drm/dp: Use large transactions for I2C over AUX

Commit Message

Simon Farnsworth Feb. 10, 2015, 6:38 p.m. UTC

Older DisplayPort to DVI-D Dual Link adapters designed by Bizlink have bugs
in their I2C over AUX implementation (fixed in newer revisions). They work
fine with Windows, but fail with Linux.

It turns out that they cannot keep an I2C transaction open unless the
previous read was 16 bytes; shorter reads can only be followed by a zero
byte transfer ending the I2C transaction.

Copy Windows's behaviour, and read 16 bytes at a time. If we get a short
reply, assume that there's a hardware bottleneck, and shrink our read size
to match. For this purpose, use the algorithm in the DisplayPort 1.2 spec,
in the hopes that it'll be closest to what Windows does.

Also provide an unsafe module parameter for testing smaller transfer sizes,
in case there are sinks out there that cannot work with Windows.

Note also that despite the previous comment in drm_dp_i2c_xfer, this speeds
up native DP EDID reads; Ville Syrjälä <[e-mail]> found
the following changes in his testing:

Device under test:     old  -> with this patch
DP->DVI (OUI 001cf8):  40ms -> 35ms
DP->VGA (OUI 0022b9):  45ms -> 38ms
Zotac DP->2xHDMI:      25ms ->  4ms
Asus PB278 monitor:    22ms ->  3ms

A back of the envelope calculation shows that peak theoretical transfer rate
for 1 byte reads is around 60 kbit/s; with 16 byte reads, this increases to
around 500 kbit/s, which explains the increase in speed.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=55228
Tested-by: Aidan Marks <[e-mail]>
Signed-off-by: Simon Farnsworth <[e-mail]>
---

v4 changes:

 * Change short reply algorithm after suggestions from Ville.

 * Expanded commit message.

 * Mark the module parameter unsafe.

 * Use clamp() to bring the module parameter into range when used.

v3 changes, after feedback from Ville and more testing of Windows:

 * Change the short reply algorithm to match Ville's description of the
   DisplayPort 1.2 spec wording.

 * Add a module parameter to set the default transfer size for
   experiments. Requested over IRC by Ville.

No-one's been able to find a device that does short replies, but experiments
show that bigger reads are faster on most devices. Ville got:

 DP->DVI (OUI 001cf8):  40ms -> 35ms
 DP->VGA (OUI 0022b9):  45ms -> 38ms
 Zotac DP->2xHDMI:      25ms ->  4ms

v2 changes, after feedback from Thierry and Ville:

 * Handle short replies. I've decided (arbitrarily) that a short reply
   results in us dropping back to the newly chosen size for the rest of this
   I2C transaction. Thus, given an attempt to read the first 16 bytes of
   EDID, and a sink that only does 4 bytes of buffering, we will see the
   following AUX transfers for the EDID read (after address is set):

   <set address, block etc>
   Read 16 bytes from I2C over AUX.
   Reply with 4 bytes
   Read 4 bytes
   Reply with 4 bytes
   Read 4 bytes
   Reply with 4 bytes
   Read 4 bytes
   Reply with 4 bytes
   <end I2C transaction>

Note that I've not looked at MST support - I have neither the DP 1.2 spec
nor any MST branch devices, so I can't test anything I write or check it
against a spec. It looks from the code, however, as if MST has the branch
device do the split from a big request into small transactions.

 drivers/gpu/drm/drm_dp_helper.c | 76 +++++++++++++++++++++++++++++++----------
 include/drm/drm_dp_helper.h     |  5 +++
 2 files changed, 63 insertions(+), 18 deletions(-)

Comments

Simon Farnsworth Feb. 10, 2015, 6:42 p.m. UTC | #1

A note:

This is *not* enough to bring us to parity with Windows with these
adapters. There's also something wrong with our HPD handling that trips them
up - I suspect, based on AUX traces from one that's happy, that it's our
handling of short HPDs that's imperfect, as the device does a short HPD when
it sees the DVI-D HPD sense change.

Simon

On Tuesday 10 February 2015 18:38:08 Simon Farnsworth wrote:
> Older DisplayPort to DVI-D Dual Link adapters designed by Bizlink have bugs
> in their I2C over AUX implementation (fixed in newer revisions). They work
> fine with Windows, but fail with Linux.
> 
> It turns out that they cannot keep an I2C transaction open unless the
> previous read was 16 bytes; shorter reads can only be followed by a zero
> byte transfer ending the I2C transaction.
> 
> Copy Windows's behaviour, and read 16 bytes at a time. If we get a short
> reply, assume that there's a hardware bottleneck, and shrink our read size
> to match. For this purpose, use the algorithm in the DisplayPort 1.2 spec,
> in the hopes that it'll be closest to what Windows does.
> 
> Also provide an unsafe module parameter for testing smaller transfer sizes,
> in case there are sinks out there that cannot work with Windows.
> 
> Note also that despite the previous comment in drm_dp_i2c_xfer, this speeds
> up native DP EDID reads; Ville Syrjälä <[e-mail]> found
> the following changes in his testing:
> 
> Device under test:     old  -> with this patch
> DP->DVI (OUI 001cf8):  40ms -> 35ms
> DP->VGA (OUI 0022b9):  45ms -> 38ms
> Zotac DP->2xHDMI:      25ms ->  4ms
> Asus PB278 monitor:    22ms ->  3ms
> 
> A back of the envelope calculation shows that peak theoretical transfer rate
> for 1 byte reads is around 60 kbit/s; with 16 byte reads, this increases to
> around 500 kbit/s, which explains the increase in speed.
> 
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=55228
> Tested-by: Aidan Marks <[e-mail]>
> Signed-off-by: Simon Farnsworth <[e-mail]>
> ---
> 
> v4 changes:
> 
>  * Change short reply algorithm after suggestions from Ville.
> 
>  * Expanded commit message.
> 
>  * Mark the module parameter unsafe.
> 
>  * Use clamp() to bring the module parameter into range when used.
> 
> v3 changes, after feedback from Ville and more testing of Windows:
> 
>  * Change the short reply algorithm to match Ville's description of the
>    DisplayPort 1.2 spec wording.
> 
>  * Add a module parameter to set the default transfer size for
>    experiments. Requested over IRC by Ville.
> 
> No-one's been able to find a device that does short replies, but experiments
> show that bigger reads are faster on most devices. Ville got:
> 
>  DP->DVI (OUI 001cf8):  40ms -> 35ms
>  DP->VGA (OUI 0022b9):  45ms -> 38ms
>  Zotac DP->2xHDMI:      25ms ->  4ms
> 
> v2 changes, after feedback from Thierry and Ville:
> 
>  * Handle short replies. I've decided (arbitrarily) that a short reply
>    results in us dropping back to the newly chosen size for the rest of this
>    I2C transaction. Thus, given an attempt to read the first 16 bytes of
>    EDID, and a sink that only does 4 bytes of buffering, we will see the
>    following AUX transfers for the EDID read (after address is set):
> 
>    <set address, block etc>
>    Read 16 bytes from I2C over AUX.
>    Reply with 4 bytes
>    Read 4 bytes
>    Reply with 4 bytes
>    Read 4 bytes
>    Reply with 4 bytes
>    Read 4 bytes
>    Reply with 4 bytes
>    <end I2C transaction>
> 
> Note that I've not looked at MST support - I have neither the DP 1.2 spec
> nor any MST branch devices, so I can't test anything I write or check it
> against a spec. It looks from the code, however, as if MST has the branch
> device do the split from a big request into small transactions.
> 
>  drivers/gpu/drm/drm_dp_helper.c | 76 +++++++++++++++++++++++++++++++----------
>  include/drm/drm_dp_helper.h     |  5 +++
>  2 files changed, 63 insertions(+), 18 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_dp_helper.c b/drivers/gpu/drm/drm_dp_helper.c
> index 79968e3..105fd66 100644
> --- a/drivers/gpu/drm/drm_dp_helper.c
> +++ b/drivers/gpu/drm/drm_dp_helper.c
> @@ -396,11 +396,13 @@ static u32 drm_dp_i2c_functionality(struct i2c_adapter *adapter)
>   * retrying the transaction as appropriate.  It is assumed that the
>   * aux->transfer function does not modify anything in the msg other than the
>   * reply field.
> + *
> + * Returns bytes transferred on success, or a negative error code on failure.
>   */
>  static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
>  {
>  	unsigned int retry;
> -	int err;
> +	int ret;
>  
>  	/*
>  	 * DP1.2 sections 2.7.7.1.5.6.1 and 2.7.7.1.6.6.1: A DP Source device
> @@ -409,14 +411,14 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
>  	 */
>  	for (retry = 0; retry < 7; retry++) {
>  		mutex_lock(&aux->hw_mutex);
> -		err = aux->transfer(aux, msg);
> +		ret = aux->transfer(aux, msg);
>  		mutex_unlock(&aux->hw_mutex);
> -		if (err < 0) {
> -			if (err == -EBUSY)
> +		if (ret < 0) {
> +			if (ret == -EBUSY)
>  				continue;
>  
> -			DRM_DEBUG_KMS("transaction failed: %d\n", err);
> -			return err;
> +			DRM_DEBUG_KMS("transaction failed: %d\n", ret);
> +			return ret;
>  		}
>  
>  
> @@ -457,9 +459,7 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
>  			 * Both native ACK and I2C ACK replies received. We
>  			 * can assume the transfer was successful.
>  			 */
> -			if (err < msg->size)
> -				return -EPROTO;
> -			return 0;
> +			return ret;
>  
>  		case DP_AUX_I2C_REPLY_NACK:
>  			DRM_DEBUG_KMS("I2C nack\n");
> @@ -482,14 +482,55 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
>  	return -EREMOTEIO;
>  }
>  
> +/*
> + * Keep retrying drm_dp_i2c_do_msg until all data has been transferred.
> + *
> + * Returns an error code on failure, or a recommended transfer size on success.
> + */
> +static int drm_dp_i2c_drain_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *orig_msg)
> +{
> +	int err, ret = orig_msg->size;
> +	struct drm_dp_aux_msg msg = *orig_msg;
> +
> +	while (msg.size > 0) {
> +		err = drm_dp_i2c_do_msg(aux, &msg);
> +		if (err <= 0)
> +			return err == 0 ? -EPROTO : err;
> +
> +		if (err < msg.size && err < ret) {
> +			DRM_DEBUG_KMS("Partial I2C reply: requested %zu bytes got %d bytes\n",
> +				      msg.size, err);
> +			ret = err;
> +		}
> +
> +		msg.size -= err;
> +		msg.buffer += err;
> +	}
> +
> +	return ret;
> +}
> +
> +/*
> + * Bizlink designed DP->DVI-D Dual Link adapters require the I2C over AUX
> + * packets to be as large as possible. If not, the I2C transactions never
> + * succeed. Hence the default is maximum.
> + */
> +static int dp_aux_i2c_transfer_size __read_mostly = DP_AUX_MAX_PAYLOAD_BYTES;
> +module_param_unsafe(dp_aux_i2c_transfer_size, int, 0644);
> +MODULE_PARM_DESC(dp_aux_i2c_transfer_size,
> +		 "Number of bytes to transfer in a single I2C over DP AUX CH message, (1-16, default 16)");
> +
>  static int drm_dp_i2c_xfer(struct i2c_adapter *adapter, struct i2c_msg *msgs,
>  			   int num)
>  {
>  	struct drm_dp_aux *aux = adapter->algo_data;
>  	unsigned int i, j;
> +	unsigned transfer_size;
>  	struct drm_dp_aux_msg msg;
>  	int err = 0;
>  
> +	dp_aux_i2c_transfer_size = clamp(dp_aux_i2c_transfer_size, 1, DP_AUX_MAX_PAYLOAD_BYTES);
> +
>  	memset(&msg, 0, sizeof(msg));
>  
>  	for (i = 0; i < num; i++) {
> @@ -507,20 +548,19 @@ static int drm_dp_i2c_xfer(struct i2c_adapter *adapter, struct i2c_msg *msgs,
>  		err = drm_dp_i2c_do_msg(aux, &msg);
>  		if (err < 0)
>  			break;
> -		/*
> -		 * Many hardware implementations support FIFOs larger than a
> -		 * single byte, but it has been empirically determined that
> -		 * transferring data in larger chunks can actually lead to
> -		 * decreased performance. Therefore each message is simply
> -		 * transferred byte-by-byte.
> +		/* We want each transaction to be as large as possible, but
> +		 * we'll go to smaller sizes if the hardware gives us a
> +		 * short reply.
>  		 */
> -		for (j = 0; j < msgs[i].len; j++) {
> +		transfer_size = dp_aux_i2c_transfer_size;
> +		for (j = 0; j < msgs[i].len; j += msg.size) {
>  			msg.buffer = msgs[i].buf + j;
> -			msg.size = 1;
> +			msg.size = min(transfer_size, msgs[i].len - j);
>  
> -			err = drm_dp_i2c_do_msg(aux, &msg);
> +			err = drm_dp_i2c_drain_msg(aux, &msg);
>  			if (err < 0)
>  				break;
> +			transfer_size = err;
>  		}
>  		if (err < 0)
>  			break;
> diff --git a/include/drm/drm_dp_helper.h b/include/drm/drm_dp_helper.h
> index 11f8c84..444d51b 100644
> --- a/include/drm/drm_dp_helper.h
> +++ b/include/drm/drm_dp_helper.h
> @@ -42,6 +42,8 @@
>   * 1.2 formally includes both eDP and DPI definitions.
>   */
>  
> +#define DP_AUX_MAX_PAYLOAD_BYTES	16
> +
>  #define DP_AUX_I2C_WRITE		0x0
>  #define DP_AUX_I2C_READ			0x1
>  #define DP_AUX_I2C_STATUS		0x2
> @@ -519,6 +521,9 @@ struct drm_dp_aux_msg {
>   * transactions. The drm_dp_aux_register_i2c_bus() function registers an
>   * I2C adapter that can be passed to drm_probe_ddc(). Upon removal, drivers
>   * should call drm_dp_aux_unregister_i2c_bus() to remove the I2C adapter.
> + * The I2C adapter uses long transfers by default; if a partial response is
> + * received, the adapter will drop down to the size given by the partial
> + * response for this transaction only.
>   *
>   * Note that the aux helper code assumes that the .transfer() function
>   * only modifies the reply field of the drm_dp_aux_msg structure.  The
>

Dave Airlie Feb. 11, 2015, 5:36 a.m. UTC | #2

On 11 February 2015 at 04:42, Simon Farnsworth
<[e-mail]> wrote:
> A note:
>
> This is *not* enough to bring us to parity with Windows with these
> adapters. There's also something wrong with our HPD handling that trips them
> up - I suspect, based on AUX traces from one that's happy, that it's our
> handling of short HPDs that's imperfect, as the device does a short HPD when
> it sees the DVI-D HPD sense change.
>

We don't handle soft irqs well outside of MST, I think the Apple VGA dongle
also causes a soft irq when you plug a vga monitor into the dongle, when
 the dongle is already plugged in.

Dave.

Daniel Vetter Feb. 11, 2015, 7:25 a.m. UTC | #3

On Wed, Feb 11, 2015 at 03:36:29PM +1000, Dave Airlie wrote:
> On 11 February 2015 at 04:42, Simon Farnsworth
> <[e-mail]> wrote:
> > A note:
> >
> > This is *not* enough to bring us to parity with Windows with these
> > adapters. There's also something wrong with our HPD handling that trips them
> > up - I suspect, based on AUX traces from one that's happy, that it's our
> > handling of short HPDs that's imperfect, as the device does a short HPD when
> > it sees the DVI-D HPD sense change.
> >
> 
> We don't handle soft irqs well outside of MST, I think the Apple VGA dongle
> also causes a soft irq when you plug a vga monitor into the dongle, when
>  the dongle is already plugged in.

Yeah we probably need some helper function to handle sst short pulse. The
problem there is that with sst the driver owns the drm connector and not
the helper like with mst, so I'm not sure how we should communicate back
to the driver what has been detected exactly. Maybe just a pile of out
pointers for edid and stuff, or we create an sst_connector_state struct or
something. Of course mst should then use the same for handling sst leafs
so that we can share all the special magic for e.g. the apple vga dongle.
-Daniel

Jani Nikula Feb. 11, 2015, 8:13 a.m. UTC | #4

On Tue, 10 Feb 2015, Simon Farnsworth <[e-mail]> wrote:
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=55228
> Tested-by: Aidan Marks <[e-mail]>

Aidan writes in the bug, "Hi Simon, I'm sorry to report that the v4
patch doesn't work for me at all, it reverts to 1024x768.  v3 was great
though."

BR,
Jani.

Ville Syrjälä Feb. 11, 2015, 12:05 p.m. UTC | #5

On Tue, Feb 10, 2015 at 06:38:08PM +0000, Simon Farnsworth wrote:
> Older DisplayPort to DVI-D Dual Link adapters designed by Bizlink have bugs
> in their I2C over AUX implementation (fixed in newer revisions). They work
> fine with Windows, but fail with Linux.
> 
> It turns out that they cannot keep an I2C transaction open unless the
> previous read was 16 bytes; shorter reads can only be followed by a zero
> byte transfer ending the I2C transaction.
> 
> Copy Windows's behaviour, and read 16 bytes at a time. If we get a short
> reply, assume that there's a hardware bottleneck, and shrink our read size
> to match. For this purpose, use the algorithm in the DisplayPort 1.2 spec,
> in the hopes that it'll be closest to what Windows does.
> 
> Also provide an unsafe module parameter for testing smaller transfer sizes,
> in case there are sinks out there that cannot work with Windows.
> 
> Note also that despite the previous comment in drm_dp_i2c_xfer, this speeds
> up native DP EDID reads; Ville Syrjälä <[e-mail]> found
> the following changes in his testing:
> 
> Device under test:     old  -> with this patch
> DP->DVI (OUI 001cf8):  40ms -> 35ms
> DP->VGA (OUI 0022b9):  45ms -> 38ms
> Zotac DP->2xHDMI:      25ms ->  4ms
> Asus PB278 monitor:    22ms ->  3ms
> 
> A back of the envelope calculation shows that peak theoretical transfer rate
> for 1 byte reads is around 60 kbit/s; with 16 byte reads, this increases to
> around 500 kbit/s, which explains the increase in speed.
> 
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=55228
> Tested-by: Aidan Marks <[e-mail]>
> Signed-off-by: Simon Farnsworth <[e-mail]>
> ---
> 
> v4 changes:
> 
>  * Change short reply algorithm after suggestions from Ville.
> 
>  * Expanded commit message.
> 
>  * Mark the module parameter unsafe.
> 
>  * Use clamp() to bring the module parameter into range when used.
> 
> v3 changes, after feedback from Ville and more testing of Windows:
> 
>  * Change the short reply algorithm to match Ville's description of the
>    DisplayPort 1.2 spec wording.
> 
>  * Add a module parameter to set the default transfer size for
>    experiments. Requested over IRC by Ville.
> 
> No-one's been able to find a device that does short replies, but experiments
> show that bigger reads are faster on most devices. Ville got:
> 
>  DP->DVI (OUI 001cf8):  40ms -> 35ms
>  DP->VGA (OUI 0022b9):  45ms -> 38ms
>  Zotac DP->2xHDMI:      25ms ->  4ms
> 
> v2 changes, after feedback from Thierry and Ville:
> 
>  * Handle short replies. I've decided (arbitrarily) that a short reply
>    results in us dropping back to the newly chosen size for the rest of this
>    I2C transaction. Thus, given an attempt to read the first 16 bytes of
>    EDID, and a sink that only does 4 bytes of buffering, we will see the
>    following AUX transfers for the EDID read (after address is set):
> 
>    <set address, block etc>
>    Read 16 bytes from I2C over AUX.
>    Reply with 4 bytes
>    Read 4 bytes
>    Reply with 4 bytes
>    Read 4 bytes
>    Reply with 4 bytes
>    Read 4 bytes
>    Reply with 4 bytes
>    <end I2C transaction>
> 
> Note that I've not looked at MST support - I have neither the DP 1.2 spec
> nor any MST branch devices, so I can't test anything I write or check it
> against a spec. It looks from the code, however, as if MST has the branch
> device do the split from a big request into small transactions.
> 
>  drivers/gpu/drm/drm_dp_helper.c | 76 +++++++++++++++++++++++++++++++----------
>  include/drm/drm_dp_helper.h     |  5 +++
>  2 files changed, 63 insertions(+), 18 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_dp_helper.c b/drivers/gpu/drm/drm_dp_helper.c
> index 79968e3..105fd66 100644
> --- a/drivers/gpu/drm/drm_dp_helper.c
> +++ b/drivers/gpu/drm/drm_dp_helper.c
> @@ -396,11 +396,13 @@ static u32 drm_dp_i2c_functionality(struct i2c_adapter *adapter)
>   * retrying the transaction as appropriate.  It is assumed that the
>   * aux->transfer function does not modify anything in the msg other than the
>   * reply field.
> + *
> + * Returns bytes transferred on success, or a negative error code on failure.
>   */
>  static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
>  {
>  	unsigned int retry;
> -	int err;
> +	int ret;
>  
>  	/*
>  	 * DP1.2 sections 2.7.7.1.5.6.1 and 2.7.7.1.6.6.1: A DP Source device
> @@ -409,14 +411,14 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
>  	 */
>  	for (retry = 0; retry < 7; retry++) {
>  		mutex_lock(&aux->hw_mutex);
> -		err = aux->transfer(aux, msg);
> +		ret = aux->transfer(aux, msg);
>  		mutex_unlock(&aux->hw_mutex);
> -		if (err < 0) {
> -			if (err == -EBUSY)
> +		if (ret < 0) {
> +			if (ret == -EBUSY)
>  				continue;
>  
> -			DRM_DEBUG_KMS("transaction failed: %d\n", err);
> -			return err;
> +			DRM_DEBUG_KMS("transaction failed: %d\n", ret);
> +			return ret;
>  		}
>  
>  
> @@ -457,9 +459,7 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
>  			 * Both native ACK and I2C ACK replies received. We
>  			 * can assume the transfer was successful.
>  			 */
> -			if (err < msg->size)
> -				return -EPROTO;
> -			return 0;
> +			return ret;
>  
>  		case DP_AUX_I2C_REPLY_NACK:
>  			DRM_DEBUG_KMS("I2C nack\n");
> @@ -482,14 +482,55 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
>  	return -EREMOTEIO;
>  }
>  
> +/*
> + * Keep retrying drm_dp_i2c_do_msg until all data has been transferred.
> + *
> + * Returns an error code on failure, or a recommended transfer size on success.
> + */
> +static int drm_dp_i2c_drain_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *orig_msg)

orig_msg should be const to make sure someone doesn't change it by
accident since that would break drm_dp_i2c_xfer().

Otherwise looks good to me, so
Reviewed-by: Ville Syrjälä <[e-mail]>

I also tested this version with a few dongles and everything seems to
work as intended.

> +{
> +	int err, ret = orig_msg->size;
> +	struct drm_dp_aux_msg msg = *orig_msg;
> +
> +	while (msg.size > 0) {
> +		err = drm_dp_i2c_do_msg(aux, &msg);
> +		if (err <= 0)
> +			return err == 0 ? -EPROTO : err;
> +
> +		if (err < msg.size && err < ret) {
> +			DRM_DEBUG_KMS("Partial I2C reply: requested %zu bytes got %d bytes\n",
> +				      msg.size, err);
> +			ret = err;
> +		}
> +
> +		msg.size -= err;
> +		msg.buffer += err;
> +	}
> +
> +	return ret;
> +}
> +
> +/*
> + * Bizlink designed DP->DVI-D Dual Link adapters require the I2C over AUX
> + * packets to be as large as possible. If not, the I2C transactions never
> + * succeed. Hence the default is maximum.
> + */
> +static int dp_aux_i2c_transfer_size __read_mostly = DP_AUX_MAX_PAYLOAD_BYTES;
> +module_param_unsafe(dp_aux_i2c_transfer_size, int, 0644);
> +MODULE_PARM_DESC(dp_aux_i2c_transfer_size,
> +		 "Number of bytes to transfer in a single I2C over DP AUX CH message, (1-16, default 16)");
> +
>  static int drm_dp_i2c_xfer(struct i2c_adapter *adapter, struct i2c_msg *msgs,
>  			   int num)
>  {
>  	struct drm_dp_aux *aux = adapter->algo_data;
>  	unsigned int i, j;
> +	unsigned transfer_size;
>  	struct drm_dp_aux_msg msg;
>  	int err = 0;
>  
> +	dp_aux_i2c_transfer_size = clamp(dp_aux_i2c_transfer_size, 1, DP_AUX_MAX_PAYLOAD_BYTES);
> +
>  	memset(&msg, 0, sizeof(msg));
>  
>  	for (i = 0; i < num; i++) {
> @@ -507,20 +548,19 @@ static int drm_dp_i2c_xfer(struct i2c_adapter *adapter, struct i2c_msg *msgs,
>  		err = drm_dp_i2c_do_msg(aux, &msg);
>  		if (err < 0)
>  			break;
> -		/*
> -		 * Many hardware implementations support FIFOs larger than a
> -		 * single byte, but it has been empirically determined that
> -		 * transferring data in larger chunks can actually lead to
> -		 * decreased performance. Therefore each message is simply
> -		 * transferred byte-by-byte.
> +		/* We want each transaction to be as large as possible, but
> +		 * we'll go to smaller sizes if the hardware gives us a
> +		 * short reply.
>  		 */
> -		for (j = 0; j < msgs[i].len; j++) {
> +		transfer_size = dp_aux_i2c_transfer_size;
> +		for (j = 0; j < msgs[i].len; j += msg.size) {
>  			msg.buffer = msgs[i].buf + j;
> -			msg.size = 1;
> +			msg.size = min(transfer_size, msgs[i].len - j);
>  
> -			err = drm_dp_i2c_do_msg(aux, &msg);
> +			err = drm_dp_i2c_drain_msg(aux, &msg);
>  			if (err < 0)
>  				break;
> +			transfer_size = err;
>  		}
>  		if (err < 0)
>  			break;
> diff --git a/include/drm/drm_dp_helper.h b/include/drm/drm_dp_helper.h
> index 11f8c84..444d51b 100644
> --- a/include/drm/drm_dp_helper.h
> +++ b/include/drm/drm_dp_helper.h
> @@ -42,6 +42,8 @@
>   * 1.2 formally includes both eDP and DPI definitions.
>   */
>  
> +#define DP_AUX_MAX_PAYLOAD_BYTES	16
> +
>  #define DP_AUX_I2C_WRITE		0x0
>  #define DP_AUX_I2C_READ			0x1
>  #define DP_AUX_I2C_STATUS		0x2
> @@ -519,6 +521,9 @@ struct drm_dp_aux_msg {
>   * transactions. The drm_dp_aux_register_i2c_bus() function registers an
>   * I2C adapter that can be passed to drm_probe_ddc(). Upon removal, drivers
>   * should call drm_dp_aux_unregister_i2c_bus() to remove the I2C adapter.
> + * The I2C adapter uses long transfers by default; if a partial response is
> + * received, the adapter will drop down to the size given by the partial
> + * response for this transaction only.
>   *
>   * Note that the aux helper code assumes that the .transfer() function
>   * only modifies the reply field of the drm_dp_aux_msg structure.  The
> -- 
> 2.1.0

Ville Syrjälä March 11, 2015, 7:28 p.m. UTC | #6

On Wed, Feb 11, 2015 at 02:05:21PM +0200, Ville Syrjälä wrote:
> On Tue, Feb 10, 2015 at 06:38:08PM +0000, Simon Farnsworth wrote:
> > Older DisplayPort to DVI-D Dual Link adapters designed by Bizlink have bugs
> > in their I2C over AUX implementation (fixed in newer revisions). They work
> > fine with Windows, but fail with Linux.
> > 
> > It turns out that they cannot keep an I2C transaction open unless the
> > previous read was 16 bytes; shorter reads can only be followed by a zero
> > byte transfer ending the I2C transaction.
> > 
> > Copy Windows's behaviour, and read 16 bytes at a time. If we get a short
> > reply, assume that there's a hardware bottleneck, and shrink our read size
> > to match. For this purpose, use the algorithm in the DisplayPort 1.2 spec,
> > in the hopes that it'll be closest to what Windows does.
> > 
> > Also provide an unsafe module parameter for testing smaller transfer sizes,
> > in case there are sinks out there that cannot work with Windows.
> > 
> > Note also that despite the previous comment in drm_dp_i2c_xfer, this speeds
> > up native DP EDID reads; Ville Syrjälä <[e-mail]> found
> > the following changes in his testing:
> > 
> > Device under test:     old  -> with this patch
> > DP->DVI (OUI 001cf8):  40ms -> 35ms
> > DP->VGA (OUI 0022b9):  45ms -> 38ms
> > Zotac DP->2xHDMI:      25ms ->  4ms
> > Asus PB278 monitor:    22ms ->  3ms
> > 
> > A back of the envelope calculation shows that peak theoretical transfer rate
> > for 1 byte reads is around 60 kbit/s; with 16 byte reads, this increases to
> > around 500 kbit/s, which explains the increase in speed.
> > 
> > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=55228
> > Tested-by: Aidan Marks <[e-mail]>
> > Signed-off-by: Simon Farnsworth <[e-mail]>
> > ---
> > 
> > v4 changes:
> > 
> >  * Change short reply algorithm after suggestions from Ville.
> > 
> >  * Expanded commit message.
> > 
> >  * Mark the module parameter unsafe.
> > 
> >  * Use clamp() to bring the module parameter into range when used.
> > 
> > v3 changes, after feedback from Ville and more testing of Windows:
> > 
> >  * Change the short reply algorithm to match Ville's description of the
> >    DisplayPort 1.2 spec wording.
> > 
> >  * Add a module parameter to set the default transfer size for
> >    experiments. Requested over IRC by Ville.
> > 
> > No-one's been able to find a device that does short replies, but experiments
> > show that bigger reads are faster on most devices. Ville got:
> > 
> >  DP->DVI (OUI 001cf8):  40ms -> 35ms
> >  DP->VGA (OUI 0022b9):  45ms -> 38ms
> >  Zotac DP->2xHDMI:      25ms ->  4ms
> > 
> > v2 changes, after feedback from Thierry and Ville:
> > 
> >  * Handle short replies. I've decided (arbitrarily) that a short reply
> >    results in us dropping back to the newly chosen size for the rest of this
> >    I2C transaction. Thus, given an attempt to read the first 16 bytes of
> >    EDID, and a sink that only does 4 bytes of buffering, we will see the
> >    following AUX transfers for the EDID read (after address is set):
> > 
> >    <set address, block etc>
> >    Read 16 bytes from I2C over AUX.
> >    Reply with 4 bytes
> >    Read 4 bytes
> >    Reply with 4 bytes
> >    Read 4 bytes
> >    Reply with 4 bytes
> >    Read 4 bytes
> >    Reply with 4 bytes
> >    <end I2C transaction>
> > 
> > Note that I've not looked at MST support - I have neither the DP 1.2 spec
> > nor any MST branch devices, so I can't test anything I write or check it
> > against a spec. It looks from the code, however, as if MST has the branch
> > device do the split from a big request into small transactions.
> > 
> >  drivers/gpu/drm/drm_dp_helper.c | 76 +++++++++++++++++++++++++++++++----------
> >  include/drm/drm_dp_helper.h     |  5 +++
> >  2 files changed, 63 insertions(+), 18 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_dp_helper.c b/drivers/gpu/drm/drm_dp_helper.c
> > index 79968e3..105fd66 100644
> > --- a/drivers/gpu/drm/drm_dp_helper.c
> > +++ b/drivers/gpu/drm/drm_dp_helper.c
> > @@ -396,11 +396,13 @@ static u32 drm_dp_i2c_functionality(struct i2c_adapter *adapter)
> >   * retrying the transaction as appropriate.  It is assumed that the
> >   * aux->transfer function does not modify anything in the msg other than the
> >   * reply field.
> > + *
> > + * Returns bytes transferred on success, or a negative error code on failure.
> >   */
> >  static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
> >  {
> >  	unsigned int retry;
> > -	int err;
> > +	int ret;
> >  
> >  	/*
> >  	 * DP1.2 sections 2.7.7.1.5.6.1 and 2.7.7.1.6.6.1: A DP Source device
> > @@ -409,14 +411,14 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
> >  	 */
> >  	for (retry = 0; retry < 7; retry++) {
> >  		mutex_lock(&aux->hw_mutex);
> > -		err = aux->transfer(aux, msg);
> > +		ret = aux->transfer(aux, msg);
> >  		mutex_unlock(&aux->hw_mutex);
> > -		if (err < 0) {
> > -			if (err == -EBUSY)
> > +		if (ret < 0) {
> > +			if (ret == -EBUSY)
> >  				continue;
> >  
> > -			DRM_DEBUG_KMS("transaction failed: %d\n", err);
> > -			return err;
> > +			DRM_DEBUG_KMS("transaction failed: %d\n", ret);
> > +			return ret;
> >  		}
> >  
> >  
> > @@ -457,9 +459,7 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
> >  			 * Both native ACK and I2C ACK replies received. We
> >  			 * can assume the transfer was successful.
> >  			 */
> > -			if (err < msg->size)
> > -				return -EPROTO;
> > -			return 0;
> > +			return ret;
> >  
> >  		case DP_AUX_I2C_REPLY_NACK:
> >  			DRM_DEBUG_KMS("I2C nack\n");
> > @@ -482,14 +482,55 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
> >  	return -EREMOTEIO;
> >  }
> >  
> > +/*
> > + * Keep retrying drm_dp_i2c_do_msg until all data has been transferred.
> > + *
> > + * Returns an error code on failure, or a recommended transfer size on success.
> > + */
> > +static int drm_dp_i2c_drain_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *orig_msg)
> 
> orig_msg should be const to make sure someone doesn't change it by
> accident since that would break drm_dp_i2c_xfer().
> 
> Otherwise looks good to me, so
> Reviewed-by: Ville Syrjälä <[e-mail]>
> 
> I also tested this version with a few dongles and everything seems to
> work as intended.

Did this fall through the cracks, or are we waiting for some resolution
to that one bug?

Daniel Vetter March 11, 2015, 9:06 p.m. UTC | #7

On Wed, Mar 11, 2015 at 09:28:20PM +0200, Ville Syrjälä wrote:
> On Wed, Feb 11, 2015 at 02:05:21PM +0200, Ville Syrjälä wrote:
> > On Tue, Feb 10, 2015 at 06:38:08PM +0000, Simon Farnsworth wrote:
> > > Older DisplayPort to DVI-D Dual Link adapters designed by Bizlink have bugs
> > > in their I2C over AUX implementation (fixed in newer revisions). They work
> > > fine with Windows, but fail with Linux.
> > > 
> > > It turns out that they cannot keep an I2C transaction open unless the
> > > previous read was 16 bytes; shorter reads can only be followed by a zero
> > > byte transfer ending the I2C transaction.
> > > 
> > > Copy Windows's behaviour, and read 16 bytes at a time. If we get a short
> > > reply, assume that there's a hardware bottleneck, and shrink our read size
> > > to match. For this purpose, use the algorithm in the DisplayPort 1.2 spec,
> > > in the hopes that it'll be closest to what Windows does.
> > > 
> > > Also provide an unsafe module parameter for testing smaller transfer sizes,
> > > in case there are sinks out there that cannot work with Windows.
> > > 
> > > Note also that despite the previous comment in drm_dp_i2c_xfer, this speeds
> > > up native DP EDID reads; Ville Syrjälä <[e-mail]> found
> > > the following changes in his testing:
> > > 
> > > Device under test:     old  -> with this patch
> > > DP->DVI (OUI 001cf8):  40ms -> 35ms
> > > DP->VGA (OUI 0022b9):  45ms -> 38ms
> > > Zotac DP->2xHDMI:      25ms ->  4ms
> > > Asus PB278 monitor:    22ms ->  3ms
> > > 
> > > A back of the envelope calculation shows that peak theoretical transfer rate
> > > for 1 byte reads is around 60 kbit/s; with 16 byte reads, this increases to
> > > around 500 kbit/s, which explains the increase in speed.
> > > 
> > > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=55228
> > > Tested-by: Aidan Marks <[e-mail]>
> > > Signed-off-by: Simon Farnsworth <[e-mail]>
> > > ---
> > > 
> > > v4 changes:
> > > 
> > >  * Change short reply algorithm after suggestions from Ville.
> > > 
> > >  * Expanded commit message.
> > > 
> > >  * Mark the module parameter unsafe.
> > > 
> > >  * Use clamp() to bring the module parameter into range when used.
> > > 
> > > v3 changes, after feedback from Ville and more testing of Windows:
> > > 
> > >  * Change the short reply algorithm to match Ville's description of the
> > >    DisplayPort 1.2 spec wording.
> > > 
> > >  * Add a module parameter to set the default transfer size for
> > >    experiments. Requested over IRC by Ville.
> > > 
> > > No-one's been able to find a device that does short replies, but experiments
> > > show that bigger reads are faster on most devices. Ville got:
> > > 
> > >  DP->DVI (OUI 001cf8):  40ms -> 35ms
> > >  DP->VGA (OUI 0022b9):  45ms -> 38ms
> > >  Zotac DP->2xHDMI:      25ms ->  4ms
> > > 
> > > v2 changes, after feedback from Thierry and Ville:
> > > 
> > >  * Handle short replies. I've decided (arbitrarily) that a short reply
> > >    results in us dropping back to the newly chosen size for the rest of this
> > >    I2C transaction. Thus, given an attempt to read the first 16 bytes of
> > >    EDID, and a sink that only does 4 bytes of buffering, we will see the
> > >    following AUX transfers for the EDID read (after address is set):
> > > 
> > >    <set address, block etc>
> > >    Read 16 bytes from I2C over AUX.
> > >    Reply with 4 bytes
> > >    Read 4 bytes
> > >    Reply with 4 bytes
> > >    Read 4 bytes
> > >    Reply with 4 bytes
> > >    Read 4 bytes
> > >    Reply with 4 bytes
> > >    <end I2C transaction>
> > > 
> > > Note that I've not looked at MST support - I have neither the DP 1.2 spec
> > > nor any MST branch devices, so I can't test anything I write or check it
> > > against a spec. It looks from the code, however, as if MST has the branch
> > > device do the split from a big request into small transactions.
> > > 
> > >  drivers/gpu/drm/drm_dp_helper.c | 76 +++++++++++++++++++++++++++++++----------
> > >  include/drm/drm_dp_helper.h     |  5 +++
> > >  2 files changed, 63 insertions(+), 18 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/drm_dp_helper.c b/drivers/gpu/drm/drm_dp_helper.c
> > > index 79968e3..105fd66 100644
> > > --- a/drivers/gpu/drm/drm_dp_helper.c
> > > +++ b/drivers/gpu/drm/drm_dp_helper.c
> > > @@ -396,11 +396,13 @@ static u32 drm_dp_i2c_functionality(struct i2c_adapter *adapter)
> > >   * retrying the transaction as appropriate.  It is assumed that the
> > >   * aux->transfer function does not modify anything in the msg other than the
> > >   * reply field.
> > > + *
> > > + * Returns bytes transferred on success, or a negative error code on failure.
> > >   */
> > >  static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
> > >  {
> > >  	unsigned int retry;
> > > -	int err;
> > > +	int ret;
> > >  
> > >  	/*
> > >  	 * DP1.2 sections 2.7.7.1.5.6.1 and 2.7.7.1.6.6.1: A DP Source device
> > > @@ -409,14 +411,14 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
> > >  	 */
> > >  	for (retry = 0; retry < 7; retry++) {
> > >  		mutex_lock(&aux->hw_mutex);
> > > -		err = aux->transfer(aux, msg);
> > > +		ret = aux->transfer(aux, msg);
> > >  		mutex_unlock(&aux->hw_mutex);
> > > -		if (err < 0) {
> > > -			if (err == -EBUSY)
> > > +		if (ret < 0) {
> > > +			if (ret == -EBUSY)
> > >  				continue;
> > >  
> > > -			DRM_DEBUG_KMS("transaction failed: %d\n", err);
> > > -			return err;
> > > +			DRM_DEBUG_KMS("transaction failed: %d\n", ret);
> > > +			return ret;
> > >  		}
> > >  
> > >  
> > > @@ -457,9 +459,7 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
> > >  			 * Both native ACK and I2C ACK replies received. We
> > >  			 * can assume the transfer was successful.
> > >  			 */
> > > -			if (err < msg->size)
> > > -				return -EPROTO;
> > > -			return 0;
> > > +			return ret;
> > >  
> > >  		case DP_AUX_I2C_REPLY_NACK:
> > >  			DRM_DEBUG_KMS("I2C nack\n");
> > > @@ -482,14 +482,55 @@ static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
> > >  	return -EREMOTEIO;
> > >  }
> > >  
> > > +/*
> > > + * Keep retrying drm_dp_i2c_do_msg until all data has been transferred.
> > > + *
> > > + * Returns an error code on failure, or a recommended transfer size on success.
> > > + */
> > > +static int drm_dp_i2c_drain_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *orig_msg)
> > 
> > orig_msg should be const to make sure someone doesn't change it by
> > accident since that would break drm_dp_i2c_xfer().
> > 
> > Otherwise looks good to me, so
> > Reviewed-by: Ville Syrjälä <[e-mail]>
> > 
> > I also tested this version with a few dongles and everything seems to
> > work as intended.
> 
> Did this fall through the cracks, or are we waiting for some resolution
> to that one bug?

Yeah it sounds a lot like wrong kernel boot from reading bugzilla. I
merged this to drm-misc, we can easily back it out if it causes real
trouble. And the speedup itself is nice already imo.
-Daniel

Patch

diff --git a/drivers/gpu/drm/drm_dp_helper.c b/drivers/gpu/drm/drm_dp_helper.c
index 79968e3..105fd66 100644
--- a/drivers/gpu/drm/drm_dp_helper.c
+++ b/drivers/gpu/drm/drm_dp_helper.c
@@ -396,11 +396,13 @@  static u32 drm_dp_i2c_functionality(struct i2c_adapter *adapter)
  * retrying the transaction as appropriate.  It is assumed that the
  * aux->transfer function does not modify anything in the msg other than the
  * reply field.
+ *
+ * Returns bytes transferred on success, or a negative error code on failure.
  */
 static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
 {
 	unsigned int retry;
-	int err;
+	int ret;
 
 	/*
 	 * DP1.2 sections 2.7.7.1.5.6.1 and 2.7.7.1.6.6.1: A DP Source device
@@ -409,14 +411,14 @@  static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
 	 */
 	for (retry = 0; retry < 7; retry++) {
 		mutex_lock(&aux->hw_mutex);
-		err = aux->transfer(aux, msg);
+		ret = aux->transfer(aux, msg);
 		mutex_unlock(&aux->hw_mutex);
-		if (err < 0) {
-			if (err == -EBUSY)
+		if (ret < 0) {
+			if (ret == -EBUSY)
 				continue;
 
-			DRM_DEBUG_KMS("transaction failed: %d\n", err);
-			return err;
+			DRM_DEBUG_KMS("transaction failed: %d\n", ret);
+			return ret;
 		}
 
 
@@ -457,9 +459,7 @@  static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
 			 * Both native ACK and I2C ACK replies received. We
 			 * can assume the transfer was successful.
 			 */
-			if (err < msg->size)
-				return -EPROTO;
-			return 0;
+			return ret;
 
 		case DP_AUX_I2C_REPLY_NACK:
 			DRM_DEBUG_KMS("I2C nack\n");
@@ -482,14 +482,55 @@  static int drm_dp_i2c_do_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *msg)
 	return -EREMOTEIO;
 }
 
+/*
+ * Keep retrying drm_dp_i2c_do_msg until all data has been transferred.
+ *
+ * Returns an error code on failure, or a recommended transfer size on success.
+ */
+static int drm_dp_i2c_drain_msg(struct drm_dp_aux *aux, struct drm_dp_aux_msg *orig_msg)
+{
+	int err, ret = orig_msg->size;
+	struct drm_dp_aux_msg msg = *orig_msg;
+
+	while (msg.size > 0) {
+		err = drm_dp_i2c_do_msg(aux, &msg);
+		if (err <= 0)
+			return err == 0 ? -EPROTO : err;
+
+		if (err < msg.size && err < ret) {
+			DRM_DEBUG_KMS("Partial I2C reply: requested %zu bytes got %d bytes\n",
+				      msg.size, err);
+			ret = err;
+		}
+
+		msg.size -= err;
+		msg.buffer += err;
+	}
+
+	return ret;
+}
+
+/*
+ * Bizlink designed DP->DVI-D Dual Link adapters require the I2C over AUX
+ * packets to be as large as possible. If not, the I2C transactions never
+ * succeed. Hence the default is maximum.
+ */
+static int dp_aux_i2c_transfer_size __read_mostly = DP_AUX_MAX_PAYLOAD_BYTES;
+module_param_unsafe(dp_aux_i2c_transfer_size, int, 0644);
+MODULE_PARM_DESC(dp_aux_i2c_transfer_size,
+		 "Number of bytes to transfer in a single I2C over DP AUX CH message, (1-16, default 16)");
+
 static int drm_dp_i2c_xfer(struct i2c_adapter *adapter, struct i2c_msg *msgs,
 			   int num)
 {
 	struct drm_dp_aux *aux = adapter->algo_data;
 	unsigned int i, j;
+	unsigned transfer_size;
 	struct drm_dp_aux_msg msg;
 	int err = 0;
 
+	dp_aux_i2c_transfer_size = clamp(dp_aux_i2c_transfer_size, 1, DP_AUX_MAX_PAYLOAD_BYTES);
+
 	memset(&msg, 0, sizeof(msg));
 
 	for (i = 0; i < num; i++) {
@@ -507,20 +548,19 @@  static int drm_dp_i2c_xfer(struct i2c_adapter *adapter, struct i2c_msg *msgs,
 		err = drm_dp_i2c_do_msg(aux, &msg);
 		if (err < 0)
 			break;
-		/*
-		 * Many hardware implementations support FIFOs larger than a
-		 * single byte, but it has been empirically determined that
-		 * transferring data in larger chunks can actually lead to
-		 * decreased performance. Therefore each message is simply
-		 * transferred byte-by-byte.
+		/* We want each transaction to be as large as possible, but
+		 * we'll go to smaller sizes if the hardware gives us a
+		 * short reply.
 		 */
-		for (j = 0; j < msgs[i].len; j++) {
+		transfer_size = dp_aux_i2c_transfer_size;
+		for (j = 0; j < msgs[i].len; j += msg.size) {
 			msg.buffer = msgs[i].buf + j;
-			msg.size = 1;
+			msg.size = min(transfer_size, msgs[i].len - j);
 
-			err = drm_dp_i2c_do_msg(aux, &msg);
+			err = drm_dp_i2c_drain_msg(aux, &msg);
 			if (err < 0)
 				break;
+			transfer_size = err;
 		}
 		if (err < 0)
 			break;
diff --git a/include/drm/drm_dp_helper.h b/include/drm/drm_dp_helper.h
index 11f8c84..444d51b 100644
--- a/include/drm/drm_dp_helper.h
+++ b/include/drm/drm_dp_helper.h
@@ -42,6 +42,8 @@ 
  * 1.2 formally includes both eDP and DPI definitions.
  */
 
+#define DP_AUX_MAX_PAYLOAD_BYTES	16
+
 #define DP_AUX_I2C_WRITE		0x0
 #define DP_AUX_I2C_READ			0x1
 #define DP_AUX_I2C_STATUS		0x2
@@ -519,6 +521,9 @@  struct drm_dp_aux_msg {
  * transactions. The drm_dp_aux_register_i2c_bus() function registers an
  * I2C adapter that can be passed to drm_probe_ddc(). Upon removal, drivers
  * should call drm_dp_aux_unregister_i2c_bus() to remove the I2C adapter.
+ * The I2C adapter uses long transfers by default; if a partial response is
+ * received, the adapter will drop down to the size given by the partial
+ * response for this transaction only.
  *
  * Note that the aux helper code assumes that the .transfer() function
  * only modifies the reply field of the drm_dp_aux_msg structure.  The