Lu Baolu wrote:
> Hi,
>
> I am writing code in rootnex to implement Intel IOMMU. The register set
> for each remapping hardware unit is placed at a 4KB-aligned memory
> mapped location. I try to map the register set with an allocated virtual
> address space so the CPU can access it.
>
> My mapping function is put in rootnex_attach(), and the function code is,
>
> static caddr_t
> iommu_mapin_reg(paddr_t addr, size_t len)
> {
> uint_t pgoffset;
> pgcnt_t npages;
> caddr_t cvaddr;
>
> if (len == 0)
> return (0);
>
> pgoffset = addr & MMU_PAGEOFFSET;
> npages = mmu_btopr(len + pgoffset);
> cvaddr = device_arena_alloc(ptob(npages), VM_NOSLEEP);
> if (cvaddr == NULL)
> return (0);
> hat_devload(kas.a_hat, cvaddr, mmu_ptob(npages), mmu_btop(addr),
> PROT_READ | PROT_WRITE, HAT_LOAD_LOCK);
>
> cmn_err(CE_CONT, "iommu_mapin_reg: 0x%" PRIx64 ", %d @ 0x%"
> PRIx64 "\n",
> (uint64_t)addr, (uint_t)len, (uint64_t)(uintptr_t)cvaddr);
>
> return (cvaddr + pgoffset);
> }
>
> The output of cmn_err() in above function during boot is,
>
> iommu_mapin_reg: 0xfe710000 @ 0xffffff049dd20000
> iommu_mapin_reg: 0xfe714000 @ 0xffffff049dd25000
> iommu_mapin_reg: 0xfe719000 @ 0xffffff049dd26000
> iommu_mapin_reg: 0xfe71a000 @ 0xffffff049dd27000
> iommu_mapin_reg: 0xfe718000 @ 0xffffff049dd28000
>
> But when I use mdb to check the mappings:
>
> [2]> 0xfe710000 \K
> 0xfe710000: 10
> [2]> 0xffffff049dd20000 /K
> kmdb: failed to read data from target: no mapping for address
> 0xffffff049dd20000:
>
> Mdb access the physical address gets the right data (which is a
> remapping unit version, 0x10), but the access through virtual address
> fails. I checked the onnv source tree and got to know that ioapic
> register address (start from 0xfee0000) is mapped in the same way. And
> mdb can access the register apace through the virtual address (I have
> verified this in mdb).
What you have looks right, what does a ::vtop report?
i.e.
[1]> 0xffffff028e96e100::vtop
virtual ffffff028e96e100 mapped to physical d8600100
MRJ
> PS. I didn't use ddi_regs_map_setup() since what I am writing is not a
> driver. Also, ddi_regs_map_setup() will at last be routed to rootnex and
> map in the similar way as my mapin function.
>
> Please advice.
>
> Thanks,
> -baolu
>
>
_______________________________________________
opensolaris-code mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
