spapr/rtas: Add MinMem to ibm, get-system-parameter RTAS call
PATCH 1/1 spapr/rtas: Add MinMem to ibm, get-system-parameter RTAS call
Add support for MinMem SPLPAR Characteristic on emulated RTAS call ibm,get-system-parameter.
MinMem represents Minimum Memory, that is described in LOPAPR as:
The minimum amount of main store that is needed to power on the partition. Minimum memory is expressed in MB of storage.
This provides a way for the OS to discern hotplugged LMBs and LMBs that have started with the VM, allowing it to better provide a way for memory hot-removal.
--- hw/ppc/spapr_rtas.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
hw/ppc/spapr_rtas.c +3 -1
@@ -276,10 +276,12 @@ static void rtas_ibm_get_system_parameter(PowerPCCPU *cpu,switch (parameter) {case RTAS_SYSPARM_SPLPAR_CHARACTERISTICS : { - char *param_val = g_strdup_printf("MaxEntCap=%d," + char *param_val = g_strdup_printf("MinMem=%" PRIu64 "," + "MaxEntCap=%d," "DesMem=%" PRIu64 "," "DesProcs=%d," "MaxPlatProcs=%d", + ms->ram_size / MiB, ms->smp.max_cpus, ms->ram_size / MiB, ms->smp.cpus,
Signature · 1 line
2.24.1
Re: [PATCH 1/1] spapr/rtas: Add MinMem to ibm,get-system-parameter RTAS call
Quoted 6 lines · Leonardo Bras
Add support for MinMem SPLPAR Characteristic on emulated RTAS call ibm,get-system-parameter.
MinMem represents Minimum Memory, that is described in LOPAPR as:
The minimum amount of main store that is needed to power on the partition. Minimum memory is expressed in MB of storage.
Where exactly does LoPAPR say that? The version I'm looking at doesn't describe MinMem all that clearly, other than to say it must be <= DesMem, which currently has the same value here.
Quoted 3 lines
This provides a way for the OS to discern hotplugged LMBs and LMBs that have started with the VM, allowing it to better provide a way for memory hot-removal.
This seems a bit dubious. Surely we should have this information from the dynamic-reconfiguration-memory stuff already? Trying to discern this from purely a number seems very fragile - wouldn't that mean making assumptions about how qemu / the host is laying out it's fixed and dynamic memory which might not be justified?
Quoted 24 lines
Signed-off-by Leonardo Bras <[email protected]>--- hw/ppc/spapr_rtas.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)hw/ppc/spapr_rtas.c +3
-1@@ -276,10 +276,12 @@ static void rtas_ibm_get_system_parameter(PowerPCCPU *cpu,switch (parameter) {case RTAS_SYSPARM_SPLPAR_CHARACTERISTICS : { - char *param_val = g_strdup_printf("MaxEntCap=%d," + char *param_val = g_strdup_printf("MinMem=%" PRIu64 "," + "MaxEntCap=%d," "DesMem=%" PRIu64 "," "DesProcs=%d," "MaxPlatProcs=%d", + ms->ram_size / MiB, ms->smp.max_cpus, ms->ram_size / MiB, ms->smp.cpus,
Signature · 4 lines
David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
Re: [PATCH 1/1] spapr/rtas: Add MinMem to ibm,get-system-parameter RTAS call
Quoted 11 lines · David Gibson
Quoted 6 lines · Leonardo Bras
Add support for MinMem SPLPAR Characteristic on emulated RTAS call ibm,get-system-parameter.
MinMem represents Minimum Memory, that is described in LOPAPR as:
The minimum amount of main store that is needed to power on the partition. Minimum memory is expressed in MB of storage.Where exactly does LoPAPR say that? The version I'm looking at doesn't describe MinMem all that clearly, other than to say it must be <= DesMem, which currently has the same value here.
Please look for "Minimum Memory". It's on Table 237. SPLPAR Terms.
Quoted 9 lines
Quoted 3 lines
This provides a way for the OS to discern hotplugged LMBs and LMBs that have started with the VM, allowing it to better provide a way for memory hot-removal.
This seems a bit dubious. Surely we should have this information from the dynamic-reconfiguration-memory stuff already? Trying to discern this from purely a number seems very fragile - wouldn't that mean making assumptions about how qemu / the host is laying out it's fixed and dynamic memory which might not be justified?
I agree.
I previously sent a RFC proposing the usage of a new flag for this same reason [1], which I thank you for positive feedback.
Even though I think using a flag is a better solution, I am also working in this other option (MinMem), that would use parameter already available on the platform, in case the new flag don't get approved.
So far, using MinMem looks like a very complex solution on kernel side, and I think it's a poor solution.
I decided to send this patch because it's a simple change to the platform support, that should cause no harm and could even be useful to other OSes.
Best regards,
Leonardo
Re: [PATCH 1/1] spapr/rtas: Add MinMem to ibm,get-system-parameter RTAS call
Quoted 9 lines · Leonardo Bras
Quoted 8 lines · David Gibson
Quoted 6 lines · Leonardo Bras
Add support for MinMem SPLPAR Characteristic on emulated RTAS call ibm,get-system-parameter.
MinMem represents Minimum Memory, that is described in LOPAPR as:
The minimum amount of main store that is needed to power on the partition. Minimum memory is expressed in MB of storage.
Quoted 5 lines
Quoted 3 lines
Where exactly does LoPAPR say that? The version I'm looking at doesn't describe MinMem all that clearly, other than to say it must be <= DesMem, which currently has the same value here.
Please look for "Minimum Memory". It's on Table 237. SPLPAR Terms.
Ah, thanks.
Hm, yes. In the way we manage VMs with KVM and qemu, I don't think we cal really draw any meaningful distinction between MinMem and DesMem, so it's reasonble for them to have the same value.
Quoted 18 lines
Quoted 9 lines
Quoted 3 lines
This provides a way for the OS to discern hotplugged LMBs and LMBs that have started with the VM, allowing it to better provide a way for memory hot-removal.
This seems a bit dubious. Surely we should have this information from the dynamic-reconfiguration-memory stuff already? Trying to discern this from purely a number seems very fragile - wouldn't that mean making assumptions about how qemu / the host is laying out it's fixed and dynamic memory which might not be justified?
I agree.
I previously sent a RFC proposing the usage of a new flag for this same reason [1], which I thank you for positive feedback.Even though I think using a flag is a better solution, I am also working in this other option (MinMem), that would use parameter already available on the platform, in case the new flag don't get approved.
Quoted 6 lines
So far, using MinMem looks like a very complex solution on kernel side, and I think it's a poor solution.
I decided to send this patch because it's a simple change to the platform support, that should cause no harm and could even be useful to other OSes.
Hm, ok.
Signature · 4 lines
David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson