usb: typec: ucsi: add support for Cypress CCGx
PATCH v10 2/2 usb: typec: ucsi: add support for Cypress CCGx
Hi Peter
Quoted 95 lines · Peter Rosin
Quoted 89 lines · Ajay Gupta
Hi Peter,
Quoted 85 lines
Quoted 37 lines
Quoted 37 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Review
Quoted 7 lines
(quoted diff) +7
-0+static int ucsi_ccg_send_data(struct ucsi_ccg *uc) { + unsigned char buf1[USBC_MSG_OUT_SIZE]; + unsigned char buf2[USBC_CONTROL_SIZE]; + int status; + u16 rab; + + memcpy(buf1, (u8 *)(uc->ppm.data) +
USBC_MSG_OUT_OFFSET,Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
sizeof(buf1));
Quoted 1 line
+ memcpy(buf2, (u8 *)(uc->ppm.data) +
USBC_CONTROL_OFFSET,Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 1 line
+sizeof(buf2));
Hmm, now that I see what this function does, instead of just seeing a bunch of magic numbers, I wonder why you make copies instead of feeding the correct section of the ppm.data buffer directly to
ccg_write, like you do below for recv?Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 1 line
Ok, will fix.
Hmm, now that I see this again, it makes me wonder why you complained about copying the buffer to fix the misunderstanding of the
i2c_transferinterface, when you already copy the buffer in the firstplace?
Quoted 4 lines
Quoted 4 lines
Copy is indeed not needed. I will fix it in next version. We will have to do copy in ccg_write()if we try to combine two write
i2c_msginto one and I want to rather stay with twoi2c_msgto avoidcopy.
Quoted 8 lines
Quoted 2 lines
Also
master_xfer()will become tricky since rab write for read alsp has to gofirst.
You are stuck with the construction of the extended buffer. See my mail in the
1/2 thread.Quoted 8 lines
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
+ rab =
CCGX_I2C_RAB_UCSI_DATA_BLOCK(USBC_VERSION_OFFSET);Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 1 line
+ status = ccg_read(uc, rab, (u8 *)(uc->ppm.data) +
USBC_VERSION_OFFSET,Quoted 1 line
+
USBC_VERSION_SIZE);E.g.
rab = CCGX_I2C_RAB_UCSI_DATA_BLOCK(offsetof(struct
ucsi_data,Quoted 21 lines
Quoted 21 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
version)); status = ccg_read(uc, rab, (u8 *)&uc->ppm.data->version, sizeof(uc->ppm.data->version));Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 5 lines
Quoted 4 lines
Hmm, but this highlights that you are not doing any endian conversion of the fields in that struct as you read/write it.
Do you need to in case you have an endian mismatch?
Looks like don't need it. I have tested it and it works as is.
Yeah, but have you tested the driver on a machine with the other byte-
sex?
Quoted 8 lines
Quoted 1 line
No, I think better to convert to desired endian.
The device has a specific endianess. The host cpu has a specific endianess. You transfer data byte-by-byte to/from a struct that appears to have multi- byte integers, e.g. the 16-bit version. You do not do any conversion that I see and you report that it works. So, there are two cases. Either
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
1. your host cpu and the device has the same endianess, and it all just works by accident
or
2. whatever is consuming the ppm data does the endian conversion for
you
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
on "the other side", and it all just works by design.
I have no idea which it is since I know nothing about whatever handles the ppm data on the other side of that
ucsi_register_ppmcall. So,I asked.
Quoted 5 lines
Quoted 5 lines
UCSI specification requires the ppm data to be in little-endian format.
Below is from the UCSI specification.
"All multiple byte fields in this specification are interpreted as and moved over the bus in little-endian order, i.e., LSB to MSB unlessotherwise specified"
Taking another peek into the UCSI spec, and I do not find any mention of any rab. So, I think the rab is out of scope for that spec. I.e. that the rab falls into this bucket:
Quoted 6 lines
This specification does not define the method to use (PCIe/ACPI/I2C/etc.) in order to interface with the PPM. It is left to individual system manufacturers to determine what bus/protocol they use to expose the PPM.
What abbreviation is rab anyway? Register Address Block?
Yes
Quoted 14 lines
Quoted 14 lines
Quoted 9 lines
Quoted 5 lines
Do we still need any conversion here? The ppm data is now directly fed for read and write both and rab should be in little endian as per macro.
#defineCCGX_RAB_UCSI_DATA_BLOCK(offset)(0xf000| ((offset) &0xff))What do you mean by "in little endian as per macro"? Should not the non-offset
0xf0byte ofCCGX_RAB_UCSI_DATA_BLOCKbe in the other byte of rabShouldn't it be always in bits D[8:15] of rab so that it gets written to buf[1] (2nd byte) in ccg_read/write() ?
Quoted 2 lines
compared to e.g. the
0x06byte ofCCGX_RAB_INTR_REG?
Quoted 7 lines
Quoted 3 lines
Quoted 2 lines
I assumed *all*
CCGX_RAB_... defines to be in cpu-native endian. Are they not?How to know/confirm this?
I don't know what you want to have on the wire, but for the code as written, a rab that is
CCGX_RAB_UCSI_DATA_BLOCK(0x10)gets the bytes 0x10,0xf0 and for a rab that isCCGX_RAB_INTR_REGyou get 0x06,0x00
I guess you meant0x10 at lower byte D0:7 and0xf0 at upper byte D8:15
Similarly,0x06 at D0:70x00 at D8:15
I don’t see any issue here. Consider a 16 bit offset where CCGX_RAB_INTR_REG is at 0x0006 and other one at 0xf010
Quoted 11 lines
What makes me a little worried is that the offset of the data-block rab appears where the significant bits of the other rabs are. However, that might not be a problem at all since the second byte of the data-block rab (
0xf0) appears to be distinct from all other rabs. It just looks like the data-block rab might have been byte-swapped when comparing it to all the other rabs you have defined.If you can run the code and see that it works, then obviously you are getting the byte-ordering of the rabs correct. No?
Yes, they work as is on multiple systems.
Thanks
Ajay
—nvpublic
Quoted 3 lines
Cheers,
Peter