On Fri, May 13, 2016 at 2:11 PM, Bharata B Rao <bharata....@gmail.com> wrote:
> On Wed, May 4, 2016 at 12:22 PM, Alexey Kardashevskiy <a...@ozlabs.ru> wrote:
>> +static void rtas_ibm_create_pe_dma_window(PowerPCCPU *cpu,
>> + sPAPRMachineState *spapr,
>> + uint32_t token, uint32_t nargs,
>> + target_ulong args,
>> + uint32_t nret, target_ulong rets)
>> +{
>> + sPAPRPHBState *sphb;
>> + sPAPRTCETable *tcet = NULL;
>> + uint32_t addr, page_shift, window_shift, liobn;
>> + uint64_t buid;
>> +
>> + if ((nargs != 5) || (nret != 4)) {
>> + goto param_error_exit;
>> + }
>> +
>> + buid = ((uint64_t)rtas_ld(args, 1) << 32) | rtas_ld(args, 2);
>> + addr = rtas_ld(args, 0);
>> + sphb = spapr_pci_find_phb(spapr, buid);
>> + if (!sphb || !sphb->ddw_enabled) {
>> + goto param_error_exit;
>> + }
>> +
>> + page_shift = rtas_ld(args, 3);
>> + window_shift = rtas_ld(args, 4);
>
> Kernel has a bug due to which wrong window_shift gets returned here. I
> have posted possible fix here:
> https://patchwork.ozlabs.org/patch/621497/
>
> I have tried to work around this issue in QEMU too
> https://lists.nongnu.org/archive/html/qemu-ppc/2016-04/msg00226.html
>
> But the above work around involves changing the memory representation
> in DT. Hence I feel until the guest kernel changes are available, a
> simpler work around would be to discard the window_shift value above
> and recalculate the right value as below:
>
> if (machine->ram_size == machine->maxram_size) {
> max_window_size = machine->ram_size;
> } else {
> MemoryHotplugState *hpms = &spapr->hotplug_memory;
> max_window_size = hpms->base + memory_region_size(&hpms->mr);
> }
> window_shift = max_window_size >> SPAPR_TCE_PAGE_SHIFT;
>
> and create DDW based on this calculated window_shift value. Does that
> sound reasonable ?
Sorry, missed mentioning earlier that incorrect DDW value here causes
memory hotplug to fail.
>
> Regards,
> Bharata.
--
http://raobharata.wordpress.com/
