Message ID | 1475722987-18644-3-git-send-email-david@gibson.dropbear.id.au (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On 06/10/2016 05:03, David Gibson wrote: > Currently the default PCI host bridge for the 'pseries' machine type is > constructed with its IO windows in the 1TiB..(1TiB + 64GiB) range in > guest memory space. This means that if > 1TiB of guest RAM is specified, > the RAM will collide with the PCI IO windows, causing serious problems. > > Problems won't be obvious until guest RAM goes a bit beyond 1TiB, because > there's a little unused space at the bottom of the area reserved for PCI, > but essentially this means that > 1TiB of RAM has never worked with the > pseries machine type. > > This patch fixes this by altering the placement of PHBs on large-RAM VMs. > Instead of always placing the first PHB at 1TiB, it is placed at the next > 1 TiB boundary after the maximum RAM address. > > Technically, this changes behaviour in a migration-breaking way for > existing machines with > 1TiB maximum memory, but since having > 1 TiB > memory was broken anyway, this seems like a reasonable trade-off. Perhaps you can add an SPAPR_COMPAT_XX property with PHB0 base to not break compatibility? I think spapr without PCI card (only VIO, for instance), should work with 1TiB+. On another side, how the SPAPR kernel manages memory? Is it possible to add an hole in RAM between 1TiB and 1TiB+64GiB to allow the kernel to register the I/O space? Laurent
On Thu, Oct 06, 2016 at 09:21:56AM +0200, Laurent Vivier wrote: > > > On 06/10/2016 05:03, David Gibson wrote: > > Currently the default PCI host bridge for the 'pseries' machine type is > > constructed with its IO windows in the 1TiB..(1TiB + 64GiB) range in > > guest memory space. This means that if > 1TiB of guest RAM is specified, > > the RAM will collide with the PCI IO windows, causing serious problems. > > > > Problems won't be obvious until guest RAM goes a bit beyond 1TiB, because > > there's a little unused space at the bottom of the area reserved for PCI, > > but essentially this means that > 1TiB of RAM has never worked with the > > pseries machine type. > > > > This patch fixes this by altering the placement of PHBs on large-RAM VMs. > > Instead of always placing the first PHB at 1TiB, it is placed at the next > > 1 TiB boundary after the maximum RAM address. > > > > Technically, this changes behaviour in a migration-breaking way for > > existing machines with > 1TiB maximum memory, but since having > 1 TiB > > memory was broken anyway, this seems like a reasonable trade-off. > > Perhaps you can add an SPAPR_COMPAT_XX property with PHB0 base to not > break compatibility? I think spapr without PCI card (only VIO, for > instance), should work with 1TiB+. No, it won't because we always create the default PHB even if we don't actually have any PCI devices. > On another side, how the SPAPR kernel manages memory? > Is it possible to add an hole in RAM between 1TiB and 1TiB+64GiB to > allow the kernel to register the I/O space? Possible, yes, but I don't really see any advantage to it.
On 06/10/2016 05:03, David Gibson wrote: > Currently the default PCI host bridge for the 'pseries' machine type is > constructed with its IO windows in the 1TiB..(1TiB + 64GiB) range in > guest memory space. This means that if > 1TiB of guest RAM is specified, > the RAM will collide with the PCI IO windows, causing serious problems. > > Problems won't be obvious until guest RAM goes a bit beyond 1TiB, because > there's a little unused space at the bottom of the area reserved for PCI, > but essentially this means that > 1TiB of RAM has never worked with the > pseries machine type. > > This patch fixes this by altering the placement of PHBs on large-RAM VMs. > Instead of always placing the first PHB at 1TiB, it is placed at the next > 1 TiB boundary after the maximum RAM address. > > Technically, this changes behaviour in a migration-breaking way for > existing machines with > 1TiB maximum memory, but since having > 1 TiB > memory was broken anyway, this seems like a reasonable trade-off. > > Signed-off-by: David Gibson <david@gibson.dropbear.id.au> Reviewed-by: Laurent Vivier <lvivier@redhat.com> > --- > hw/ppc/spapr.c | 5 ++++- > 1 file changed, 4 insertions(+), 1 deletion(-) > > diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c > index f6e9c2a..9f3e004 100644 > --- a/hw/ppc/spapr.c > +++ b/hw/ppc/spapr.c > @@ -2376,12 +2376,15 @@ static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index, > unsigned n_dma, uint32_t *liobns, Error **errp) > { > const uint64_t base_buid = 0x800000020000000ULL; > - const hwaddr phb0_base = 0x10000000000ULL; /* 1 TiB */ > const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */ > const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */ > const hwaddr pio_offset = 0x80000000; /* 2 GiB */ > const uint32_t max_index = 255; > + const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */ > > + uint64_t max_hotplug_addr = spapr->hotplug_memory.base + > + memory_region_size(&spapr->hotplug_memory.mr); > + hwaddr phb0_base = QEMU_ALIGN_UP(max_hotplug_addr, phb0_alignment); > hwaddr phb_base; > int i; > >
diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c index f6e9c2a..9f3e004 100644 --- a/hw/ppc/spapr.c +++ b/hw/ppc/spapr.c @@ -2376,12 +2376,15 @@ static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index, unsigned n_dma, uint32_t *liobns, Error **errp) { const uint64_t base_buid = 0x800000020000000ULL; - const hwaddr phb0_base = 0x10000000000ULL; /* 1 TiB */ const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */ const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */ const hwaddr pio_offset = 0x80000000; /* 2 GiB */ const uint32_t max_index = 255; + const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */ + uint64_t max_hotplug_addr = spapr->hotplug_memory.base + + memory_region_size(&spapr->hotplug_memory.mr); + hwaddr phb0_base = QEMU_ALIGN_UP(max_hotplug_addr, phb0_alignment); hwaddr phb_base; int i;
Currently the default PCI host bridge for the 'pseries' machine type is constructed with its IO windows in the 1TiB..(1TiB + 64GiB) range in guest memory space. This means that if > 1TiB of guest RAM is specified, the RAM will collide with the PCI IO windows, causing serious problems. Problems won't be obvious until guest RAM goes a bit beyond 1TiB, because there's a little unused space at the bottom of the area reserved for PCI, but essentially this means that > 1TiB of RAM has never worked with the pseries machine type. This patch fixes this by altering the placement of PHBs on large-RAM VMs. Instead of always placing the first PHB at 1TiB, it is placed at the next 1 TiB boundary after the maximum RAM address. Technically, this changes behaviour in a migration-breaking way for existing machines with > 1TiB maximum memory, but since having > 1 TiB memory was broken anyway, this seems like a reasonable trade-off. Signed-off-by: David Gibson <david@gibson.dropbear.id.au> --- hw/ppc/spapr.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)