Re: [Qemu-block] [PATCH v3 0/9] ahci: enable migration
Am 30.04.2015 um 20:07 hat John Snow geschrieben: The day we all feared is here, and I am proposing we allow the migration of the AHCI device tentatively for the 2.4 development window. There are some more NCQ migration tests are needed, but I felt that it was important to get migration enabled as close to the start of the 2.4 development window as possible. If the NCQ patches don't pan out by the time the 2.4 freeze occurs, we can revert the migration boolean and add a conditional around the ahci tests that rely on the migration feature being enabled. I am justifying this checkin based on a series of ping-pong migration tests I ran under heavy load (using google's stressapptest) and saw over 300 successful migrations without a single failure. This series does a few things: (1) Add migration facilities to libqos (2) Enable AHCI and ICH9 migration (3) Add a series of migration tests to ahci-test Reviewed-by: Kevin Wolf kw...@redhat.com I think besides the NCQ tests, we'll definitely also want to test migration with READ DMA in flight (this series tests only WRITE DMA and FLUSH CACHE). Probably also discard. Nice to have, but less important, would be the other variants that exist, like the EXT version of each command, and READ/WRITE SECTOR. But all of that can be added in a follow-up series, it's not a reason to hold up what's already there. Kevin
Re: [Qemu-block] [PATCH v3 0/9] ahci: enable migration
On 05/04/2015 08:29 AM, Kevin Wolf wrote: Am 30.04.2015 um 20:07 hat John Snow geschrieben: The day we all feared is here, and I am proposing we allow the migration of the AHCI device tentatively for the 2.4 development window. There are some more NCQ migration tests are needed, but I felt that it was important to get migration enabled as close to the start of the 2.4 development window as possible. If the NCQ patches don't pan out by the time the 2.4 freeze occurs, we can revert the migration boolean and add a conditional around the ahci tests that rely on the migration feature being enabled. I am justifying this checkin based on a series of ping-pong migration tests I ran under heavy load (using google's stressapptest) and saw over 300 successful migrations without a single failure. This series does a few things: (1) Add migration facilities to libqos (2) Enable AHCI and ICH9 migration (3) Add a series of migration tests to ahci-test Reviewed-by: Kevin Wolf kw...@redhat.com I think besides the NCQ tests, we'll definitely also want to test migration with READ DMA in flight (this series tests only WRITE DMA and FLUSH CACHE). Probably also discard. Added to the todo list, thanks. Nice to have, but less important, would be the other variants that exist, like the EXT version of each command, and READ/WRITE SECTOR. But all of that can be added in a follow-up series, it's not a reason to hold up what's already there. Kevin Thanks! --js
[Qemu-block] [PATCH v3 0/9] ahci: enable migration
The day we all feared is here, and I am proposing we allow the migration of the AHCI device tentatively for the 2.4 development window. There are some more NCQ migration tests are needed, but I felt that it was important to get migration enabled as close to the start of the 2.4 development window as possible. If the NCQ patches don't pan out by the time the 2.4 freeze occurs, we can revert the migration boolean and add a conditional around the ahci tests that rely on the migration feature being enabled. I am justifying this checkin based on a series of ping-pong migration tests I ran under heavy load (using google's stressapptest) and saw over 300 successful migrations without a single failure. This series does a few things: (1) Add migration facilities to libqos (2) Enable AHCI and ICH9 migration (3) Add a series of migration tests to ahci-test v3: - Rebase and resend for 2.4. - Minor style guide fix. v2: - Added a URI parameter to the migrate() helper - Adjust ahci_shutdown to set qtest context for itself - Make sure verify() is part of ahci_migrate() and redundant calls are eliminated - Add new helpers to make tests with blkdebug injections more succint - Change the flush migrate test to not load the blkdebug rule on the destination host - Modify the migrate() function so that it does not poll the VM for migration status if it can rely on RESUME events. - New patch: Repair the ahci_command_set_offset helper. - New test: Test DMA halt and resume. - New test: Test DMA halt, migrate, and resume. == For convenience, this branch is available at: https://github.com/jnsnow/qemu.git branch ahci-migration-test https://github.com/jnsnow/qemu/tree/ahci-migration-test This version is tagged ahci-migration-test-v3: https://github.com/jnsnow/qemu/releases/tag/ahci-migration-test-v3 == John Snow (9): libqos/ahci: Add halted command helpers libqos/ahci: Fix sector set method libqos: Add migration helpers ich9/ahci: Enable Migration qtest/ahci: Add migration test qtest/ahci: add migrate dma test qtest/ahci: add flush migrate test qtest/ahci: add halted dma test qtest/ahci: add migrate halted dma test hw/ide/ahci.c | 1 - hw/ide/ich.c | 1 - tests/ahci-test.c | 318 +- tests/libqos/ahci.c | 34 +- tests/libqos/ahci.h | 3 + tests/libqos/libqos.c | 84 + tests/libqos/libqos.h | 2 + tests/libqos/malloc.c | 74 +--- tests/libqos/malloc.h | 1 + 9 files changed, 496 insertions(+), 22 deletions(-) -- 2.1.0
