在 2025/5/28 下午5:53, Stefano Garzarella 写道:
On Wed, May 28, 2025 at 05:34:49PM +0800, Qunqin Zhao wrote:
在 2025/5/28 下午5:24, Qunqin Zhao 写道:
在 2025/5/28 下午5:00, Stefano Garzarella 写道:
On Wed, May 28, 2025 at 04:42:05PM +0800, Qunqin Zhao wrote:
在 2025/5/28 下午3:57, Stefano Garzarella 写道:
+ chip = tpmm_chip_alloc(dev, &tpm_loongson_ops);
+ if (IS_ERR(chip))
+ return PTR_ERR(chip);
+ chip->flags = TPM_CHIP_FLAG_TPM2 | TPM_CHIP_FLAG_IRQ;
Why setting TPM_CHIP_FLAG_IRQ?
When tpm_engine completes TPM_CC* command,
the hardware will indeed trigger an interrupt to the kernel.
IIUC that is hidden by loongson_se_send_engine_cmd(), that for this
driver is completely synchronous, no?
IIUC this driver is similar to ftpm and svsm where the send is
synchronous so having .status, .cancel, etc. set to 0 should be
enough to call .recv() just after send() in tpm_try_transmit().
See commit 980a573621ea ("tpm: Make
chip->{status,cancel,req_canceled} opt")
The send callback would wait until the TPM_CC* command complete.
We don't need a poll.
Right, that's what I was saying too, send() is synchronous (as in
ftpm and svsm). The polling in tpm_try_transmit() is already
skipped since we are setting .status = 0, .req_complete_mask = 0,
.req_complete_val = 0, etc. so IMHO this is exactly the same of
ftpm and svsm, so we don't need to set TPM_CHIP_FLAG_IRQ.
I see, but why not skip polling directly in "if (chip->flags &
TPM_CHIP_FLAG_IRQ)" instead of do while?
I mean, why not skip polling directly in "if (chip->flags &
TPM_CHIP_FLAG_IRQ)"?
And In my opinion, TPM_CHIP_FLAG_SYNC and TPM_CHIP_FLAG_IRQ are
essentially the same, only with different names.
When TPM_CHIP_FLAG_SYNC is defined, the .recv() is not invoked and
.send() will send the command and retrieve the response. For some
driver like ftpm this will save an extra copy/buffer.
I need to copy the data to the DMA data buffer. So my suggestion is to
let the vendor specific driver decide whether to use the SYNC or IRQ flag.
IRQ flag is fine for me.
Thanks,
Qunqin.
Stefano
