On Fri, May 30, 2025 at 03:26:55PM +0100, Jonathan Cameron wrote: > On Tue, 20 May 2025 14:39:47 +0100 > Jonathan Cameron via <qemu-devel@nongnu.org> wrote: > > > On Thu, 8 May 2025 00:00:56 +0000 > > anisa.su...@gmail.com wrote: > > > > > From: Anisa Su <anisa...@samsung.com> > > > > > > This patchset adds support for 6 FM API DCD Management commands > > > (0x5600-0x5605) > > > according to the CXL r3.2 Spec. It is based on the following branch: > > > https://gitlab.com/jic23/qemu/-/tree/cxl-2025-02-20. > > > > Nice work - this is very clean and well presented. > > > > I would like Fan to take a look an provide some tags as DCD modeling in > > QEMU is more in his area of expertise than mine! > > > > I think this series is much more ready for upstream than much of > > my staging cxl tree so I'll rebase it to go near the front. > > > > Today I'm focused on getting test cases to Richard for the TCG issues but > > after that I'll spin a new tree (probably pushed out under a name > > that makes it clear there is a known nasty problem though!) > > Anisa, > > Ideally I'd like to get the majority of this upstream once you post > a v3. To do that I'd like to break the dependence on mctp_i2c in patch 2. > These commands are equally valid via a switch-cci (and that path to the > fm owned ld_mctpcci is already upstream). > > If you could send it as two series. One of which sits on upstream and > doesn't include the i2c_mctp part and the other of which is applied once > that support is available in my tree that would be great. I'll squash > the i2c_mctp part into the patch that adds support for that in the > first place. > > I'll deal with rebasing needed for other command introductions etc. > > There are just enough open questions in this series that I haven't done > this rework directly (though I did start doing it - hence the missing > include note in one of the patches). > > Thanks, > Sounds good, will do :)
Thanks, Anisa > Jonathan > > > > Jonathan > > > > > > > > > > The code was tested with libcxlmi, which runs in the QEMU VM and sends > > > 56xxh > > > commands to the device (QEMU emulated) through MCTP messages over I2C > > > bus. To perform end-to-end tests, both MCTP and DCD support are needed > > > for the kernel, so the needed MCTP patches are applied on top of Ira's DCD > > > branch https://github.com/weiny2/linux-kernel/tree/dcd-v4-2024-12-11. > > > > > > For the tests of commands 0x5600 (Get DCD Info), 0x5601 (Get Host DC > > > Region > > > Config), and 0x5603 (Get DC Region Extent Lists), DCD kernel code is not > > > involved. > > > The libcxlmi test program is used to send the command to the device and > > > results > > > are collected and verified. > > > > > > For command 0x5602 (Set DC Region Config): device creates an event record > > > with type > > > DC_EVENT_REGION_CONFIG_UPDATED and triggers an interrupt to the host > > > if the configuration changes as a result of the command. Currently, the > > > kernel > > > version used to test this only supports Add/Release type events. Thus, > > > this > > > request essentially gets ignored but did not cause problems besides the > > > host > > > not knowing about the configuration change when tested. > > > > > > For the command 0x5604 (Initiate DC Add) and 0x5605 (Initiate DC > > > Release), the > > > tests involve libcxlmi test program (acting as the FM), kernel DCD > > > code (host) and QEMU device. The test workflow follows that in cxl r3.2 > > > section > > > 7.6.7.6.5 and 7.6.7.6.6. More specifically, the tests involve following > > > steps, > > > 1. Start a VM with CXL topology: > > > https://github.com/moking/cxl-test-tool/blob/main/utils/cxl.py#L54. > > > 2. Load the CXL related drivers in the VM; > > > 3. Create a DC region for the DCD device attached. > > > 4. add/release DC extents by sending 0x5604 and 0x5605 respectively > > > through > > > the out-of-tree libcxlmi test program > > > (https://github.com/anisa-su993/libcxlmi/blob/dcd_management_cmds/tests/test-fmapi.c). > > > 5. Check and verify the extents by retrieving the extents list through > > > command 0x5603 in the test program. > > > > > > The remaining 3 commands in this series (0x5606-0x5608) are related to > > > tags > > > and sharing, thus have not been implemented. > > > > > > Changes > > > ================================================================================ > > > v1 -> v2: > > > 1. Feedback from Jonathan Cameron on v1 > > > Addressed general style concerns (newlines/spacing, minor refactoring, > > > etc.) > > > 1.1. Changes Related to 0x5600 - FMAPI Get DCD Info > > > - Squashed prepatory patch adding supported_blk_sizes_bitmask > > > - Added new prepatory patch moving opcodes enum from > > > cxl-mailbox-utils.c to > > > new header file opcodes.h > > > Needed for the check in i2c_mctp_cxl.c to ensure the FMAPI Commands > > > (0x51 - 0x59) are bound with MCTP_MT_CXL_FMAPI. By moving the enum, > > > the hardcoded values (0x51, 0x59) can be replaced with their > > > enumerators. > > > - Bug fix to return Add/Release Extent Selection Policy bitmasks > > > correctly > > > 1.2. Changes Related to 0x5601 - FMAPI Get Host Region Config > > > - Prepatory patch to add dsmas_flags to CXLDCRegion struct was > > > modified to > > > store the booleans dsmas_flags is made up of instead of copying it > > > from the > > > CDAT for that region. Values hardcoded for unsupported flags. > > > - Build the returned dsmas_flags from the new booleans. > > > 1.3. Changes Related to 0x5602 - FMAPI Set DC Region Config > > > - Added locking for CXLDCRegion bitmap for the case that extents are > > > being > > > added/released via a different CCI than that of the FM-enabled CCI. > > > - Prepatory patch created for the above (quite short, can be squashed > > > if > > > preferred) > > > - Added a check to verify that the requested block_size is supported > > > by the > > > region by looking at region->supported_blk_sizes_bitmask > > > - Instead of event_record validity flag being cleared, set to 1 > > > - Fixed bug of forgetting to update region->block_size > > > 1.4. Changes Related to 0x5603 - FMAPI Get DC Region Extents > > > - Minor refactoring of loop filling in response payload extents > > > > > > 2. Feedback from Fan Ni and Jonathan Cameron on v1 > > > 2.1. Changes Related to 0x5604 - FMAPI Initiate DC Add > > > - Remove redundant storage of extents in event_rec_exts > > > - Refactor event record creation into helper function for re-use by > > > release > > > - Return event_record.available_extents > > > (total_extents_available - num_pending - num_accepted) instead of > > > leaving it blank > > > 2.2. Changes Related to 0x5605 - FMAPI Initiate DC Release > > > - Remove redundant storage of extents in event_rec_exts/redundant 2nd > > > loop > > > - Add #define for removal_policy_bitmask instead of hardcoding 0x7 > > > > > > Anisa Su (10): > > > cxl-mailbox-utils: Move opcodes enum to new header file > > > cxl-mailbox-utils: 0x5600 - FMAPI Get DCD Info > > > cxl/type3: Add dsmas_flags to CXLDCRegion struct > > > cxl-mailbox-utils: 0x5601 - FMAPI Get Host Region Config > > > cxl_events.h: Move definition for dynamic_capacity_uuid and enum for > > > DC event types > > > hw/cxl_type3: Add DC Region bitmap lock > > > cxl-mailbox-utils: 0x5602 - FMAPI Set DC Region Config > > > cxl-mailbox-utils: 0x5603 - FMAPI Get DC Region Extent Lists > > > cxl-mailbox-utils: 0x5604 - FMAPI Initiate DC Add > > > cxl-mailbox-utils: 0x5605 - FMAPI Initiate DC Release > > > > > > hw/cxl/cxl-mailbox-utils.c | 649 +++++++++++++++++++++++++++++++---- > > > hw/cxl/i2c_mctp_cxl.c | 6 +- > > > hw/mem/cxl_type3.c | 41 ++- > > > include/hw/cxl/cxl_device.h | 24 ++ > > > include/hw/cxl/cxl_events.h | 15 + > > > include/hw/cxl/cxl_mailbox.h | 6 + > > > include/hw/cxl/cxl_opcodes.h | 72 ++++ > > > 7 files changed, 724 insertions(+), 89 deletions(-) > > > create mode 100644 include/hw/cxl/cxl_opcodes.h > > > > > > > >
