netdev - RE: [RDMA RFC v6 00/16] Intel RDMA Driver Updates 2020-05-19

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <9DD61F30A802C4429A01CA4200E302A7EE04047F@fmsmsx124.amr.corp.intel.com>
Date:   Wed, 27 May 2020 01:58:01 +0000
From:   "Saleem, Shiraz" <shiraz.saleem@...el.com>
To:     Jason Gunthorpe <jgg@...lanox.com>,
        "Kirsher, Jeffrey T" <jeffrey.t.kirsher@...el.com>
CC:     "dledford@...hat.com" <dledford@...hat.com>,
        "davem@...emloft.net" <davem@...emloft.net>,
        "gregkh@...uxfoundation.org" <gregkh@...uxfoundation.org>,
        "linux-rdma@...r.kernel.org" <linux-rdma@...r.kernel.org>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        "nhorman@...hat.com" <nhorman@...hat.com>,
        "sassmann@...hat.com" <sassmann@...hat.com>,
        "poswald@...e.com" <poswald@...e.com>
Subject: RE: [RDMA RFC v6 00/16] Intel RDMA Driver Updates 2020-05-19

> Subject: Re: [RDMA RFC v6 00/16] Intel RDMA Driver Updates 2020-05-19
> 
> On Wed, May 20, 2020 at 12:03:59AM -0700, Jeff Kirsher wrote:
> > This patch set adds a unified Intel Ethernet Protocol Driver for RDMA
> > that supports a new network device E810 (iWARP and RoCEv2 capable) and
> > the existing X722 iWARP device. The driver architecture provides the
> > extensibility for future generations of Intel HW supporting RDMA.
> >
> > This driver replaces the legacy X722 driver i40iw and extends the ABI
> > already defined for i40iw. It is backward compatible with legacy X722
> > rdma-core provider (libi40iw).
> >
> > This series was built against the rdma for-next branch.  This series
> > is dependant upon the v4 100GbE Intel Wired LAN Driver Updates
> > 2020-05-19
> > 12 patch series, which adds virtual_bus interface and ice/i40e LAN
> > driver changes.
> >
> > v5-->v6:
> > *Convert irdma destroy QP to a synchronous API *Drop HMC obj macros
> > for use counts like IRDMA_INC_SD_REFCNT et al.
> > *cleanup unneccesary 'mem' variable in irdma_create_qp *cleanup unused
> > headers such as linux/moduleparam.h et. al *set kernel_ver in
> > irdma_ualloc_resp struct to current ABI ver. Placeholder to  support
> > user-space compatbility checks in future *GENMASK/FIELD_PREP scheme to
> > set WQE descriptor fields considered for irdma  driver but decision to
> > drop. The FIELD_PREP macro cannot be used on the device  bitfield mask
> > array maintained for common WQE descriptors and initialized  based on
> > HW generation. The macro expects compile time constants only.
> 
> The request was to use GENMASK for the #define constants. If you move to a
> code environment then the spot the constant appears in the C code should be
> FIELD_PREP'd into the something dynamic code can use.
>

Maybe I am missing something here, but from what I understood,
the vantage point of using GENMASK for the masks
was so that we could get rid of open coding the shift constants and use the
FIELD_PREP macro to place the value in the field of a descriptor.
This should work for the static masks. So something like --

-#define IRDMA_UDA_QPSQ_INLINEDATALEN_S 48
-#define IRDMA_UDA_QPSQ_INLINEDATALEN_M \
-       ((u64)0xff << IRDMA_UDA_QPSQ_INLINEDATALEN_S)
+#define IRDMA_UDA_QPSQ_INLINEDATALEN_M GENMASK_ULL(55, 48)

-#define LS_64(val, field)      (((u64)(val) << field ## _S) & (field ## _M))
+#define LS_64(val, field)      (FIELD_PREP(val,(field ## _M)))

However we have device's dynamically computed bitfield mask array and shifts
for some WQE descriptor fields --
see icrdma_init_hw.c
https://lore.kernel.org/linux-rdma/20200520070415.3392210-3-jeffrey.t.kirsher@intel.com/#Z30drivers:infiniband:hw:irdma:icrdma_hw.c

we still need to use the custom macro FLD_LS_64 without FIELD_PREP in this case
as FIELD_PREP expects compile time constants.
+#define FLD_LS_64(dev, val, field)	\
+	(((u64)(val) << (dev)->hw_shifts[field ## _S]) & (dev)->hw_masks[field 
+## _M])
And the shifts are still required for these fields which causes a bit of
inconsistency