On Thu, May 17, 2018 at 11:18:28AM +0530, Raju Rangoju wrote:
>
> + /* The support for new kernel mode SRQ extends the t4_cqe struct
> + * size to 64 Bytes from existing 32 Bytes. But, this change brings
> + * in the library/kernel ABI and 32/64 Byte CQE compatibility issues.
> + *
> + * To address these issues, the user <-> kernel response message
> + * c4iw_create_cq_resp's 'reserved' field has been renamed to 'flags'.
> + *
> + * Newer version of libcxgb4 can pre-initialize the 'flags' field in
> + * the request struct with the flag C4IW_64B_CQE. And a new iw_cxgb4
> + * can detect the new libcxgb4 by checking for the C4IW_64B_CQE flag
> + * in the structure c4iw_create_cq received from userspace.
> + * New kernel driver can not work with older userspace library,
> + * c4iw_create_cq() will log a warning message to upgrade libcxgb4 and
> + * aborts the cq creation in this case.
> + *
> + * Likewise, new iw_cxgb4 which supports 64B CQE will set the flag
> + * C4IW_64B_CQE in c4iw_create_cq_resp struct. So, if the response
> + * from the create_cq verb results in the flags field still being zero,
> + * libcxgb4 can destroy the cq and print a warning and fail the user
> + * create cq verb.
> + */
> +
> + if (ib_context && udata->inlen < sizeof(ucmd)) {
> + pr_err_once("WARNING: Downlevel libcxgb4. RDMA cannot be supported with this driver/lib combination. Please update libcxgb4.\n");
> + return ERR_PTR(-EINVAL);
> + }
Nope, hard NAK on this idea.
The driver must continue to support current user space with full
compatibility.
And user space must generally support all past kernels as well.
Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
