| Message ID | a75bef32-8bbb-4471-b26b-981b06173cd5@suse.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | VT-d: SATC handling and ATS tidying | expand |
On Mon, Feb 05, 2024 at 02:56:36PM +0100, Jan Beulich wrote: > All callers only care about boolean outcome. For this there's no point > in allocating a duplicate of the respective DRHD structure; a simple > boolean suffices (which eventually may wantg to become a count, such ^ want > that the "any ATS devices assigned state" can also clear again). With > that boolean, remove respective parameters from internal helper > functions right away, as those have access to the flag through another > parameter. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> AFAICT the intention is that this is a non-functional change? > > --- a/xen/drivers/passthrough/vtd/extern.h > +++ b/xen/drivers/passthrough/vtd/extern.h > @@ -65,8 +65,6 @@ struct acpi_drhd_unit *ioapic_to_drhd(un > struct acpi_drhd_unit *hpet_to_drhd(unsigned int hpet_id); > struct acpi_rhsa_unit *drhd_to_rhsa(const struct acpi_drhd_unit *drhd); > > -struct acpi_drhd_unit *find_ats_dev_drhd(struct vtd_iommu *iommu); > - > int ats_device(const struct pci_dev *, const struct acpi_drhd_unit *); > > int dev_invalidate_iotlb(struct vtd_iommu *iommu, u16 did, > --- a/xen/drivers/passthrough/vtd/iommu.c > +++ b/xen/drivers/passthrough/vtd/iommu.c > @@ -624,8 +624,7 @@ int cf_check vtd_flush_iotlb_reg( > } > > static int __must_check iommu_flush_iotlb_global(struct vtd_iommu *iommu, > - bool flush_non_present_entry, > - bool flush_dev_iotlb) > + bool flush_non_present_entry) > { > int status; > > @@ -633,7 +632,7 @@ static int __must_check iommu_flush_iotl > vtd_ops_preamble_quirk(iommu); > > status = iommu->flush.iotlb(iommu, 0, 0, 0, DMA_TLB_GLOBAL_FLUSH, > - flush_non_present_entry, flush_dev_iotlb); > + flush_non_present_entry, iommu->flush_dev_iotlb); Any reason to not also remove the parameter from here also? As the handler gets iommu passed as the first parameter anyway. Thanks, Roger.
On 08.02.2024 18:31, Roger Pau Monné wrote: > On Mon, Feb 05, 2024 at 02:56:36PM +0100, Jan Beulich wrote: >> All callers only care about boolean outcome. For this there's no point >> in allocating a duplicate of the respective DRHD structure; a simple >> boolean suffices (which eventually may wantg to become a count, such > ^ want >> that the "any ATS devices assigned state" can also clear again). With >> that boolean, remove respective parameters from internal helper >> functions right away, as those have access to the flag through another >> parameter. >> >> Signed-off-by: Jan Beulich <jbeulich@suse.com> > > AFAICT the intention is that this is a non-functional change? No functional effect intended, yes. Added such a sentence. >> --- a/xen/drivers/passthrough/vtd/extern.h >> +++ b/xen/drivers/passthrough/vtd/extern.h >> @@ -65,8 +65,6 @@ struct acpi_drhd_unit *ioapic_to_drhd(un >> struct acpi_drhd_unit *hpet_to_drhd(unsigned int hpet_id); >> struct acpi_rhsa_unit *drhd_to_rhsa(const struct acpi_drhd_unit *drhd); >> >> -struct acpi_drhd_unit *find_ats_dev_drhd(struct vtd_iommu *iommu); >> - >> int ats_device(const struct pci_dev *, const struct acpi_drhd_unit *); >> >> int dev_invalidate_iotlb(struct vtd_iommu *iommu, u16 did, >> --- a/xen/drivers/passthrough/vtd/iommu.c >> +++ b/xen/drivers/passthrough/vtd/iommu.c >> @@ -624,8 +624,7 @@ int cf_check vtd_flush_iotlb_reg( >> } >> >> static int __must_check iommu_flush_iotlb_global(struct vtd_iommu *iommu, >> - bool flush_non_present_entry, >> - bool flush_dev_iotlb) >> + bool flush_non_present_entry) >> { >> int status; >> >> @@ -633,7 +632,7 @@ static int __must_check iommu_flush_iotl >> vtd_ops_preamble_quirk(iommu); >> >> status = iommu->flush.iotlb(iommu, 0, 0, 0, DMA_TLB_GLOBAL_FLUSH, >> - flush_non_present_entry, flush_dev_iotlb); >> + flush_non_present_entry, iommu->flush_dev_iotlb); > > Any reason to not also remove the parameter from here also? As the handler > gets iommu passed as the first parameter anyway. Indeed, yet then the patch would have grown quite a bit. I think I meant to have a respective post-commit-message remark, but then forgot to actually put one there. Once (if) this change has gone in, a follow-on patch could further tidy tings. (The "right away" in the description was kind of meant to indicate that.) Jan
On Fri, Feb 09, 2024 at 08:06:07AM +0100, Jan Beulich wrote: > On 08.02.2024 18:31, Roger Pau Monné wrote: > > On Mon, Feb 05, 2024 at 02:56:36PM +0100, Jan Beulich wrote: > >> All callers only care about boolean outcome. For this there's no point > >> in allocating a duplicate of the respective DRHD structure; a simple > >> boolean suffices (which eventually may wantg to become a count, such > > ^ want > >> that the "any ATS devices assigned state" can also clear again). With > >> that boolean, remove respective parameters from internal helper > >> functions right away, as those have access to the flag through another > >> parameter. > >> > >> Signed-off-by: Jan Beulich <jbeulich@suse.com> > > > > AFAICT the intention is that this is a non-functional change? > > No functional effect intended, yes. Added such a sentence. Acked-by: Roger Pau Monné <roger.pau@citrix.com> > > >> --- a/xen/drivers/passthrough/vtd/extern.h > >> +++ b/xen/drivers/passthrough/vtd/extern.h > >> @@ -65,8 +65,6 @@ struct acpi_drhd_unit *ioapic_to_drhd(un > >> struct acpi_drhd_unit *hpet_to_drhd(unsigned int hpet_id); > >> struct acpi_rhsa_unit *drhd_to_rhsa(const struct acpi_drhd_unit *drhd); > >> > >> -struct acpi_drhd_unit *find_ats_dev_drhd(struct vtd_iommu *iommu); > >> - > >> int ats_device(const struct pci_dev *, const struct acpi_drhd_unit *); > >> > >> int dev_invalidate_iotlb(struct vtd_iommu *iommu, u16 did, > >> --- a/xen/drivers/passthrough/vtd/iommu.c > >> +++ b/xen/drivers/passthrough/vtd/iommu.c > >> @@ -624,8 +624,7 @@ int cf_check vtd_flush_iotlb_reg( > >> } > >> > >> static int __must_check iommu_flush_iotlb_global(struct vtd_iommu *iommu, > >> - bool flush_non_present_entry, > >> - bool flush_dev_iotlb) > >> + bool flush_non_present_entry) > >> { > >> int status; > >> > >> @@ -633,7 +632,7 @@ static int __must_check iommu_flush_iotl > >> vtd_ops_preamble_quirk(iommu); > >> > >> status = iommu->flush.iotlb(iommu, 0, 0, 0, DMA_TLB_GLOBAL_FLUSH, > >> - flush_non_present_entry, flush_dev_iotlb); > >> + flush_non_present_entry, iommu->flush_dev_iotlb); > > > > Any reason to not also remove the parameter from here also? As the handler > > gets iommu passed as the first parameter anyway. > > Indeed, yet then the patch would have grown quite a bit. I think I > meant to have a respective post-commit-message remark, but then > forgot to actually put one there. Once (if) this change has gone in, > a follow-on patch could further tidy tings. (The "right away" in the > description was kind of meant to indicate that.) Would you mind adding a sentence to the commit message that the vtd_iommu hooks are not modified in order to avoid the patch growing too much? Otherwise it it's not clear why those are not also converted. Thanks, Roger.
--- a/xen/drivers/passthrough/vtd/extern.h +++ b/xen/drivers/passthrough/vtd/extern.h @@ -65,8 +65,6 @@ struct acpi_drhd_unit *ioapic_to_drhd(un struct acpi_drhd_unit *hpet_to_drhd(unsigned int hpet_id); struct acpi_rhsa_unit *drhd_to_rhsa(const struct acpi_drhd_unit *drhd); -struct acpi_drhd_unit *find_ats_dev_drhd(struct vtd_iommu *iommu); - int ats_device(const struct pci_dev *, const struct acpi_drhd_unit *); int dev_invalidate_iotlb(struct vtd_iommu *iommu, u16 did, --- a/xen/drivers/passthrough/vtd/iommu.c +++ b/xen/drivers/passthrough/vtd/iommu.c @@ -624,8 +624,7 @@ int cf_check vtd_flush_iotlb_reg( } static int __must_check iommu_flush_iotlb_global(struct vtd_iommu *iommu, - bool flush_non_present_entry, - bool flush_dev_iotlb) + bool flush_non_present_entry) { int status; @@ -633,7 +632,7 @@ static int __must_check iommu_flush_iotl vtd_ops_preamble_quirk(iommu); status = iommu->flush.iotlb(iommu, 0, 0, 0, DMA_TLB_GLOBAL_FLUSH, - flush_non_present_entry, flush_dev_iotlb); + flush_non_present_entry, iommu->flush_dev_iotlb); /* undo platform specific errata workarounds */ vtd_ops_postamble_quirk(iommu); @@ -642,8 +641,7 @@ static int __must_check iommu_flush_iotl } static int __must_check iommu_flush_iotlb_dsi(struct vtd_iommu *iommu, u16 did, - bool flush_non_present_entry, - bool flush_dev_iotlb) + bool flush_non_present_entry) { int status; @@ -651,7 +649,7 @@ static int __must_check iommu_flush_iotl vtd_ops_preamble_quirk(iommu); status = iommu->flush.iotlb(iommu, did, 0, 0, DMA_TLB_DSI_FLUSH, - flush_non_present_entry, flush_dev_iotlb); + flush_non_present_entry, iommu->flush_dev_iotlb); /* undo platform specific errata workarounds */ vtd_ops_postamble_quirk(iommu); @@ -661,26 +659,23 @@ static int __must_check iommu_flush_iotl static int __must_check iommu_flush_iotlb_psi(struct vtd_iommu *iommu, u16 did, u64 addr, unsigned int order, - bool flush_non_present_entry, - bool flush_dev_iotlb) + bool flush_non_present_entry) { int status; /* Fallback to domain selective flush if no PSI support */ if ( !cap_pgsel_inv(iommu->cap) ) - return iommu_flush_iotlb_dsi(iommu, did, flush_non_present_entry, - flush_dev_iotlb); + return iommu_flush_iotlb_dsi(iommu, did, flush_non_present_entry); /* Fallback to domain selective flush if size is too big */ if ( order > cap_max_amask_val(iommu->cap) ) - return iommu_flush_iotlb_dsi(iommu, did, flush_non_present_entry, - flush_dev_iotlb); + return iommu_flush_iotlb_dsi(iommu, did, flush_non_present_entry); /* apply platform specific errata workarounds */ vtd_ops_preamble_quirk(iommu); status = iommu->flush.iotlb(iommu, did, addr, order, DMA_TLB_PSI_FLUSH, - flush_non_present_entry, flush_dev_iotlb); + flush_non_present_entry, iommu->flush_dev_iotlb); /* undo platform specific errata workarounds */ vtd_ops_postamble_quirk(iommu); @@ -692,7 +687,6 @@ static int __must_check iommu_flush_all( { struct acpi_drhd_unit *drhd; struct vtd_iommu *iommu; - bool flush_dev_iotlb; int rc = 0; flush_local(FLUSH_CACHE); @@ -703,8 +697,7 @@ static int __must_check iommu_flush_all( iommu = drhd->iommu; context_rc = iommu_flush_context_global(iommu, 0); - flush_dev_iotlb = !!find_ats_dev_drhd(iommu); - iotlb_rc = iommu_flush_iotlb_global(iommu, 0, flush_dev_iotlb); + iotlb_rc = iommu_flush_iotlb_global(iommu, 0); /* * The current logic for returns: @@ -734,7 +727,6 @@ static int __must_check cf_check iommu_f struct domain_iommu *hd = dom_iommu(d); struct acpi_drhd_unit *drhd; struct vtd_iommu *iommu; - bool flush_dev_iotlb; int iommu_domid; int ret = 0; @@ -762,21 +754,18 @@ static int __must_check cf_check iommu_f if ( !test_bit(iommu->index, hd->arch.vtd.iommu_bitmap) ) continue; - flush_dev_iotlb = !!find_ats_dev_drhd(iommu); iommu_domid = get_iommu_did(d->domain_id, iommu, !d->is_dying); if ( iommu_domid == -1 ) continue; if ( !page_count || (page_count & (page_count - 1)) || dfn_eq(dfn, INVALID_DFN) || !IS_ALIGNED(dfn_x(dfn), page_count) ) - rc = iommu_flush_iotlb_dsi(iommu, iommu_domid, - 0, flush_dev_iotlb); + rc = iommu_flush_iotlb_dsi(iommu, iommu_domid, 0); else rc = iommu_flush_iotlb_psi(iommu, iommu_domid, dfn_to_daddr(dfn), get_order_from_pages(page_count), - !(flush_flags & IOMMU_FLUSHF_modified), - flush_dev_iotlb); + !(flush_flags & IOMMU_FLUSHF_modified)); if ( rc > 0 ) iommu_flush_write_buffer(iommu); @@ -1488,7 +1477,6 @@ int domain_context_mapping_one( uint16_t seg = iommu->drhd->segment, prev_did = 0; struct domain *prev_dom = NULL; int rc, ret; - bool flush_dev_iotlb; if ( QUARANTINE_SKIP(domain, pgd_maddr) ) return 0; @@ -1637,8 +1625,7 @@ int domain_context_mapping_one( rc = iommu_flush_context_device(iommu, prev_did, PCI_BDF(bus, devfn), DMA_CCMD_MASK_NOBIT, !prev_dom); - flush_dev_iotlb = !!find_ats_dev_drhd(iommu); - ret = iommu_flush_iotlb_dsi(iommu, prev_did, !prev_dom, flush_dev_iotlb); + ret = iommu_flush_iotlb_dsi(iommu, prev_did, !prev_dom); /* * The current logic for returns: @@ -1882,7 +1869,6 @@ int domain_context_unmap_one( struct context_entry *context, *context_entries; u64 maddr; int iommu_domid, rc, ret; - bool flush_dev_iotlb; ASSERT(pcidevs_locked()); spin_lock(&iommu->lock); @@ -1908,8 +1894,7 @@ int domain_context_unmap_one( PCI_BDF(bus, devfn), DMA_CCMD_MASK_NOBIT, 0); - flush_dev_iotlb = !!find_ats_dev_drhd(iommu); - ret = iommu_flush_iotlb_dsi(iommu, iommu_domid, 0, flush_dev_iotlb); + ret = iommu_flush_iotlb_dsi(iommu, iommu_domid, 0); /* * The current logic for returns: --- a/xen/drivers/passthrough/vtd/iommu.h +++ b/xen/drivers/passthrough/vtd/iommu.h @@ -484,6 +484,7 @@ struct vtd_iommu { spinlock_t register_lock; /* protect iommu register handling */ u64 root_maddr; /* root entry machine address */ nodeid_t node; + bool flush_dev_iotlb; struct msi_desc msi; struct acpi_drhd_unit *drhd; --- a/xen/drivers/passthrough/vtd/x86/ats.c +++ b/xen/drivers/passthrough/vtd/x86/ats.c @@ -28,22 +28,8 @@ #include "../extern.h" #include "../../ats.h" -static LIST_HEAD(ats_dev_drhd_units); - -struct acpi_drhd_unit *find_ats_dev_drhd(struct vtd_iommu *iommu) -{ - struct acpi_drhd_unit *drhd; - list_for_each_entry ( drhd, &ats_dev_drhd_units, list ) - { - if ( drhd->iommu == iommu ) - return drhd; - } - return NULL; -} - int ats_device(const struct pci_dev *pdev, const struct acpi_drhd_unit *drhd) { - struct acpi_drhd_unit *ats_drhd; const struct acpi_satc_unit *satc; int pos; @@ -61,17 +47,10 @@ int ats_device(const struct pci_dev *pde if ( opt_ats < 0 && (!satc || !satc->atc_required) ) return 0; - ats_drhd = find_ats_dev_drhd(drhd->iommu); pos = pci_find_ext_capability(pdev->sbdf, PCI_EXT_CAP_ID_ATS); + if ( pos ) + drhd->iommu->flush_dev_iotlb = true; - if ( pos && (ats_drhd == NULL) ) - { - ats_drhd = xmalloc(struct acpi_drhd_unit); - if ( !ats_drhd ) - return -ENOMEM; - *ats_drhd = *drhd; - list_add_tail(&ats_drhd->list, &ats_dev_drhd_units); - } return pos; }
All callers only care about boolean outcome. For this there's no point in allocating a duplicate of the respective DRHD structure; a simple boolean suffices (which eventually may wantg to become a count, such that the "any ATS devices assigned state" can also clear again). With that boolean, remove respective parameters from internal helper functions right away, as those have access to the flag through another parameter. Signed-off-by: Jan Beulich <jbeulich@suse.com>