| Message ID | 20210402233702.3291792-1-seanjc@google.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | ccp: KVM: SVM: Use stack for SEV command buffers | expand |
Hi Sean, On 4/2/21 6:36 PM, Sean Christopherson wrote: > While doing minor KVM cleanup to account various kernel allocations, I > noticed that all of the SEV command buffers are allocated via kmalloc(), > even for commands whose payloads is smaller than a pointer. After much > head scratching, the only reason I could come up with for dynamically > allocating the command data is CONFIG_VMAP_STACK=y. > > This series teaches __sev_do_cmd_locked() to gracefully handle vmalloc'd > command buffers by copying such buffers an internal buffer before sending > the command to the PSP. The SEV driver and KVM are then converted to use > the stack for all command buffers. Thanks for the series. Post SNP series, I was going to move all the command buffer allocation to the stack. You are ahead of me :). I can certainly build upon your series. The behavior of the SEV-legacy command is changed when SNP firmware is in the INIT state. All the legacy commands that cause a firmware to write to memory must be in the firmware state before issuing the command. One of my patch in the SNP series is using an internal memory before sending the command to the PSP. Looking forward to the SNP support, may I ask you to remove the vmalloc'd buffer check and use a page for the internal buffer ? In SNP series, I can simply transition the internal page to firmware state before issuing the command. > The first patch is optional, I included it in case someone wants to > backport it to stable kernels. It wouldn't actually fix bugs, but it > would make debugging issues a lot easier if they did pop up. > > Tested everything except sev_ioctl_do_pek_import(), I don't know anywhere > near enough about the PSP to give it the right input. > > Based on kvm/queue, commit f96be2deac9b ("KVM: x86: Support KVM VMs > sharing SEV context") to avoid a minor conflict. > > Sean Christopherson (5): > crypto: ccp: Detect and reject vmalloc addresses destined for PSP > crypto: ccp: Reject SEV commands with mismatching command buffer > crypto: ccp: Play nice with vmalloc'd memory for SEV command structs > crypto: ccp: Use the stack for small SEV command buffers > KVM: SVM: Allocate SEV command structures on local stack > > arch/x86/kvm/svm/sev.c | 262 +++++++++++++---------------------- > drivers/crypto/ccp/sev-dev.c | 161 ++++++++++----------- > drivers/crypto/ccp/sev-dev.h | 7 + > 3 files changed, 184 insertions(+), 246 deletions(-) >
While doing minor KVM cleanup to account various kernel allocations, I noticed that all of the SEV command buffers are allocated via kmalloc(), even for commands whose payloads is smaller than a pointer. After much head scratching, the only reason I could come up with for dynamically allocating the command data is CONFIG_VMAP_STACK=y. This series teaches __sev_do_cmd_locked() to gracefully handle vmalloc'd command buffers by copying such buffers an internal buffer before sending the command to the PSP. The SEV driver and KVM are then converted to use the stack for all command buffers. The first patch is optional, I included it in case someone wants to backport it to stable kernels. It wouldn't actually fix bugs, but it would make debugging issues a lot easier if they did pop up. Tested everything except sev_ioctl_do_pek_import(), I don't know anywhere near enough about the PSP to give it the right input. Based on kvm/queue, commit f96be2deac9b ("KVM: x86: Support KVM VMs sharing SEV context") to avoid a minor conflict. Sean Christopherson (5): crypto: ccp: Detect and reject vmalloc addresses destined for PSP crypto: ccp: Reject SEV commands with mismatching command buffer crypto: ccp: Play nice with vmalloc'd memory for SEV command structs crypto: ccp: Use the stack for small SEV command buffers KVM: SVM: Allocate SEV command structures on local stack arch/x86/kvm/svm/sev.c | 262 +++++++++++++---------------------- drivers/crypto/ccp/sev-dev.c | 161 ++++++++++----------- drivers/crypto/ccp/sev-dev.h | 7 + 3 files changed, 184 insertions(+), 246 deletions(-)