Arm VM Multicore Improvements #84
Conversation
Apave24
force-pushed
the
arm-vm-multicore
branch
from
4b6fc4b
to
e828a2c
Compare
Apave24
force-pushed
the
arm-vm-multicore
branch
from
e828a2c
to
882e6c3
Compare
| smc_set_return_value(®s, PSCI_SUCCESS); | ||
| } else { | ||
| ZF_LOGE("[vCPU %u] no unused physical core left", vcpu->vcpu_id); |
There was a problem hiding this comment.
...or starting failed. I think separating the errors is worth it, because that make debugging easier when something fails here or got out if sync somehow.
| @@ -92,6 +92,15 @@ int vm_assign_vcpu_target(vm_vcpu_t *vcpu, int target_cpu) | |||
| ZF_LOGE("Failed to assign target cpu - A VCPU is already assigned to core %d", target_cpu); | |||
| return -1; | |||
| } | |||
|
|
|||
| #if CONFIG_MAX_NUM_NODES > 1 | |||
| int err = seL4_TCB_SetAffinity(vcpu->tcb.tcb.cptr, target_cpu); | |||
There was a problem hiding this comment.
I think to commit is ok. But it makes this generic code now, how do other architectures cope with this change then? Do we gave to change something in the x86 VMM part also?
There was a problem hiding this comment.
I believe vm_assign_vcpu_target is currently only used by ARM
|
Since this PR has multiple commits that seem independent, but have dependencies on other PRs, could you split this up that we can merge things independently and avoid unnecessary blocking? |
Apave24
force-pushed
the
arm-vm-multicore
branch
from
882e6c3
to
1f1f86f
Compare
Dropped commit |
|
Bumping this PR. I believe everything has been addressed. |
|
@Apave24 could you rebase this, then we also get another CI run there to see if this passes. |
Apave24
force-pushed
the
arm-vm-multicore
branch
from
1f1f86f
to
f3b8e13
Compare
chrisguikema
force-pushed
the
arm-vm-multicore
branch
from
f3b8e13
to
61bba4c
Compare
Calling seL4_TCB_SetAffinity in the vm_assign_vcpu_target function allows the vcpu to correctly run on the cpu it is assigned to. Previously the vcpu was getting assigned to the same core as it's ID. Signed-off-by: Alex Pavey <[email protected]>
Previously the target_cpu member was being used to determine if a vcpu was started in conjunction with the is_vcpu_online function. This is unneccessary and has been repurposed to keep track of whether or not the vcpu has been previously assigned a pcpu or not. Signed-off-by: Alex Pavey <[email protected]>
Signed-off-by: Alex Pavey <[email protected]>
chrisguikema
force-pushed
the
arm-vm-multicore
branch
from
61bba4c
to
2094fef
Compare
|
@axel-h @abrandnewusername @kent-mcleod Hey guys, just wanted to ping on this PR. This series of PRs makes it so that when the VM kicks off a secondary core, the VMM pins the VCPU to a PCPU defined in the |
These commits allow for more flexible core configurations for Arm VM systems