On 2022/08/11 12:05, matoro wrote: > Just a small update, the module stuff turned out to be a separate, > unrelated regression. I bisected that one also (applying these reverts > each time to allow me to boot) and reported it to Masahiro, who put in a > fix for it here: > https://lore.kernel.org/all/20220809141117.641543-1-masahiroy@xxxxxxxxxx/ > . So you can ignore that stuff. Are these two commits still planned to > be reverted? Revert should be a last resort (I really want to get rid of all these sparse warnings !). Let me first try to generate a fix for you to test. > > -------- Original Message -------- > Subject: Re: [PATCH v3 0/5] Fix mpt3sas driver sparse warnings > Date: 2022-08-02 19:36 > From: Damien Le Moal <damien.lemoal@xxxxxxxxxxxxxxxxxx> > To: matoro <matoro_mailinglist_kernel@xxxxxxxxx>, Sreekanth Reddy > <sreekanth.reddy@xxxxxxxxxxxx> > > On 8/3/22 05:27, matoro wrote: >> Hi folks, sorry for the lateness, unfortunately this is in fact broken >> on BE. I use mpt3sas on sparc and my drives fail to come up on 5.19, >> bisected to this patchset. Reverting both of the endian-related >> commits, b4efbec4c2a75b619fae4e8768be379e88c78687 and >> 7ab4d2441b952977556672c2fe3f4c2a698cbb37, allows it to boot. However, >> after booting, I can't load any modules - everything errors with >> "disagrees about version of symbol module_layout". I have completely >> wiped out kernel sources, the module tree, and the kernel image, >> rebuilding both from scratch with ONLY the revert patch applied, but I >> still can't load any modules. Presumably it would work with >> CONFIG_MODVERSIONS=n, but these CRC checks are there for a reason and I >> can't tell if it has something to do with the revert or not. > > For b4efbec4c2a75b619fae4e8768be379e88c78687, removing the cpu_to_le32() > call results in the bytes actually being reversed by writel()/readl() > for > your BE machine. So it looks like the values that need to be written to > the HBA have to be in CPU endian, not le32. Should be easy to fix. > And for 7ab4d2441b952977556672c2fe3f4c2a698cbb37, this looks like the > same > problem. > > I will be traveling and busy this week, but I can have a look at a fix > next Monday. If the Broadcom folks can send a fix faster than that, that > is of course welcome :) > >> >> $ modprobe --dump-modversions >> /lib/modules/5.19.0-gentoo-sparc64/kernel/fs/openpromfs/openpromfs.ko >> | >> grep "module_layout" >> 0xa6c23707 module_layout >> $ grep "module_layout" /usr/src/linux/Module.symvers >> 0xa6c23707 module_layout vmlinux EXPORT_SYMBOL >> >> If you need real hardware access and do not have any on hand, please >> reach out and I can provide temporary access! >> >> Here is the full error on vanilla 5.19: >> >> mpt3sas version 42.100.00.00 loaded >> mpt2sas_cm0: 64 BIT PCI BUS DMA ADDRESSING SUPPORTED, total mem >> (32650280 kB) >> mpt2sas_cm0: _base_wait_for_doorbell_not_used: failed due to timeout >> count(5000), doorbell_reg(18000000)! >> mpt2sas_cm0: CurrentHostPageSize is 0: Setting default host page size >> to >> 4k >> mpt2sas_cm0: MSI-X vectors supported: 1 >> no of cores: 64, max_msix_vectors: -1 >> mpt2sas_cm0: 0 1 1 >> mpt2sas_cm0: pci_alloc_irq_vectors failed (r=-22) !!! >> mpt2sas_cm0: High IOPs queues : disabled >> mpt2sas0: IO-APIC enabled: IRQ 4 >> mpt2sas_cm0: iomem(0x0000084100000000), mapped(0x(____ptrval____)), >> size(16384) >> mpt2sas_cm0: ioport(0x0000085100000000), size(256) >> mpt2sas_cm0: doorbell is in use (line=6869) >> mpt2sas_cm0: _base_get_ioc_facts: handshake failed (r=-14) >> mpt2sas_cm0: failure at >> drivers/scsi/mpt3sas/mpt3sas_scsih.c:12336/_scsih_probe()! >> mpt2sas_cm1: 64 BIT PCI BUS DMA ADDRESSING SUPPORTED, total mem >> (32650280 kB) >> mpt2sas_cm1: _base_wait_for_doorbell_not_used: failed due to timeout >> count(5000), doorbell_reg(18000000)! >> mpt2sas_cm1: CurrentHostPageSize is 0: Setting default host page size >> to >> 4k >> mpt2sas_cm1: MSI-X vectors supported: 1 >> no of cores: 64, max_msix_vectors: -1 >> mpt2sas_cm1: 0 1 1 >> mpt2sas_cm1: pci_alloc_irq_vectors failed (r=-22) !!! >> mpt2sas_cm1: High IOPs queues : disabled >> mpt2sas1: IO-APIC enabled: IRQ 5 >> mpt2sas_cm1: iomem(0x0000084120000000), mapped(0x(____ptrval____)), >> size(16384) >> mpt2sas_cm1: ioport(0x0000085100002000), size(256) >> mpt2sas_cm1: doorbell is in use (line=6869) >> mpt2sas_cm1: _base_get_ioc_facts: handshake failed (r=-14) >> mpt2sas_cm1: failure at >> drivers/scsi/mpt3sas/mpt3sas_scsih.c:12336/_scsih_probe()! >> >> -------- Original Message -------- >> Subject: Re: [PATCH v3 0/5] Fix mpt3sas driver sparse warnings >> Date: 2022-03-09 01:35 >> From: Sreekanth Reddy <sreekanth.reddy@xxxxxxxxxxxx> >> To: "Martin K. Petersen" <martin.petersen@xxxxxxxxxx> >> >> On Wed, Mar 9, 2022 at 9:26 AM Martin K. Petersen >> <martin.petersen@xxxxxxxxxx> wrote: >>> >>> >>> Sreekanth, >>> >>>> This series fix (remove) all sparse warnings generated when compiling >>>> the mpt3sas driver. All warnings are related to __iomem access and >>>> endianness. >>> >>> Please review this series and validate the patch 5 modification. >> >> Martin, >> This patch set looks good, but before acknowledging this patch set I >> just wanted to do some basic testing on a big endian machine. >> Currently I don't have a big endian machine, internally I am checking >> to get access to big endian machines. Meanwhile if anyone does a basic >> testing on any big endian machine then please let me know. I will add >> the acknowledgement signature. >> >> Thanks, >> Sreekanth >> >>> >>> Thanks! >>> >>> -- >>> Martin K. Petersen Oracle Linux Engineering > -- Damien Le Moal Western Digital Research
