On Mon, Dec 16, 2024 at 7:56:45PM +0200 Ilpo Järvinen <ilpo.jarvinen@xxxxxxxxxxxxxxx> wrote: > From: Ilpo Järvinen <ilpo.jarvinen@xxxxxxxxxxxxxxx> > > Hi all, > > This series focuses on PCI resource fitting and assignment algorithms. > I've further changes in works to enable handling resizable BARs better during > resource fitting built on top of these, but that's still WIP and this series seems > way too large as is to have more stuff included. > > First there are small tweaks and fixes to the relaxed tail alignment code and > applying the lessons learned to other similar cases. They are sort of > independent of the rest. Then a large set of pure cleanups and refactoring that > are not intended to make any functional changes. > Finally, starting from "PCI: Extend enable to check for any optional resource" are > again patches that aim to make behavioral changes to fix bridge window sizing > to consider expansion ROM as an optional resource (to fix a remove/rescan > cycle issue) and improve resource fitting algorithm in general. > > The series includes one of the change from Michał Winiarski > <michal.winiarski@xxxxxxxxx> as these changes also touch the same IOV checks. > > Please let me know if you'd prefer me to order the changes differently or split it > into smaller chunks. > > > I've extensively tested this series over the hosts in our lab which have quite > heterogeneous PCI setup each. There were no losses of any important resource. > Without pci=realloc, there's some churn in which of the disabled expansion > ROMs gets a scarce memory space assigned (with pci=realloc, they are all > assigned large enough bridge window). > > > Ilpo Järvinen (24): > PCI: Remove add_align overwrite unrelated to size0 > PCI: size0 is unrelated to add_align > PCI: Simplify size1 assignment logic > PCI: Optional bridge window size too may need relaxing > PCI: Fix old_size lower bound in calculate_iosize() too > PCI: Use SZ_* instead of literals in setup-bus.c > PCI: resource_set_range/size() conversions > PCI: Check resource_size() separately > PCI: Add pci_resource_num() helper > PCI: Add dev & res local variables to resource assignment funcs > PCI: Converge return paths in __assign_resources_sorted() > PCI: Refactor pdev_sort_resources() & __dev_sort_resources() > PCI: Use while loop and break instead of gotos > PCI: Rename retval to ret > PCI: Consolidate assignment loop next round preparation > PCI: Remove wrong comment from pci_reassign_resource() > PCI: Add restore_dev_resource() > PCI: Extend enable to check for any optional resource > PCI: Always have realloc_head in __assign_resources_sorted() > PCI: Indicate optional resource assignment failures > PCI: Add debug print when releasing resources before retry > PCI: Use res->parent to check is resource is assigned > PCI: Perform reset_resource() and build fail list in sync > PCI: Rework optional resource handling > > Michał Winiarski (1): > PCI: Add a helper to identify IOV resources > > drivers/pci/pci.h | 44 +++- > drivers/pci/setup-bus.c | 566 +++++++++++++++++++++++----------------- > drivers/pci/setup-res.c | 8 +- > 3 files changed, 364 insertions(+), 254 deletions(-) > Hi, all This series has undergone testing on the following configurations: - Lenovo ThinkSystem SR630 V4 equipped with Intel Granite Rapids CPUs. - The latest upstream kernel v6.14-rc2 and the stable kernel 6.13.2. - With "pci=realloc" appended to the kernel command line. - Red Hat Enterprise Linux 10.0 Beta. Test results: - All patches were applied cleanly and the build process was successful. - The assignment for the ROM BAR of downstream devices was successful. - The bridge window has been adjusted to fit the downstream resources. Tested-by: Xiaochun Lee <lixc17@xxxxxxxxxx> > -- > 2.39.5 >
