| Message ID | 20180806123445.1459-1-peter.maydell@linaro.org |
|---|---|
| Headers | show |
| Series | Arm migration fixes for 3.0 | expand |
On 08/06/2018 05:34 AM, Peter Maydell wrote: > Peter Maydell (5): > hw/intc/arm_gicv3_common: Give no-migration-shift-bug subsection a > needed function > hw/intc/arm_gicv3_common: Combine duplicate .subsections in > vmstate_gicv3_cpu > target/arm: Add dummy needed functions to M profile vmstate > subsections > hw/intc/arm_gicv3_common: Move post_load hooks to top-level VMSD > hw/intc/arm_gicv3_common: Move gicd shift bug handling to > gicv3_post_load Reviewed-by: Richard Henderson <richard.henderson@linaro.org> r~
On 6 August 2018 at 15:06, Richard Henderson <richard.henderson@linaro.org> wrote: > On 08/06/2018 05:34 AM, Peter Maydell wrote: >> Peter Maydell (5): >> hw/intc/arm_gicv3_common: Give no-migration-shift-bug subsection a >> needed function >> hw/intc/arm_gicv3_common: Combine duplicate .subsections in >> vmstate_gicv3_cpu >> target/arm: Add dummy needed functions to M profile vmstate >> subsections >> hw/intc/arm_gicv3_common: Move post_load hooks to top-level VMSD >> hw/intc/arm_gicv3_common: Move gicd shift bug handling to >> gicv3_post_load > > Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Thanks; series applied to master for rc4. -- PMM
This patchset primarily fixes problems with Arm migration induced by a bug in the core vmstate handling of subsections: currently the migration code incorrectly treats a subsection with no .needed function pointer as if it was the subsection list terminator -- it is ignored and so is everything after it. I did an audit of all uses of subsections in QEMU, and found that we had four which didn't define a .needed function (assuming that this meant "always needed", same as the semantics for not providing a .needed function for a toplevel vmsd). This patchset fixes them all up by providing a dummy needed function. It also fixes an error in vmstate_gicv3_cpu which was accidentally initializing .subsections twice and so ignoring one of the subsections. Patches 1..3 are the same as for v1 and have been reviewed. Patches 4 and 5 fix a further bug which I discovered during testing of 1..3: the GICv3 migration structs had several uses of the pre_load and post_load hooks which assumed that the hooks were run whether the subsection was present or not. In fact the migration code only runs the hooks when the subsection is present in the incoming migration stream, and so for correct behaviour we need to move the code we were running in the hooks up to the parent VMSDs. I've now tested this by doing a save/load of: 2.12.0 QEMU -> new QEMU new QEMU -> new QEMU An on-trunk QEMU prior to these bugfixes won't migrate to a QEMU with the bugfixes, but we never released any final QEMU version with the bugs in. thanks -- PMM Peter Maydell (5): hw/intc/arm_gicv3_common: Give no-migration-shift-bug subsection a needed function hw/intc/arm_gicv3_common: Combine duplicate .subsections in vmstate_gicv3_cpu target/arm: Add dummy needed functions to M profile vmstate subsections hw/intc/arm_gicv3_common: Move post_load hooks to top-level VMSD hw/intc/arm_gicv3_common: Move gicd shift bug handling to gicv3_post_load hw/intc/arm_gicv3_common.c | 284 ++++++++++++++++++------------------- target/arm/machine.c | 3 + 2 files changed, 145 insertions(+), 142 deletions(-)