[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <000001d18373$b5c920f0$215b62d0$@emc.com>
Date: Mon, 21 Mar 2016 09:15:15 -0400
From: "Allen Hubbe" <Allen.Hubbe@....com>
To: "'kbuild test robot'" <fengguang.wu@...el.com>,
"'Xiangliang Yu'" <Xiangliang.Yu@....com>
Cc: <kbuild-all@...org>, <linux-kernel@...r.kernel.org>,
"'Jon Mason'" <jdmason@...zu.us>
Subject: RE: drivers/ntb/hw/amd/ntb_hw_amd.c:367:29: sparse: cast removes address space of expression
> >> drivers/ntb/hw/amd/ntb_hw_amd.c:367:29: sparse: cast removes address space of
> expression
> drivers/ntb/hw/amd/ntb_hw_amd.c:427:31: sparse: cast removes address space of
> expression
> 360 static int amd_ntb_peer_db_addr(struct ntb_dev *ntb,
> 361 phys_addr_t *db_addr,
> 362 resource_size_t *db_size)
> 363 {
> 364 struct amd_ntb_dev *ndev = ntb_ndev(ntb);
> 365
> 366 if (db_addr)
> > 367 *db_addr = (phys_addr_t)(ndev->peer_mmio + AMD_DBREQ_OFFSET);
This is a good warning. The code here does look wrong (both lines 367 and 427).
Instead of peer_mmio, the code here should use pci_resource_start(pcidev, 0) + AMD_PEER_OFFSET. For comparison, in ntb_hw_intel the offset is based on a physical address stored in peer_addr, not the virtual mapping of that address in peer_mmio.
Look at ntb_hw_amd around lines 1016 where self_mmio and peer_mmio are set up. The register physical address is never saved in the ndev object, only the virtual address is saved, therefore the physical address would need to be retrieved with pci_resoruce_start wherever it is used in the ntb_hw_amd driver.
Alternatively, the offending functions could be deleted until they are added later, after they have been tested. Note that the ntb_foo_addr functions are an optional part of the api for drivers to implement.
Allen
Powered by blists - more mailing lists