DMAIntFunc(IDEDMA *, int);
>  typedef void DMARestartFunc(void *, int, int);
>  
> +struct unreported_events {
> +    bool eject_request;
> +    bool new_media;
> +};
> +
>  /* NOTE: IDEState represents in fact one drive */
>  struct IDEState {
>      IDEBus *bus;
> @@ -408,6 +413,7 @@ struct IDEState {
>      BlockDriverState *bs;
>      char version[9];
>      /* ATAPI specific */
> +    struct unreported_events events;
>      uint8_t sense_key;
>      uint8_t asc;
>      uint8_t cdrom_changed;

Kevin