Hi Alexander,
On Fri, 2017-12-08 at 19:46 +0100, Alexander Steffen wrote:
> There do not seem to be any real limits one the amount/duration of wait
> states that the TPM is allowed to generate. So at least give the TPM some
> time to clean up the situation that caused the wait states instead of
> retrying the transfers as fast as possible.
Without this patch, the TPM performance on the pi is amazing! A
thousand extends takes ~6.5 seconds. Unfortunately, the same thousand
extends with this patch takes > 2+ minutes. TPM_TIMEOUT (5 msecs) is
a really long time. Why so long?
Mimi
> Signed-off-by: Alexander Steffen <Alexander.Steffen@xxxxxxxxxxxx>
> ---
> drivers/char/tpm/tpm_tis_spi.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/char/tpm/tpm_tis_spi.c b/drivers/char/tpm/tpm_tis_spi.c
> index 2cc6aa9..e53c9c3 100644
> --- a/drivers/char/tpm/tpm_tis_spi.c
> +++ b/drivers/char/tpm/tpm_tis_spi.c
> @@ -88,7 +88,7 @@ static int tpm_tis_spi_transfer(struct tpm_tis_data *data, u32 addr, u16 len,
>
> if ((phy->iobuf[3] & 0x01) == 0) {
> // handle SPI wait states
> - for (i = 0; i < TPM_RETRY; i++) {
> + for (i = 1; i <= TPM_RETRY; i++) {
> phy->iobuf[0] = addr;
> spi_xfer.len = 1;
> spi_message_init(&m);
> @@ -98,9 +98,10 @@ static int tpm_tis_spi_transfer(struct tpm_tis_data *data, u32 addr, u16 len,
> goto exit;
> if (phy->iobuf[0] & 0x01)
> break;
> + tpm_msleep(i * TPM_TIMEOUT);
> }
>
> - if (i == TPM_RETRY) {
> + if (i > TPM_RETRY) {
> spi_xfer.tx_buf = NULL;
> spi_xfer.rx_buf = NULL;
> spi_xfer.len = 0;
