Re: [PATCH 4/8] i3c: mipi-i3c-hci: Add a quirk to clear the TX start threshold

[Frank Li <Frank.li@oss.nxp.com>]

On Tue, Sep 01, 2026 at 07:35:31PM +0800, Billy Tsai wrote:
> The DATA_TX_START_THLD field of the PIO data buffer threshold register
> holds off the start of a TX transfer until the FIFO holds a set amount
> of data, which in PIO mode cuts down on the number of software writes.
> The field resets to 0x1, requiring at least (2 ^ 2) DWORDs (16 bytes)
> in the FIFO before transmission starts.
>
> HCI controllers that support both PIO and DMA can expose the two as
> separate register blocks; nothing in the specification says selecting
> DMA mode disables the PIO block's own gating logic. On ASPEED
> platforms it doesn't: DATA_TX_START_THLD still holds up transfer
> start regardless of which mode feeds the FIFO, and in DMA mode that
> threshold is never satisfied for some transfer sizes:
>
>  - 1-4 bytes: uses the Immediate Data Transfer Command.
>  - 13+ bytes: since the hardware fetches data in 4-byte chunks, a
>    13-byte transfer fetches 16 bytes into the FIFO and reaches the
>    threshold.
>  - 5-12 bytes: the threshold is never reached and the transfer stalls.
>
> Add HCI_QUIRK_TX_START_THLD to clear the field whenever DMA mode is
> selected. The clear lives in i3c_hci_set_io_mode() rather than in
> probe so the reset-and-restore recovery path also reapplies it after
> a controller soft reset. Move the PIO Access Area register and
> bitfield definitions from pio.c to a new pio.h so this quirk can reuse
> PIO_DATA_BUFFER_THLD_CTRL and DATA_TX_START_THLD instead of
> redefining them.
>
> Signed-off-by: Billy Tsai <billy_tsai@aspeedtech.com>
> Assisted-by: Claude:claude-fable-5
> ---
>  drivers/i3c/master/mipi-i3c-hci/core.c |  14 +++++
>  drivers/i3c/master/mipi-i3c-hci/hci.h  |   1 +
>  drivers/i3c/master/mipi-i3c-hci/pio.c  |  97 +----------------------------
>  drivers/i3c/master/mipi-i3c-hci/pio.h  | 109 +++++++++++++++++++++++++++++++++
>  4 files changed, 125 insertions(+), 96 deletions(-)
>
> diff --git a/drivers/i3c/master/mipi-i3c-hci/core.c b/drivers/i3c/master/mipi-i3c-hci/core.c
> index c03c3a9cbe4f..2290a889701c 100644
> --- a/drivers/i3c/master/mipi-i3c-hci/core.c
> +++ b/drivers/i3c/master/mipi-i3c-hci/core.c
> @@ -26,6 +26,7 @@
>  #include "cmd.h"
>  #include "dat.h"
>  #include "ibi.h"
> +#include "pio.h"
>
>  /*
>   * Host Controller Capabilities and Operation Registers
> @@ -823,6 +824,19 @@ static int i3c_hci_set_io_mode(struct i3c_hci *hci, bool dma)
>  	else
>  		reg_set(HC_CONTROL, HC_CONTROL_PIO_MODE);
>
> +	/*
> +	 * On the ASPEED AST2700 the TX start threshold gates transfer start
> +	 * even in DMA mode. Clear it so DMA transfers are not held back
> +	 * waiting for a PIO FIFO level that will never be reached.
> +	 */
> +	if (dma && (hci->quirks & HCI_QUIRK_TX_START_THLD) && hci->PIO_regs) {
> +		void __iomem *thld_reg = hci->PIO_regs + PIO_DATA_BUFFER_THLD_CTRL;
> +		u32 thld_val = readl(thld_reg);
> +
> +		thld_val &= ~DATA_TX_START_THLD;
> +		writel(thld_val, thld_reg);

don't move these defination, original isolate is quite good. Add API or
callback to implement clean DATA_TX_START_THLD.

Frank