Skip to contentSkip to conversation

cleanup swiotlb initialization v4

PATCH In review Christoph Hellwig 7 participants 38 messages Mar 1, 2022 – Mar 9, 2022

Subsystems (23)
Arm (32-bit) x86 PowerPC PCI +19 more subsystems
lore
Share & tools
lore
#0 lore raw

Hi all,

this series tries to clean up the swiotlb initialization, including that of swiotlb-xen. To get there is also removes the x86 iommu table infrastructure that massively obsfucates the initialization path.

Git tree:

git://git.infradead.org/users/hch/misc.git swiotlb-init-cleanup

Gitweb:

http://git.infradead.org/users/hch/misc.git/shortlog/refs/heads/swiotlb-init-cleanup

Changes since v3:
- fix a compilation issue on some powerpc configfs
- fix and cleanup how forced bounce buffering is enabled for guest memory encryption

Changes since v2:
- make ppc_swiotlb_flags actually work again
- also force enable swiotlb for guest encrypted memory to cater to hyperv which doesn't set the host encrypted memory flag

Changes since v1:
- skip IOMMU initialization on Xen PV kernels
- various small whitespace / typo fixes

Diffstat:

#1 lore raw

PATCH 1/12 dma-direct: use is_swiotlb_active in dma_direct_map_page

Use the more specific is_swiotlb_active check instead of checking the global swiotlb_force variable.

---
 kernel/dma/direct.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/dma/direct.h +1 -1
index 4632b0f4f72eb..4dc16e08c7e1a 100644
@@ -91,7 +91,7 @@ static inline dma_addr_t dma_direct_map_page(struct device *dev,
 		return swiotlb_map(dev, phys, size, dir, attrs);
 
 	if (unlikely(!dma_capable(dev, dma_addr, size, true))) {
-		if (swiotlb_force != SWIOTLB_NO_FORCE)
+		if (is_swiotlb_active(dev))
 			return swiotlb_map(dev, phys, size, dir, attrs);
 
 		dev_WARN_ONCE(dev, 1,
Signature · 1 line
2.30.2
#2 lore raw

PATCH 2/12 swiotlb: make swiotlb_exit a no-op if SWIOTLB_FORCE is set

In reply to Christoph Hellwig

If force bouncing is enabled we can't release the buffers.

---
 kernel/dma/swiotlb.c | 3 +++
 1 file changed, 3 insertions(+)
kernel/dma/swiotlb.c +3 -0
index bfc56cb217059..64b390136f9ef 100644
@@ -378,6 +378,9 @@ void __init swiotlb_exit(void)
 	unsigned long tbl_vaddr;
 	size_t tbl_size, slots_size;
 
+	if (swiotlb_force = SWIOTLB_FORCE)
+		return;
+
 	if (!mem->nslabs)
 		return;
 
Signature · 1 line
2.30.2
#3 lore raw

PATCH 3/12 swiotlb: simplify swiotlb_max_segment

In reply to Christoph Hellwig

Remove the bogus Xen override that was usually larger than the actual size and just calculate the value on demand. Note that swiotlb_max_segment still doesn't make sense as an interface and should eventually be removed.

---
 drivers/xen/swiotlb-xen.c |  2 --
 include/linux/swiotlb.h   |  1 -
 kernel/dma/swiotlb.c      | 20 +++-----------------
 3 files changed, 3 insertions(+), 20 deletions(-)
drivers/xen/swiotlb-xen.c +0 -2
index 47aebd98f52f5..485cd06ed39e7 100644
@@ -202,7 +202,6 @@ int xen_swiotlb_init(void)
 	rc = swiotlb_late_init_with_tbl(start, nslabs);
 	if (rc)
 		return rc;
-	swiotlb_set_max_segment(PAGE_SIZE);
 	return 0;
 error:
 	if (nslabs > 1024 && repeat--) {
@@ -254,7 +253,6 @@ void __init xen_swiotlb_init_early(void)
 
 	if (swiotlb_init_with_tbl(start, nslabs, true))
 		panic("Cannot allocate SWIOTLB buffer");
-	swiotlb_set_max_segment(PAGE_SIZE);
 }
 #endif /* CONFIG_X86 */
 
include/linux/swiotlb.h +0 -1
index f6c3638255d54..9fb3a568f0c51 100644
@@ -164,7 +164,6 @@ static inline void swiotlb_adjust_size(unsigned long size)
 #endif /* CONFIG_SWIOTLB */
 
 extern void swiotlb_print_info(void);
-extern void swiotlb_set_max_segment(unsigned int);
 
 #ifdef CONFIG_DMA_RESTRICTED_POOL
 struct page *swiotlb_alloc(struct device *dev, size_t size);
kernel/dma/swiotlb.c +3 -17
index 64b390136f9ef..e31c235b5fd9e 100644
@@ -75,12 +75,6 @@ struct io_tlb_mem io_tlb_default_mem;
 
 phys_addr_t swiotlb_unencrypted_base;
 
-/*
- * Max segment that we can provide which (if pages are contingous) will
- * not be bounced (unless SWIOTLB_FORCE is set).
- */
-static unsigned int max_segment;
-
 static unsigned long default_nslabs = IO_TLB_DEFAULT_SIZE >> IO_TLB_SHIFT;
 
 static int __init
@@ -104,18 +98,12 @@ early_param("swiotlb", setup_io_tlb_npages);
 
 unsigned int swiotlb_max_segment(void)
 {
-	return io_tlb_default_mem.nslabs ? max_segment : 0;
+	if (!io_tlb_default_mem.nslabs)
+		return 0;
+	return rounddown(io_tlb_default_mem.nslabs << IO_TLB_SHIFT, PAGE_SIZE);
 }
 EXPORT_SYMBOL_GPL(swiotlb_max_segment);
 
-void swiotlb_set_max_segment(unsigned int val)
-{
-	if (swiotlb_force = SWIOTLB_FORCE)
-		max_segment = 1;
-	else
-		max_segment = rounddown(val, PAGE_SIZE);
-}
-
 unsigned long swiotlb_size_or_default(void)
 {
 	return default_nslabs << IO_TLB_SHIFT;
@@ -267,7 +255,6 @@ int __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose)
 
 	if (verbose)
 		swiotlb_print_info();
-	swiotlb_set_max_segment(mem->nslabs << IO_TLB_SHIFT);
 	return 0;
 }
 
@@ -368,7 +355,6 @@ swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs)
 	swiotlb_init_io_tlb_mem(mem, virt_to_phys(tlb), nslabs, true);
 
 	swiotlb_print_info();
-	swiotlb_set_max_segment(mem->nslabs << IO_TLB_SHIFT);
 	return 0;
 }
 
Signature · 1 line
2.30.2
#4 lore raw

PATCH 4/12 swiotlb: rename swiotlb_late_init_with_default_size

In reply to Christoph Hellwig

swiotlb_late_init_with_default_size is an overly verbose name that doesn't even catch what the function is doing, given that the size is not just a default but the actual requested size.

Rename it to swiotlb_init_late.

---
 arch/x86/pci/sta2x11-fixup.c | 2 +-
 include/linux/swiotlb.h      | 2 +-
 kernel/dma/swiotlb.c         | 6 ++----
 3 files changed, 4 insertions(+), 6 deletions(-)
arch/x86/pci/sta2x11-fixup.c +1 -1
index 101081ad64b6d..e0c039a75b2db 100644
@@ -57,7 +57,7 @@ static void sta2x11_new_instance(struct pci_dev *pdev)
 		int size = STA2X11_SWIOTLB_SIZE;
 		/* First instance: register your own swiotlb area */
 		dev_info(&pdev->dev, "Using SWIOTLB (size %i)\n", size);
-		if (swiotlb_late_init_with_default_size(size))
+		if (swiotlb_init_late(size))
 			dev_emerg(&pdev->dev, "init swiotlb failed\n");
 	}
 	list_add(&instance->list, &sta2x11_instance_list);
include/linux/swiotlb.h +1 -1
index 9fb3a568f0c51..b48b26bfa0edb 100644
@@ -40,7 +40,7 @@ extern void swiotlb_init(int verbose);
 int swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose);
 unsigned long swiotlb_size_or_default(void);
 extern int swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs);
-extern int swiotlb_late_init_with_default_size(size_t default_size);
+int swiotlb_init_late(size_t size);
 extern void __init swiotlb_update_mem_attributes(void);
 
 phys_addr_t swiotlb_tbl_map_single(struct device *hwdev, phys_addr_t phys,
kernel/dma/swiotlb.c +2 -3
index e31c235b5fd9e..0b1992c355f36 100644
@@ -290,11 +290,9 @@ swiotlb_init(int verbose)
  * initialize the swiotlb later using the slab allocator if needed.
  * This should be just like above, but with some error catching.
  */
-int
-swiotlb_late_init_with_default_size(size_t default_size)
+int swiotlb_init_late(size_t size)
 {
-	unsigned long nslabs -		ALIGN(default_size >> IO_TLB_SHIFT, IO_TLB_SEGSIZE);
+	unsigned long nslabs = ALIGN(size >> IO_TLB_SHIFT, IO_TLB_SEGSIZE);
 	unsigned long bytes;
 	unsigned char *vstart = NULL;
 	unsigned int order;
Signature · 1 line
2.30.2
#5 lore raw

PATCH 5/12 swiotlb: pass a gfp_mask argument to swiotlb_init_late

In reply to Christoph Hellwig

Let the caller chose a zone to allocate from. This will be used later on by the xen-swiotlb initialization on arm.

---
 arch/x86/pci/sta2x11-fixup.c | 2 +-
 include/linux/swiotlb.h      | 2 +-
 kernel/dma/swiotlb.c         | 4 ++--
 3 files changed, 4 insertions(+), 4 deletions(-)
arch/x86/pci/sta2x11-fixup.c +1 -1
index e0c039a75b2db..c7e6faf59a861 100644
@@ -57,7 +57,7 @@ static void sta2x11_new_instance(struct pci_dev *pdev)
 		int size = STA2X11_SWIOTLB_SIZE;
 		/* First instance: register your own swiotlb area */
 		dev_info(&pdev->dev, "Using SWIOTLB (size %i)\n", size);
-		if (swiotlb_init_late(size))
+		if (swiotlb_init_late(size, GFP_DMA))
 			dev_emerg(&pdev->dev, "init swiotlb failed\n");
 	}
 	list_add(&instance->list, &sta2x11_instance_list);
include/linux/swiotlb.h +1 -1
index b48b26bfa0edb..1befd6b2ccf5e 100644
@@ -40,7 +40,7 @@ extern void swiotlb_init(int verbose);
 int swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose);
 unsigned long swiotlb_size_or_default(void);
 extern int swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs);
-int swiotlb_init_late(size_t size);
+int swiotlb_init_late(size_t size, gfp_t gfp_mask);
 extern void __init swiotlb_update_mem_attributes(void);
 
 phys_addr_t swiotlb_tbl_map_single(struct device *hwdev, phys_addr_t phys,
kernel/dma/swiotlb.c +2 -2
index 0b1992c355f36..9a38ea3a46e9f 100644
@@ -290,7 +290,7 @@ swiotlb_init(int verbose)
  * initialize the swiotlb later using the slab allocator if needed.
  * This should be just like above, but with some error catching.
  */
-int swiotlb_init_late(size_t size)
+int swiotlb_init_late(size_t size, gfp_t gfp_mask)
 {
 	unsigned long nslabs = ALIGN(size >> IO_TLB_SHIFT, IO_TLB_SEGSIZE);
 	unsigned long bytes;
@@ -309,7 +309,7 @@ int swiotlb_init_late(size_t size)
 	bytes = nslabs << IO_TLB_SHIFT;
 
 	while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) {
-		vstart = (void *)__get_free_pages(GFP_DMA | __GFP_NOWARN,
+		vstart = (void *)__get_free_pages(gfp_mask | __GFP_NOWARN,
 						  order);
 		if (vstart)
 			break;
Signature · 1 line
2.30.2
#6 lore raw

PATCH 6/12 MIPS/octeon: use swiotlb_init instead of open coding it

In reply to Christoph Hellwig

Use the generic swiotlb initialization helper instead of open coding it.

---
 arch/mips/cavium-octeon/dma-octeon.c | 15 ++-------------
 arch/mips/pci/pci-octeon.c           |  2 +-
 2 files changed, 3 insertions(+), 14 deletions(-)
arch/mips/cavium-octeon/dma-octeon.c +2 -13
index df70308db0e69..fb7547e217263 100644
@@ -186,15 +186,12 @@ phys_addr_t dma_to_phys(struct device *dev, dma_addr_t daddr)
 	return daddr;
 }
 
-char *octeon_swiotlb;
-
 void __init plat_swiotlb_setup(void)
 {
 	phys_addr_t start, end;
 	phys_addr_t max_addr;
 	phys_addr_t addr_size;
 	size_t swiotlbsize;
-	unsigned long swiotlb_nslabs;
 	u64 i;
 
 	max_addr = 0;
@@ -236,15 +233,7 @@ void __init plat_swiotlb_setup(void)
 	if (OCTEON_IS_OCTEON2() && max_addr >= 0x100000000ul)
 		swiotlbsize = 64 * (1<<20);
 #endif
-	swiotlb_nslabs = swiotlbsize >> IO_TLB_SHIFT;
-	swiotlb_nslabs = ALIGN(swiotlb_nslabs, IO_TLB_SEGSIZE);
-	swiotlbsize = swiotlb_nslabs << IO_TLB_SHIFT;
-
-	octeon_swiotlb = memblock_alloc_low(swiotlbsize, PAGE_SIZE);
-	if (!octeon_swiotlb)
-		panic("%s: Failed to allocate %zu bytes align=%lx\n",
-		      __func__, swiotlbsize, PAGE_SIZE);
 
-	if (swiotlb_init_with_tbl(octeon_swiotlb, swiotlb_nslabs, 1) = -ENOMEM)
-		panic("Cannot allocate SWIOTLB buffer");
+	swiotlb_adjust_size(swiotlbsize);
+	swiotlb_init(1);
 }
arch/mips/pci/pci-octeon.c +1 -0
index fc29b85cfa926..e457a18cbdc59 100644
@@ -664,7 +664,7 @@ static int __init octeon_pci_setup(void)
 
 		/* BAR1 movable regions contiguous to cover the swiotlb */
 		octeon_bar1_pci_phys -			virt_to_phys(octeon_swiotlb) & ~((1ull << 22) - 1);
+			io_tlb_default_mem.start & ~((1ull << 22) - 1);
 
 		for (index = 0; index < 32; index++) {
 			union cvmx_pci_bar1_indexx bar1_index;
Signature · 1 line
2.30.2
maintainer Acked-by on v1 6/12
#7 lore raw

Re: [PATCH 06/12] MIPS/octeon: use swiotlb_init instead of open coding it

Quoted 7 lines · Christoph Hellwig

Use the generic swiotlb initialization helper instead of open coding it.

---
 arch/mips/cavium-octeon/dma-octeon.c | 15 ++-------------
 arch/mips/pci/pci-octeon.c           |  2 +-
 2 files changed, 3 insertions(+), 14 deletions(-)
Signature · 2 lines
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]
#8 lore raw

PATCH 7/12 x86: remove the IOMMU table infrastructure

In reply to Christoph Hellwig

The IOMMU table tries to separate the different IOMMUs into different backends, but actually requires various cross calls.

Rewrite the code to do the generic swiotlb/swiotlb-xen setup directly in pci-dma.c and then just call into the IOMMU drivers.

---
 arch/ia64/include/asm/iommu_table.h    |   7 --
 arch/x86/include/asm/dma-mapping.h     |   1 -
 arch/x86/include/asm/gart.h            |   5 +-
 arch/x86/include/asm/iommu.h           |   6 ++
 arch/x86/include/asm/iommu_table.h     | 102 -----------------------
 arch/x86/include/asm/swiotlb.h         |  30 -------
 arch/x86/include/asm/xen/swiotlb-xen.h |   2 -
 arch/x86/kernel/Makefile               |   2 -
 arch/x86/kernel/amd_gart_64.c          |   5 +-
 arch/x86/kernel/aperture_64.c          |  14 ++--
 arch/x86/kernel/pci-dma.c              | 107 ++++++++++++++++++++-----
 arch/x86/kernel/pci-iommu_table.c      |  77 ------------------
 arch/x86/kernel/pci-swiotlb.c          |  77 ------------------
 arch/x86/kernel/tboot.c                |   1 -
 arch/x86/kernel/vmlinux.lds.S          |  12 ---
 arch/x86/xen/Makefile                  |   2 -
 arch/x86/xen/pci-swiotlb-xen.c         |  96 ----------------------
 drivers/iommu/amd/init.c               |   6 --
 drivers/iommu/amd/iommu.c              |   5 +-
 drivers/iommu/intel/dmar.c             |   6 +-
 include/linux/dmar.h                   |   6 +-
 21 files changed, 110 insertions(+), 459 deletions(-)
 delete mode 100644 arch/ia64/include/asm/iommu_table.h
 delete mode 100644 arch/x86/include/asm/iommu_table.h
 delete mode 100644 arch/x86/include/asm/swiotlb.h
 delete mode 100644 arch/x86/kernel/pci-iommu_table.c
 delete mode 100644 arch/x86/kernel/pci-swiotlb.c
 delete mode 100644 arch/x86/xen/pci-swiotlb-xen.c
arch/ia64/include/asm/iommu_table.h deleted +0 -7
deleted file mode 100644
index cc96116ac276a..0000000000000
@@ -1,7 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0 */
-#ifndef _ASM_IA64_IOMMU_TABLE_H
-#define _ASM_IA64_IOMMU_TABLE_H
-
-#define IOMMU_INIT_POST(_detect)
-
-#endif /* _ASM_IA64_IOMMU_TABLE_H */
arch/x86/include/asm/dma-mapping.h +0 -1
index bb1654fe0ce74..256fd8115223d 100644
@@ -9,7 +9,6 @@
 
 #include <linux/scatterlist.h>
 #include <asm/io.h>
-#include <asm/swiotlb.h>
 
 extern int iommu_merge;
 extern int panic_on_overflow;
arch/x86/include/asm/gart.h +2 -3
index 3185565743459..5af8088a10df6 100644
@@ -38,7 +38,7 @@ extern int gart_iommu_aperture_disabled;
 extern void early_gart_iommu_check(void);
 extern int gart_iommu_init(void);
 extern void __init gart_parse_options(char *);
-extern int gart_iommu_hole_init(void);
+void gart_iommu_hole_init(void);
 
 #else
 #define gart_iommu_aperture            0
@@ -51,9 +51,8 @@ static inline void early_gart_iommu_check(void)
 static inline void gart_parse_options(char *options)
 {
 }
-static inline int gart_iommu_hole_init(void)
+static inline void gart_iommu_hole_init(void)
 {
-	return -ENODEV;
 }
 #endif
 
arch/x86/include/asm/iommu.h +6 -0
index bf1ed2ddc74bd..dba89ed40d38d 100644
@@ -9,6 +9,12 @@
 extern int force_iommu, no_iommu;
 extern int iommu_detected;
 
+#ifdef CONFIG_SWIOTLB
+extern bool x86_swiotlb_enable;
+#else
+#define x86_swiotlb_enable false
+#endif
+
 /* 10 seconds */
 #define DMAR_OPERATION_TIMEOUT ((cycles_t) tsc_khz*10*1000)
 
arch/x86/include/asm/iommu_table.h deleted +0 -102

102 lines not on this page · raw message

arch/x86/include/asm/swiotlb.h deleted +0 -30
deleted file mode 100644
index ff6c92eff035a..0000000000000
@@ -1,30 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0 */
-#ifndef _ASM_X86_SWIOTLB_H
-#define _ASM_X86_SWIOTLB_H
-
-#include <linux/swiotlb.h>
-
-#ifdef CONFIG_SWIOTLB
-extern int swiotlb;
-extern int __init pci_swiotlb_detect_override(void);
-extern int __init pci_swiotlb_detect_4gb(void);
-extern void __init pci_swiotlb_init(void);
-extern void __init pci_swiotlb_late_init(void);
-#else
-#define swiotlb 0
-static inline int pci_swiotlb_detect_override(void)
-{
-	return 0;
-}
-static inline int pci_swiotlb_detect_4gb(void)
-{
-	return 0;
-}
-static inline void pci_swiotlb_init(void)
-{
-}
-static inline void pci_swiotlb_late_init(void)
-{
-}
-#endif
-#endif /* _ASM_X86_SWIOTLB_H */
arch/x86/include/asm/xen/swiotlb-xen.h +0 -2
index 66b4ddde77430..e5a90b42e4dde 100644
@@ -3,10 +3,8 @@
 #define _ASM_X86_SWIOTLB_XEN_H
 
 #ifdef CONFIG_SWIOTLB_XEN
-extern int __init pci_xen_swiotlb_detect(void);
 extern int pci_xen_swiotlb_init_late(void);
 #else
-#define pci_xen_swiotlb_detect NULL
 static inline int pci_xen_swiotlb_init_late(void) { return -ENXIO; }
 #endif
 
arch/x86/kernel/Makefile +0 -2
index 6aef9ee28a394..2851d4f0aa0d2 100644
@@ -71,7 +71,6 @@ obj-y			+= bootflag.o e820.o
 obj-y			+= pci-dma.o quirks.o topology.o kdebugfs.o
 obj-y			+= alternative.o i8253.o hw_breakpoint.o
 obj-y			+= tsc.o tsc_msr.o io_delay.o rtc.o
-obj-y			+= pci-iommu_table.o
 obj-y			+= resource.o
 obj-y			+= irqflags.o
 obj-y			+= static_call.o
@@ -136,7 +135,6 @@ obj-$(CONFIG_PCSPKR_PLATFORM)	+= pcspeaker.o
 
 obj-$(CONFIG_X86_CHECK_BIOS_CORRUPTION) += check.o
 
-obj-$(CONFIG_SWIOTLB)			+= pci-swiotlb.o
 obj-$(CONFIG_OF)			+= devicetree.o
 obj-$(CONFIG_UPROBES)			+= uprobes.o
 
arch/x86/kernel/amd_gart_64.c +1 -4
index ed837383de5c8..194d54eed5376 100644
@@ -38,11 +38,9 @@
 #include <asm/iommu.h>
 #include <asm/gart.h>
 #include <asm/set_memory.h>
-#include <asm/swiotlb.h>
 #include <asm/dma.h>
 #include <asm/amd_nb.h>
 #include <asm/x86_init.h>
-#include <asm/iommu_table.h>
 
 static unsigned long iommu_bus_base;	/* GART remapping area (physical) */
 static unsigned long iommu_size;	/* size of remapping area bytes */
@@ -808,7 +806,7 @@ int __init gart_iommu_init(void)
 	flush_gart();
 	dma_ops = &gart_dma_ops;
 	x86_platform.iommu_shutdown = gart_iommu_shutdown;
-	swiotlb = 0;
+	x86_swiotlb_enable = false;
 
 	return 0;
 }
@@ -842,4 +840,3 @@ void __init gart_parse_options(char *p)
 		}
 	}
 }
-IOMMU_INIT_POST(gart_iommu_hole_init);
arch/x86/kernel/aperture_64.c +5 -9
index af3ba08b684b5..7a5630d904b23 100644
@@ -392,7 +392,7 @@ void __init early_gart_iommu_check(void)
 
 static int __initdata printed_gart_size_msg;
 
-int __init gart_iommu_hole_init(void)
+void __init gart_iommu_hole_init(void)
 {
 	u32 agp_aper_base = 0, agp_aper_order = 0;
 	u32 aper_size, aper_alloc = 0, aper_order = 0, last_aper_order = 0;
@@ -401,11 +401,11 @@ int __init gart_iommu_hole_init(void)
 	int i, node;
 
 	if (!amd_gart_present())
-		return -ENODEV;
+		return;
 
 	if (gart_iommu_aperture_disabled || !fix_aperture ||
 	    !early_pci_allowed())
-		return -ENODEV;
+		return;
 
 	pr_info("Checking aperture...\n");
 
@@ -491,10 +491,8 @@ int __init gart_iommu_hole_init(void)
 			 * and fixed up the northbridge
 			 */
 			exclude_from_core(last_aper_base, last_aper_order);
-
-			return 1;
 		}
-		return 0;
+		return;
 	}
 
 	if (!fallback_aper_force) {
@@ -527,7 +525,7 @@ int __init gart_iommu_hole_init(void)
 			panic("Not enough memory for aperture");
 		}
 	} else {
-		return 0;
+		return;
 	}
 
 	/*
@@ -561,6 +559,4 @@ int __init gart_iommu_hole_init(void)
 	}
 
 	set_up_gart_resume(aper_order, aper_alloc);
-
-	return 1;
 }
arch/x86/kernel/pci-dma.c +86 -21

144 lines not on this page · raw message

arch/x86/kernel/pci-iommu_table.c deleted +0 -77

77 lines not on this page · raw message

arch/x86/kernel/pci-swiotlb.c deleted +0 -77

77 lines not on this page · raw message

arch/x86/kernel/tboot.c +0 -1
index f9af561c3cd4f..0c1154a1c4032 100644
@@ -24,7 +24,6 @@
 #include <asm/processor.h>
 #include <asm/bootparam.h>
 #include <asm/pgalloc.h>
-#include <asm/swiotlb.h>
 #include <asm/fixmap.h>
 #include <asm/proto.h>
 #include <asm/setup.h>
arch/x86/kernel/vmlinux.lds.S +0 -12
index 27f830345b6f0..bbe910c15b293 100644
@@ -306,18 +306,6 @@ SECTIONS
 		*(.altinstr_replacement)
 	}
 
-	/*
-	 * struct iommu_table_entry entries are injected in this section.
-	 * It is an array of IOMMUs which during run time gets sorted depending
-	 * on its dependency order. After rootfs_initcall is complete
-	 * this section can be safely removed.
-	 */
-	.iommu_table : AT(ADDR(.iommu_table) - LOAD_OFFSET) {
-		__iommu_table = .;
-		*(.iommu_table)
-		__iommu_table_end = .;
-	}
-
 	. = ALIGN(8);
 	.apicdrivers : AT(ADDR(.apicdrivers) - LOAD_OFFSET) {
 		__apicdrivers = .;
arch/x86/xen/Makefile +0 -2
index 4953260e281c3..3c5b52fbe4a7f 100644
@@ -47,6 +47,4 @@ obj-$(CONFIG_XEN_DEBUG_FS)	+= debugfs.o
 
 obj-$(CONFIG_XEN_PV_DOM0)	+= vga.o
 
-obj-$(CONFIG_SWIOTLB_XEN)	+= pci-swiotlb-xen.o
-
 obj-$(CONFIG_XEN_EFI)		+= efi.o
arch/x86/xen/pci-swiotlb-xen.c deleted +0 -96

96 lines not on this page · raw message

drivers/iommu/amd/init.c +0 -6
index b10fb52ea4428..721300cf90207 100644
@@ -27,7 +27,6 @@
 #include <asm/apic.h>
 #include <asm/gart.h>
 #include <asm/x86_init.h>
-#include <asm/iommu_table.h>
 #include <asm/io_apic.h>
 #include <asm/irq_remapping.h>
 #include <asm/set_memory.h>
@@ -3237,11 +3236,6 @@ __setup("ivrs_ioapic",		parse_ivrs_ioapic);
 __setup("ivrs_hpet",		parse_ivrs_hpet);
 __setup("ivrs_acpihid",		parse_ivrs_acpihid);
 
-IOMMU_INIT_FINISH(amd_iommu_detect,
-		  gart_iommu_hole_init,
-		  NULL,
-		  NULL);
-
 bool amd_iommu_v2_supported(void)
 {
 	return amd_iommu_v2_present;
drivers/iommu/amd/iommu.c +4 -1
index 461f1844ed1fb..541a7b8315a12 100644
@@ -1834,7 +1834,10 @@ void amd_iommu_domain_update(struct protection_domain *domain)
 
 static void __init amd_iommu_init_dma_ops(void)
 {
-	swiotlb = (iommu_default_passthrough() || sme_me_mask) ? 1 : 0;
+	if (iommu_default_passthrough() || sme_me_mask)
+		x86_swiotlb_enable = true;
+	else
+		x86_swiotlb_enable = false;
 }
 
 int __init amd_iommu_init_api(void)
drivers/iommu/intel/dmar.c +1 -5
index 915bff76fe965..29bee4b210c5b 100644
@@ -30,7 +30,6 @@
 #include <linux/numa.h>
 #include <linux/limits.h>
 #include <asm/irq_remapping.h>
-#include <asm/iommu_table.h>
 #include <trace/events/intel_iommu.h>
 
 #include "../irq_remapping.h"
@@ -913,7 +912,7 @@ dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg)
 	return 0;
 }
 
-int __init detect_intel_iommu(void)
+void __init detect_intel_iommu(void)
 {
 	int ret;
 	struct dmar_res_callback validate_drhd_cb = {
@@ -946,8 +945,6 @@ int __init detect_intel_iommu(void)
 		dmar_tbl = NULL;
 	}
 	up_write(&dmar_global_lock);
-
-	return ret ? ret : 1;
 }
 
 static void unmap_iommu(struct intel_iommu *iommu)
@@ -2165,7 +2162,6 @@ static int __init dmar_free_unused_resources(void)
 }
 
 late_initcall(dmar_free_unused_resources);
-IOMMU_INIT_POST(detect_intel_iommu);
 
 /*
  * DMAR Hotplug Support
include/linux/dmar.h +5 -1
index 45e903d847335..cbd714a198a0a 100644
@@ -121,7 +121,7 @@ extern int dmar_remove_dev_scope(struct dmar_pci_notify_info *info,
 				 u16 segment, struct dmar_dev_scope *devices,
 				 int count);
 /* Intel IOMMU detection */
-extern int detect_intel_iommu(void);
+void detect_intel_iommu(void);
 extern int enable_drhd_fault_handling(void);
 extern int dmar_device_add(acpi_handle handle);
 extern int dmar_device_remove(acpi_handle handle);
@@ -197,6 +197,10 @@ static inline bool dmar_platform_optin(void)
 	return false;
 }
 
+static inline void detect_intel_iommu(void)
+{
+}
+
 #endif /* CONFIG_DMAR_TABLE */
 
 struct irte {
Signature · 1 line
2.30.2
#9 lore raw

PATCH 8/12 x86: centralize setting SWIOTLB_FORCE when guest memory encryption is enabled

In reply to Christoph Hellwig

Move enabling SWIOTLB_FORCE for guest memory encryption into common code.

---
 arch/x86/kernel/cpu/mshyperv.c | 8 --------
 arch/x86/kernel/pci-dma.c      | 7 +++++++
 arch/x86/mm/mem_encrypt_amd.c  | 3 ---
 3 files changed, 7 insertions(+), 11 deletions(-)
arch/x86/kernel/cpu/mshyperv.c +0 -8
index 5a99f993e6392..568274917f1cd 100644
@@ -336,14 +336,6 @@ static void __init ms_hyperv_init_platform(void)
 			swiotlb_unencrypted_base = ms_hyperv.shared_gpa_boundary;
 #endif
 		}
-
-#ifdef CONFIG_SWIOTLB
-		/*
-		 * Enable swiotlb force mode in Isolation VM to
-		 * use swiotlb bounce buffer for dma transaction.
-		 */
-		swiotlb_force = SWIOTLB_FORCE;
-#endif
 	}
 
 	if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
arch/x86/kernel/pci-dma.c +7 -0
index 2ac0ef9c2fb76..7ab7002758396 100644
@@ -53,6 +53,13 @@ static void __init pci_swiotlb_detect(void)
 	if (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
 		x86_swiotlb_enable = true;
 
+	/*
+	 * Guest with guest memory encryption must always do I/O through a
+	 * bounce buffer as the hypervisor can't access arbitrary VM memory.
+	 */
+	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
+		swiotlb_force = SWIOTLB_FORCE;
+
 	if (swiotlb_force = SWIOTLB_FORCE)
 		x86_swiotlb_enable = true;
 }
arch/x86/mm/mem_encrypt_amd.c +0 -3
index 2b2d018ea3450..a72942d569cf9 100644
@@ -191,9 +191,6 @@ void __init sme_early_init(void)
 	/* Update the protection map with memory encryption mask */
 	for (i = 0; i < ARRAY_SIZE(protection_map); i++)
 		protection_map[i] = pgprot_encrypted(protection_map[i]);
-
-	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
-		swiotlb_force = SWIOTLB_FORCE;
 }
 
 void __init sev_setup_arch(void)
Signature · 1 line
2.30.2
#10 lore raw

Re: [PATCH 08/12] x86: centralize setting SWIOTLB_FORCE when guest memory encryption is enabled

T24gMDEvMDMvMjAyMiAxMDo1MywgQ2hyaXN0b3BoIEhlbGx3aWcgd3JvdGU6DQo+IGRpZmYgLS1n aXQgYS9hcmNoL3g4Ni9rZXJuZWwvcGNpLWRtYS5jIGIvYXJjaC94ODYva2VybmVsL3BjaS1kbWEu Yw0KPiBpbmRleCAyYWMwZWY5YzJmYjc2Li43YWI3MDAyNzU4Mzk2IDEwMDY0NA0KPiAtLS0gYS9h cmNoL3g4Ni9rZXJuZWwvcGNpLWRtYS5jDQo+ICsrKyBiL2FyY2gveDg2L2tlcm5lbC9wY2ktZG1h LmMNCj4gQEAgLTUzLDYgKzUzLDEzIEBAIHN0YXRpYyB2b2lkIF9faW5pdCBwY2lfc3dpb3RsYl9k ZXRlY3Qodm9pZCkNCj4gIAlpZiAoY2NfcGxhdGZvcm1faGFzKENDX0FUVFJfSE9TVF9NRU1fRU5D UllQVCkpDQo+ICAJCXg4Nl9zd2lvdGxiX2VuYWJsZSA9IHRydWU7DQo+ICANCj4gKwkvKg0KPiAr CSAqIEd1ZXN0IHdpdGggZ3Vlc3QgbWVtb3J5IGVuY3J5cHRpb24gbXVzdCBhbHdheXMgZG8gSS9P IHRocm91Z2ggYQ0KPiArCSAqIGJvdW5jZSBidWZmZXIgYXMgdGhlIGh5cGVydmlzb3IgY2FuJ3Qg YWNjZXNzIGFyYml0cmFyeSBWTSBtZW1vcnkuDQoNClRoaXMgaXNuJ3QgcmVhbGx5ICJtdXN0Ii7C oCBUaGUgZ3Vlc3QgaXMgcGVyZmVjdGx5IGNhcGFibGUgb2Ygc2hhcmluZw0KbWVtb3J5IHdpdGgg dGhlIGh5cGVydmlzb3IuDQoNCkl0J3MganVzdCB0aGF0IGZvciBub3csIGJvdW5jZSBidWZmZXJp bmcgaXMgYWxsZWdlZGx5IGZhc3RlciwgYW5kIHRoZQ0Kc2ltcGxlIHdheSBvZiBnZXR0aW5nIGl0 IHdvcmtpbmcuDQoNCn5BbmRyZXcNCg=

#11 lore raw

Re: [PATCH 08/12] x86: centralize setting SWIOTLB_FORCE when guest memory encryption is enabled

Quoted 5 lines · Andrew Cooper

This isn't really "must".  The guest is perfectly capable of sharing memory with the hypervisor.

It's just that for now, bounce buffering is allegedly faster, and the simple way of getting it working.

Yeah, I guess you щould just share/unshare on demand. But given that this isn't implemented it is a must in the current kernel. But if you want a different wording suggest one and I'll put it in.

#12 lore raw

PATCH 9/12 swiotlb: make the swiotlb_init interface more useful

In reply to Christoph Hellwig

Pass a bool to pass if swiotlb needs to be enabled based on the addressing needs and replace the verbose argument with a set of flags, including one to force enable bounce buffering.

Note that this patch removes the possibility to force xen-swiotlb use using swiotlb=force on the command line on x86 (arm and arm64 never supported that), but this interface will be restored shortly.

---
 arch/arm/mm/init.c                     |  6 +----
 arch/arm64/mm/init.c                   |  6 +----
 arch/ia64/mm/init.c                    |  4 +--
 arch/mips/cavium-octeon/dma-octeon.c   |  2 +-
 arch/mips/loongson64/dma.c             |  2 +-
 arch/mips/sibyte/common/dma.c          |  2 +-
 arch/powerpc/mm/mem.c                  |  3 ++-
 arch/powerpc/platforms/pseries/setup.c |  3 ---
 arch/riscv/mm/init.c                   |  8 +-----
 arch/s390/mm/init.c                    |  3 +--
 arch/x86/kernel/pci-dma.c              | 15 ++++++-----
 drivers/xen/swiotlb-xen.c              |  4 +--
 include/linux/swiotlb.h                | 15 ++++++-----
 include/trace/events/swiotlb.h         | 29 ++++++++-------------
 kernel/dma/swiotlb.c                   | 35 ++++++++++++++------------
 15 files changed, 55 insertions(+), 82 deletions(-)
arch/arm/mm/init.c +1 -5
index 6d0cb0f7bc54b..73f30d278b565 100644
@@ -312,11 +312,7 @@ static void __init free_highpages(void)
 void __init mem_init(void)
 {
 #ifdef CONFIG_ARM_LPAE
-	if (swiotlb_force = SWIOTLB_FORCE ||
-	    max_pfn > arm_dma_pfn_limit)
-		swiotlb_init(1);
-	else
-		swiotlb_force = SWIOTLB_NO_FORCE;
+	swiotlb_init(max_pfn > arm_dma_pfn_limit, SWIOTLB_VERBOSE);
 #endif
 
 	set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
arch/arm64/mm/init.c +1 -5
index db63cc885771a..52102adda3d28 100644
@@ -373,11 +373,7 @@ void __init bootmem_init(void)
  */
 void __init mem_init(void)
 {
-	if (swiotlb_force = SWIOTLB_FORCE ||
-	    max_pfn > PFN_DOWN(arm64_dma_phys_limit))
-		swiotlb_init(1);
-	else if (!xen_swiotlb_detect())
-		swiotlb_force = SWIOTLB_NO_FORCE;
+	swiotlb_init(max_pfn > PFN_DOWN(arm64_dma_phys_limit), SWIOTLB_VERBOSE);
 
 	/* this will put all unused low memory onto the freelists */
 	memblock_free_all();
arch/ia64/mm/init.c +1 -3
index 5d165607bf354..3c3e15b22608f 100644
@@ -437,9 +437,7 @@ mem_init (void)
 		if (iommu_detected)
 			break;
 #endif
-#ifdef CONFIG_SWIOTLB
-		swiotlb_init(1);
-#endif
+		swiotlb_init(true, SWIOTLB_VERBOSE);
 	} while (0);
 
 #ifdef CONFIG_FLATMEM
arch/mips/cavium-octeon/dma-octeon.c +1 -1
index fb7547e217263..9fbba6a8fa4c5 100644
@@ -235,5 +235,5 @@ void __init plat_swiotlb_setup(void)
 #endif
 
 	swiotlb_adjust_size(swiotlbsize);
-	swiotlb_init(1);
+	swiotlb_init(true, SWIOTLB_VERBOSE);
 }
arch/mips/loongson64/dma.c +1 -1
index 364f2f27c8723..8220a1bc0db64 100644
@@ -24,5 +24,5 @@ phys_addr_t dma_to_phys(struct device *dev, dma_addr_t daddr)
 
 void __init plat_swiotlb_setup(void)
 {
-	swiotlb_init(1);
+	swiotlb_init(true, SWIOTLB_VERBOSE);
 }
arch/mips/sibyte/common/dma.c +1 -1
index eb47a94f3583e..c5c2c782aff68 100644
@@ -10,5 +10,5 @@
 
 void __init plat_swiotlb_setup(void)
 {
-	swiotlb_init(1);
+	swiotlb_init(true, SWIOTLB_VERBOSE);
 }
arch/powerpc/mm/mem.c +2 -1
index 8e301cd8925b2..e1519e2edc656 100644
@@ -17,6 +17,7 @@
 #include <linux/suspend.h>
 #include <linux/dma-direct.h>
 
+#include <asm/swiotlb.h>
 #include <asm/machdep.h>
 #include <asm/rtas.h>
 #include <asm/kasan.h>
@@ -251,7 +252,7 @@ void __init mem_init(void)
 	if (is_secure_guest())
 		svm_swiotlb_init();
 	else
-		swiotlb_init(0);
+		swiotlb_init(ppc_swiotlb_enable, 0);
 #endif
 
 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
arch/powerpc/platforms/pseries/setup.c +0 -3
index 83a04d967a59f..45d637ab58261 100644
@@ -838,9 +838,6 @@ static void __init pSeries_setup_arch(void)
 	}
 
 	ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
-
-	if (swiotlb_force = SWIOTLB_FORCE)
-		ppc_swiotlb_enable = 1;
 }
 
 static void pseries_panic(char *str)
arch/riscv/mm/init.c +1 -7
index c27294128e182..6cdbb62672fe5 100644
@@ -118,13 +118,7 @@ void __init mem_init(void)
 	BUG_ON(!mem_map);
 #endif /* CONFIG_FLATMEM */
 
-#ifdef CONFIG_SWIOTLB
-	if (swiotlb_force = SWIOTLB_FORCE ||
-	    max_pfn > PFN_DOWN(dma32_phys_limit))
-		swiotlb_init(1);
-	else
-		swiotlb_force = SWIOTLB_NO_FORCE;
-#endif
+	swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
 	high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
 	memblock_free_all();
 
arch/s390/mm/init.c +1 -2
index 86ffd0d51fd59..6fb6bf64326f9 100644
@@ -185,8 +185,7 @@ static void pv_init(void)
 		return;
 
 	/* make sure bounce buffers are shared */
-	swiotlb_force = SWIOTLB_FORCE;
-	swiotlb_init(1);
+	swiotlb_init(true, SWIOTLB_FORCE | SWIOTLB_VERBOSE);
 	swiotlb_update_mem_attributes();
 }
 
arch/x86/kernel/pci-dma.c +7 -8
index 7ab7002758396..e0def4b1c3181 100644
@@ -39,6 +39,7 @@ int iommu_detected __read_mostly = 0;
 
 #ifdef CONFIG_SWIOTLB
 bool x86_swiotlb_enable;
+static unsigned int x86_swiotlb_flags;
 
 static void __init pci_swiotlb_detect(void)
 {
@@ -57,16 +58,16 @@ static void __init pci_swiotlb_detect(void)
 	 * Guest with guest memory encryption must always do I/O through a
 	 * bounce buffer as the hypervisor can't access arbitrary VM memory.
 	 */
-	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
-		swiotlb_force = SWIOTLB_FORCE;
-
-	if (swiotlb_force = SWIOTLB_FORCE)
+	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) {
 		x86_swiotlb_enable = true;
+		x86_swiotlb_flags |= SWIOTLB_FORCE;
+	}
 }
 #else
 static inline void __init pci_swiotlb_detect(void)
 {
 }
+#define x86_swiotlb_flags 0
 #endif /* CONFIG_SWIOTLB */
 
 #ifdef CONFIG_SWIOTLB_XEN
@@ -74,8 +75,7 @@ static bool xen_swiotlb;
 
 static void __init pci_xen_swiotlb_init(void)
 {
-	if (!xen_initial_domain() && !x86_swiotlb_enable &&
-	    swiotlb_force != SWIOTLB_FORCE)
+	if (!xen_initial_domain() && !x86_swiotlb_enable)
 		return;
 	x86_swiotlb_enable = false;
 	xen_swiotlb = true;
@@ -119,8 +119,7 @@ void __init pci_iommu_alloc(void)
 	gart_iommu_hole_init();
 	amd_iommu_detect();
 	detect_intel_iommu();
-	if (x86_swiotlb_enable)
-		swiotlb_init(0);
+	swiotlb_init(x86_swiotlb_enable, x86_swiotlb_flags);
 }
 
 /*
drivers/xen/swiotlb-xen.c +2 -2
index 485cd06ed39e7..c2da3eb4826e8 100644
@@ -251,7 +251,7 @@ void __init xen_swiotlb_init_early(void)
 		panic("%s (rc:%d)", xen_swiotlb_error(XEN_SWIOTLB_EFIXUP), rc);
 	}
 
-	if (swiotlb_init_with_tbl(start, nslabs, true))
+	if (swiotlb_init_with_tbl(start, nslabs, SWIOTLB_VERBOSE))
 		panic("Cannot allocate SWIOTLB buffer");
 }
 #endif /* CONFIG_X86 */
@@ -376,7 +376,7 @@ static dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page,
 	/*
 	 * Oh well, have to allocate and map a bounce buffer.
 	 */
-	trace_swiotlb_bounced(dev, dev_addr, size, swiotlb_force);
+	trace_swiotlb_bounced(dev, dev_addr, size);
 
 	map = swiotlb_tbl_map_single(dev, phys, size, size, 0, dir, attrs);
 	if (map = (phys_addr_t)DMA_MAPPING_ERROR)
include/linux/swiotlb.h +7 -8
index 1befd6b2ccf5e..dcecf953f7997 100644
@@ -13,11 +13,8 @@ struct device;
 struct page;
 struct scatterlist;
 
-enum swiotlb_force {
-	SWIOTLB_NORMAL,		/* Default - depending on HW DMA mask etc. */
-	SWIOTLB_FORCE,		/* swiotlb=force */
-	SWIOTLB_NO_FORCE,	/* swiotlb=noforce */
-};
+#define SWIOTLB_VERBOSE	(1 << 0) /* verbose initialization */
+#define SWIOTLB_FORCE	(1 << 1) /* force bounce buffering */
 
 /*
  * Maximum allowable number of contiguous slabs to map,
@@ -36,8 +33,7 @@ enum swiotlb_force {
 /* default to 64MB */
 #define IO_TLB_DEFAULT_SIZE (64UL<<20)
 
-extern void swiotlb_init(int verbose);
-int swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose);
+int swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, unsigned int flags);
 unsigned long swiotlb_size_or_default(void);
 extern int swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs);
 int swiotlb_init_late(size_t size, gfp_t gfp_mask);
@@ -126,13 +122,16 @@ static inline bool is_swiotlb_force_bounce(struct device *dev)
 	return mem && mem->force_bounce;
 }
 
+void swiotlb_init(bool addressing_limited, unsigned int flags);
 void __init swiotlb_exit(void);
 unsigned int swiotlb_max_segment(void);
 size_t swiotlb_max_mapping_size(struct device *dev);
 bool is_swiotlb_active(struct device *dev);
 void __init swiotlb_adjust_size(unsigned long size);
 #else
-#define swiotlb_force SWIOTLB_NO_FORCE
+static inline void swiotlb_init(bool addressing_limited, unsigned int flags)
+{
+}
 static inline bool is_swiotlb_buffer(struct device *dev, phys_addr_t paddr)
 {
 	return false;
include/trace/events/swiotlb.h +10 -19
index 705be43b71ab0..da05c9ebd224a 100644
@@ -8,20 +8,15 @@
 #include <linux/tracepoint.h>
 
 TRACE_EVENT(swiotlb_bounced,
-
-	TP_PROTO(struct device *dev,
-		 dma_addr_t dev_addr,
-		 size_t size,
-		 enum swiotlb_force swiotlb_force),
-
-	TP_ARGS(dev, dev_addr, size, swiotlb_force),
+	TP_PROTO(struct device *dev, dma_addr_t dev_addr, size_t size),
+	TP_ARGS(dev, dev_addr, size),
 
 	TP_STRUCT__entry(
-		__string(	dev_name,	dev_name(dev)		)
-		__field(	u64,	dma_mask			)
-		__field(	dma_addr_t,	dev_addr		)
-		__field(	size_t,	size				)
-		__field(	enum swiotlb_force,	swiotlb_force	)
+		__string(dev_name, dev_name(dev))
+		__field(u64, dma_mask)
+		__field(dma_addr_t, dev_addr)
+		__field(size_t, size)
+		__field(bool, force)
 	),
 
 	TP_fast_assign(
@@ -29,19 +24,15 @@ TRACE_EVENT(swiotlb_bounced,
 		__entry->dma_mask = (dev->dma_mask ? *dev->dma_mask : 0);
 		__entry->dev_addr = dev_addr;
 		__entry->size = size;
-		__entry->swiotlb_force = swiotlb_force;
+		__entry->force = is_swiotlb_force_bounce(dev);
 	),
 
-	TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx "
-		"size=%zu %s",
+	TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx size=%zu %s",
 		__get_str(dev_name),
 		__entry->dma_mask,
 		(unsigned long long)__entry->dev_addr,
 		__entry->size,
-		__print_symbolic(__entry->swiotlb_force,
-			{ SWIOTLB_NORMAL,	"NORMAL" },
-			{ SWIOTLB_FORCE,	"FORCE" },
-			{ SWIOTLB_NO_FORCE,	"NO_FORCE" }))
+		__entry->force ? "FORCE" : "NORMAL")
 );
 
 #endif /*  _TRACE_SWIOTLB_H */
kernel/dma/swiotlb.c +19 -16
index 9a38ea3a46e9f..1a40c71c4d51a 100644
@@ -69,7 +69,8 @@
 
 #define INVALID_PHYS_ADDR (~(phys_addr_t)0)
 
-enum swiotlb_force swiotlb_force;
+static bool swiotlb_force_bounce;
+static bool swiotlb_force_disable;
 
 struct io_tlb_mem io_tlb_default_mem;
 
@@ -88,9 +89,9 @@ setup_io_tlb_npages(char *str)
 	if (*str = ',')
 		++str;
 	if (!strcmp(str, "force"))
-		swiotlb_force = SWIOTLB_FORCE;
+		swiotlb_force_bounce = true;
 	else if (!strcmp(str, "noforce"))
-		swiotlb_force = SWIOTLB_NO_FORCE;
+		swiotlb_force_disable = true;
 
 	return 0;
 }
@@ -211,7 +212,7 @@ static void swiotlb_init_io_tlb_mem(struct io_tlb_mem *mem, phys_addr_t start,
 	mem->index = 0;
 	mem->late_alloc = late_alloc;
 
-	if (swiotlb_force = SWIOTLB_FORCE)
+	if (swiotlb_force_bounce)
 		mem->force_bounce = true;
 
 	spin_lock_init(&mem->lock);
@@ -233,12 +234,13 @@ static void swiotlb_init_io_tlb_mem(struct io_tlb_mem *mem, phys_addr_t start,
 	return;
 }
 
-int __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose)
+int __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs,
+		unsigned int flags)
 {
 	struct io_tlb_mem *mem = &io_tlb_default_mem;
 	size_t alloc_size;
 
-	if (swiotlb_force = SWIOTLB_NO_FORCE)
+	if (swiotlb_force_disable)
 		return 0;
 
 	/* protect against double initialization */
@@ -252,8 +254,9 @@ int __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose)
 		      __func__, alloc_size, PAGE_SIZE);
 
 	swiotlb_init_io_tlb_mem(mem, __pa(tlb), nslabs, false);
+	mem->force_bounce = flags & SWIOTLB_FORCE;
 
-	if (verbose)
+	if (flags & SWIOTLB_VERBOSE)
 		swiotlb_print_info();
 	return 0;
 }
@@ -262,20 +265,21 @@ int __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose)
  * Statically reserve bounce buffer space and initialize bounce buffer data
  * structures for the software IO TLB used to implement the DMA API.
  */
-void  __init
-swiotlb_init(int verbose)
+void __init swiotlb_init(bool addressing_limit, unsigned int flags)
 {
 	size_t bytes = PAGE_ALIGN(default_nslabs << IO_TLB_SHIFT);
 	void *tlb;
 
-	if (swiotlb_force = SWIOTLB_NO_FORCE)
+	if (!addressing_limit && !swiotlb_force_bounce)
+		return;
+	if (swiotlb_force_disable)
 		return;
 
 	/* Get IO TLB memory from the low pages */
 	tlb = memblock_alloc_low(bytes, PAGE_SIZE);
 	if (!tlb)
 		goto fail;
-	if (swiotlb_init_with_tbl(tlb, default_nslabs, verbose))
+	if (swiotlb_init_with_tbl(tlb, default_nslabs, flags))
 		goto fail_free_mem;
 	return;
 
@@ -298,7 +302,7 @@ int swiotlb_init_late(size_t size, gfp_t gfp_mask)
 	unsigned int order;
 	int rc = 0;
 
-	if (swiotlb_force = SWIOTLB_NO_FORCE)
+	if (swiotlb_force_disable)
 		return 0;
 
 	/*
@@ -337,7 +341,7 @@ swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs)
 	struct io_tlb_mem *mem = &io_tlb_default_mem;
 	unsigned long bytes = nslabs << IO_TLB_SHIFT;
 
-	if (swiotlb_force = SWIOTLB_NO_FORCE)
+	if (swiotlb_force_disable)
 		return 0;
 
 	/* protect against double initialization */
@@ -362,7 +366,7 @@ void __init swiotlb_exit(void)
 	unsigned long tbl_vaddr;
 	size_t tbl_size, slots_size;
 
-	if (swiotlb_force = SWIOTLB_FORCE)
+	if (swiotlb_force_bounce)
 		return;
 
 	if (!mem->nslabs)
@@ -709,8 +713,7 @@ dma_addr_t swiotlb_map(struct device *dev, phys_addr_t paddr, size_t size,
 	phys_addr_t swiotlb_addr;
 	dma_addr_t dma_addr;
 
-	trace_swiotlb_bounced(dev, phys_to_dma(dev, paddr), size,
-			      swiotlb_force);
+	trace_swiotlb_bounced(dev, phys_to_dma(dev, paddr), size);
 
 	swiotlb_addr = swiotlb_tbl_map_single(dev, paddr, size, size, 0, dir,
 			attrs);
Signature · 1 line
2.30.2
#13 lore raw

PATCH 10/12 swiotlb: add a SWIOTLB_ANY flag to lift the low memory restriction

In reply to Christoph Hellwig

Power SVM wants to allocate a swiotlb buffer that is not restricted to low memory for the trusted hypervisor scheme. Consolidate the support for this into the swiotlb_init interface by adding a new flag.

---
 arch/powerpc/include/asm/svm.h       |  4 ----
 arch/powerpc/include/asm/swiotlb.h   |  1 +
 arch/powerpc/kernel/dma-swiotlb.c    |  1 +
 arch/powerpc/mm/mem.c                |  5 +----
 arch/powerpc/platforms/pseries/svm.c | 26 +-------------------------
 include/linux/swiotlb.h              |  1 +
 kernel/dma/swiotlb.c                 |  9 +++++++--
 7 files changed, 12 insertions(+), 35 deletions(-)
arch/powerpc/include/asm/svm.h +0 -4
index 7546402d796af..85580b30aba48 100644
@@ -15,8 +15,6 @@ static inline bool is_secure_guest(void)
 	return mfmsr() & MSR_S;
 }
 
-void __init svm_swiotlb_init(void);
-
 void dtl_cache_ctor(void *addr);
 #define get_dtl_cache_ctor()	(is_secure_guest() ? dtl_cache_ctor : NULL)
 
@@ -27,8 +25,6 @@ static inline bool is_secure_guest(void)
 	return false;
 }
 
-static inline void svm_swiotlb_init(void) {}
-
 #define get_dtl_cache_ctor() NULL
 
 #endif /* CONFIG_PPC_SVM */
arch/powerpc/include/asm/swiotlb.h +1 -0
index 3c1a1cd161286..4203b5e0a88ed 100644
@@ -9,6 +9,7 @@
 #include <linux/swiotlb.h>
 
 extern unsigned int ppc_swiotlb_enable;
+extern unsigned int ppc_swiotlb_flags;
 
 #ifdef CONFIG_SWIOTLB
 void swiotlb_detect_4g(void);
arch/powerpc/kernel/dma-swiotlb.c +1 -0
index fc7816126a401..ba256c37bcc0f 100644
@@ -10,6 +10,7 @@
 #include <asm/swiotlb.h>
 
 unsigned int ppc_swiotlb_enable;
+unsigned int ppc_swiotlb_flags;
 
 void __init swiotlb_detect_4g(void)
 {
arch/powerpc/mm/mem.c +1 -4
index e1519e2edc656..a4d65418c30a9 100644
@@ -249,10 +249,7 @@ void __init mem_init(void)
 	 * back to to-down.
 	 */
 	memblock_set_bottom_up(true);
-	if (is_secure_guest())
-		svm_swiotlb_init();
-	else
-		swiotlb_init(ppc_swiotlb_enable, 0);
+	swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags);
 #endif
 
 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
arch/powerpc/platforms/pseries/svm.c +1 -25
index c5228f4969eb2..3b4045d508ec8 100644
@@ -28,7 +28,7 @@ static int __init init_svm(void)
 	 * need to use the SWIOTLB buffer for DMA even if dma_capable() says
 	 * otherwise.
 	 */
-	swiotlb_force = SWIOTLB_FORCE;
+	ppc_swiotlb_flags |= SWIOTLB_ANY | SWIOTLB_FORCE;
 
 	/* Share the SWIOTLB buffer with the host. */
 	swiotlb_update_mem_attributes();
@@ -37,30 +37,6 @@ static int __init init_svm(void)
 }
 machine_early_initcall(pseries, init_svm);
 
-/*
- * Initialize SWIOTLB. Essentially the same as swiotlb_init(), except that it
- * can allocate the buffer anywhere in memory. Since the hypervisor doesn't have
- * any addressing limitation, we don't need to allocate it in low addresses.
- */
-void __init svm_swiotlb_init(void)
-{
-	unsigned char *vstart;
-	unsigned long bytes, io_tlb_nslabs;
-
-	io_tlb_nslabs = (swiotlb_size_or_default() >> IO_TLB_SHIFT);
-	io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
-
-	bytes = io_tlb_nslabs << IO_TLB_SHIFT;
-
-	vstart = memblock_alloc(PAGE_ALIGN(bytes), PAGE_SIZE);
-	if (vstart && !swiotlb_init_with_tbl(vstart, io_tlb_nslabs, false))
-		return;
-
-
-	memblock_free(vstart, PAGE_ALIGN(io_tlb_nslabs << IO_TLB_SHIFT));
-	panic("SVM: Cannot allocate SWIOTLB buffer");
-}
-
 int set_memory_encrypted(unsigned long addr, int numpages)
 {
 	if (!cc_platform_has(CC_ATTR_MEM_ENCRYPT))
include/linux/swiotlb.h +1 -0
index dcecf953f7997..ee655f2e4d28b 100644
@@ -15,6 +15,7 @@ struct scatterlist;
 
 #define SWIOTLB_VERBOSE	(1 << 0) /* verbose initialization */
 #define SWIOTLB_FORCE	(1 << 1) /* force bounce buffering */
+#define SWIOTLB_ANY	(1 << 2) /* allow any memory for the buffer */
 
 /*
  * Maximum allowable number of contiguous slabs to map,
kernel/dma/swiotlb.c +7 -2
index 1a40c71c4d51a..77cf73dc20a78 100644
@@ -275,8 +275,13 @@ void __init swiotlb_init(bool addressing_limit, unsigned int flags)
 	if (swiotlb_force_disable)
 		return;
 
-	/* Get IO TLB memory from the low pages */
-	tlb = memblock_alloc_low(bytes, PAGE_SIZE);
+	/*
+	 * By default allocate the bonuce buffer memory from low memory.
+	 */
+	if (flags & SWIOTLB_ANY)
+		tlb = memblock_alloc(bytes, PAGE_SIZE);
+	else
+		tlb = memblock_alloc_low(bytes, PAGE_SIZE);
 	if (!tlb)
 		goto fail;
 	if (swiotlb_init_with_tbl(tlb, default_nslabs, flags))
Signature · 1 line
2.30.2
#14 lore raw

RE: [PATCH 10/12] swiotlb: add a SWIOTLB_ANY flag to lift the low memory restriction

From: Christoph Hellwig <[email protected]> Sent: Tuesday, March 1, 2022 2:53 AM

Quoted 4 lines

Power SVM wants to allocate a swiotlb buffer that is not restricted to low memory for the trusted hypervisor scheme. Consolidate the support for this into the swiotlb_init interface by adding a new flag.

Hyper-V Isolated VMs want to do the same thing of not restricting the swiotlb buffer to low memory. That's what Tianyu Lan's patch set[1] is proposing. Hyper-V synthetic devices have no DMA addressing limitations, and the likelihood of using a PCI pass-thru device with addressing limitations in an Isolated VM seems vanishingly small.

So could use of the SWIOTLB_ANY flag be generalized? Let Hyper-V init code set the flag before swiotlb_init() is called. Or provide a CONFIG variable that Hyper-V Isolated VMs could set.

Michael

[1] https://lore.kernel.org/lkml/[email protected]/

Quoted 153 lines
---
 arch/powerpc/include/asm/svm.h       |  4 ----
 arch/powerpc/include/asm/swiotlb.h   |  1 +
 arch/powerpc/kernel/dma-swiotlb.c    |  1 +
 arch/powerpc/mm/mem.c                |  5 +----
 arch/powerpc/platforms/pseries/svm.c | 26 +-------------------------
 include/linux/swiotlb.h              |  1 +
 kernel/dma/swiotlb.c                 |  9 +++++++--
 7 files changed, 12 insertions(+), 35 deletions(-)
arch/powerpc/include/asm/svm.h +0 -4
index 7546402d796af..85580b30aba48 100644
@@ -15,8 +15,6 @@ static inline bool is_secure_guest(void)
 	return mfmsr() & MSR_S;
 }
 
-void __init svm_swiotlb_init(void);
-
 void dtl_cache_ctor(void *addr);
 #define get_dtl_cache_ctor()	(is_secure_guest() ? dtl_cache_ctor : NULL)
 
@@ -27,8 +25,6 @@ static inline bool is_secure_guest(void)
 	return false;
 }
 
-static inline void svm_swiotlb_init(void) {}
-
 #define get_dtl_cache_ctor() NULL
 
 #endif /* CONFIG_PPC_SVM */
arch/powerpc/include/asm/swiotlb.h +1 -0
index 3c1a1cd161286..4203b5e0a88ed 100644
@@ -9,6 +9,7 @@
 #include <linux/swiotlb.h>
 
 extern unsigned int ppc_swiotlb_enable;
+extern unsigned int ppc_swiotlb_flags;
 
 #ifdef CONFIG_SWIOTLB
 void swiotlb_detect_4g(void);
arch/powerpc/kernel/dma- renamed from arch/powerpc/kernel/dma-swiotlb.c +0 -0

swiotlb.c
index fc7816126a401..ba256c37bcc0f 100644

arch/powerpc/kernel/dma-swiotlb.c +1 -0
@@ -10,6 +10,7 @@
 #include <asm/swiotlb.h>
 
 unsigned int ppc_swiotlb_enable;
+unsigned int ppc_swiotlb_flags;
 
 void __init swiotlb_detect_4g(void)
 {
arch/powerpc/mm/mem.c index renamed from arch/powerpc/mm/mem.c +0 -0

e1519e2edc656..a4d65418c30a9 100644

arch/powerpc/mm/mem.c +1 -4
@@ -249,10 +249,7 @@ void __init mem_init(void)
 	 * back to to-down.
 	 */
 	memblock_set_bottom_up(true);
-	if (is_secure_guest())
-		svm_swiotlb_init();
-	else
-		swiotlb_init(ppc_swiotlb_enable, 0);
+	swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags);
 #endif
 
 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE); diff --git

a/arch/powerpc/platforms/pseries/svm.c b/arch/powerpc/platforms/pseries/svm.c index c5228f4969eb2..3b4045d508ec8 100644

arch/powerpc/platforms/pseries/svm.c +1 -25
@@ -28,7 +28,7 @@ static int __init init_svm(void)
 	 * need to use the SWIOTLB buffer for DMA even if dma_capable() says
 	 * otherwise.
 	 */
-	swiotlb_force = SWIOTLB_FORCE;
+	ppc_swiotlb_flags |= SWIOTLB_ANY | SWIOTLB_FORCE;
 
 	/* Share the SWIOTLB buffer with the host. */
 	swiotlb_update_mem_attributes();
@@ -37,30 +37,6 @@ static int __init init_svm(void)  }  machine_early_initcall(pseries,
 init_svm);
 
-/*
- * Initialize SWIOTLB. Essentially the same as swiotlb_init(), except that it
- * can allocate the buffer anywhere in memory. Since the hypervisor doesn't have
- * any addressing limitation, we don't need to allocate it in low addresses.
- */
-void __init svm_swiotlb_init(void)
-{
-	unsigned char *vstart;
-	unsigned long bytes, io_tlb_nslabs;
-
-	io_tlb_nslabs = (swiotlb_size_or_default() >> IO_TLB_SHIFT);
-	io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
-
-	bytes = io_tlb_nslabs << IO_TLB_SHIFT;
-
-	vstart = memblock_alloc(PAGE_ALIGN(bytes), PAGE_SIZE);
-	if (vstart && !swiotlb_init_with_tbl(vstart, io_tlb_nslabs, false))
-		return;
-
-
-	memblock_free(vstart, PAGE_ALIGN(io_tlb_nslabs << IO_TLB_SHIFT));
-	panic("SVM: Cannot allocate SWIOTLB buffer");
-}
-
 int set_memory_encrypted(unsigned long addr, int numpages)  {
 	if (!cc_platform_has(CC_ATTR_MEM_ENCRYPT))
include/linux/swiotlb.h index renamed from include/linux/swiotlb.h +0 -0

dcecf953f7997..ee655f2e4d28b 100644

include/linux/swiotlb.h +1 -0
@@ -15,6 +15,7 @@ struct scatterlist;
 
 #define SWIOTLB_VERBOSE	(1 << 0) /* verbose initialization */
 #define SWIOTLB_FORCE	(1 << 1) /* force bounce buffering */
+#define SWIOTLB_ANY	(1 << 2) /* allow any memory for the buffer */
 
 /*
  * Maximum allowable number of contiguous slabs to map, diff --git

a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c index 1a40c71c4d51a..77cf73dc20a78 100644

kernel/dma/swiotlb.c +7 -2
@@ -275,8 +275,13 @@ void __init swiotlb_init(bool addressing_limit, unsigned int
 flags)
 	if (swiotlb_force_disable)
 		return;
 
-	/* Get IO TLB memory from the low pages */
-	tlb = memblock_alloc_low(bytes, PAGE_SIZE);
+	/*
+	 * By default allocate the bonuce buffer memory from low memory.
+	 */
+	if (flags & SWIOTLB_ANY)
+		tlb = memblock_alloc(bytes, PAGE_SIZE);
+	else
+		tlb = memblock_alloc_low(bytes, PAGE_SIZE);
 	if (!tlb)
 		goto fail;

if (swiotlb_init_with_tbl(tlb, default_nslabs, flags))
--
2.30.2

#15 lore raw

Re: [PATCH 10/12] swiotlb: add a SWIOTLB_ANY flag to lift the low memory restriction

Hi Michael,

Quoted 12 lines · Michael Kelley (LINUX)

From: Christoph Hellwig <[email protected]> Sent: Tuesday, March 1, 2022 2:53 AM

Quoted 4 lines

Power SVM wants to allocate a swiotlb buffer that is not restricted to low memory for the trusted hypervisor scheme. Consolidate the support for this into the swiotlb_init interface by adding a new flag.

Hyper-V Isolated VMs want to do the same thing of not restricting the swiotlb buffer to low memory. That's what Tianyu Lan's patch set[1] is proposing. Hyper-V synthetic devices have no DMA addressing limitations, and the likelihood of using a PCI pass-thru device with addressing limitations in an Isolated VM seems vanishingly small.

Quoted 3 lines

So could use of the SWIOTLB_ANY flag be generalized? Let Hyper-V init code set the flag before swiotlb_init() is called. Or provide a CONFIG variable that Hyper-V Isolated VMs could set.

I used to send 64-bit swiotlb, while at that time people thought it was the same as Restricted DMA patchset.

https://lore.kernel.org/all/[email protected]/

However, I do not think Restricted DMA patchset is going to supports 64-bit (or high memory) DMA. Is this what you are looking for?

Dongli Zhang

Quoted 160 lines

Michael

[1] https://urldefense.com/v3/__https://lore.kernel.org/lkml/[email protected]/__;!!ACWV5N9M2RV99hQ!fUx4fMgdQIrqJDDy-pbv9xMeyHX0rC6iN8176LWjylI2_lsjy03gysm0-lAbV1Yb7_g$

Quoted 153 lines
---
 arch/powerpc/include/asm/svm.h       |  4 ----
 arch/powerpc/include/asm/swiotlb.h   |  1 +
 arch/powerpc/kernel/dma-swiotlb.c    |  1 +
 arch/powerpc/mm/mem.c                |  5 +----
 arch/powerpc/platforms/pseries/svm.c | 26 +-------------------------
 include/linux/swiotlb.h              |  1 +
 kernel/dma/swiotlb.c                 |  9 +++++++--
 7 files changed, 12 insertions(+), 35 deletions(-)
arch/powerpc/include/asm/svm.h +0 -4
index 7546402d796af..85580b30aba48 100644
@@ -15,8 +15,6 @@ static inline bool is_secure_guest(void)
 	return mfmsr() & MSR_S;
 }
 
-void __init svm_swiotlb_init(void);
-
 void dtl_cache_ctor(void *addr);
 #define get_dtl_cache_ctor()	(is_secure_guest() ? dtl_cache_ctor : NULL)
 
@@ -27,8 +25,6 @@ static inline bool is_secure_guest(void)
 	return false;
 }
 
-static inline void svm_swiotlb_init(void) {}
-
 #define get_dtl_cache_ctor() NULL
 
 #endif /* CONFIG_PPC_SVM */
arch/powerpc/include/asm/swiotlb.h +1 -0
index 3c1a1cd161286..4203b5e0a88ed 100644
@@ -9,6 +9,7 @@
 #include <linux/swiotlb.h>
 
 extern unsigned int ppc_swiotlb_enable;
+extern unsigned int ppc_swiotlb_flags;
 
 #ifdef CONFIG_SWIOTLB
 void swiotlb_detect_4g(void);
arch/powerpc/kernel/dma- renamed from arch/powerpc/kernel/dma-swiotlb.c +0 -0

swiotlb.c
index fc7816126a401..ba256c37bcc0f 100644

arch/powerpc/kernel/dma-swiotlb.c +1 -0
@@ -10,6 +10,7 @@
 #include <asm/swiotlb.h>
 
 unsigned int ppc_swiotlb_enable;
+unsigned int ppc_swiotlb_flags;
 
 void __init swiotlb_detect_4g(void)
 {
arch/powerpc/mm/mem.c index renamed from arch/powerpc/mm/mem.c +0 -0

e1519e2edc656..a4d65418c30a9 100644

arch/powerpc/mm/mem.c +1 -4
@@ -249,10 +249,7 @@ void __init mem_init(void)
 	 * back to to-down.
 	 */
 	memblock_set_bottom_up(true);
-	if (is_secure_guest())
-		svm_swiotlb_init();
-	else
-		swiotlb_init(ppc_swiotlb_enable, 0);
+	swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags);
 #endif
 
 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE); diff --git

a/arch/powerpc/platforms/pseries/svm.c b/arch/powerpc/platforms/pseries/svm.c index c5228f4969eb2..3b4045d508ec8 100644

arch/powerpc/platforms/pseries/svm.c +1 -25
@@ -28,7 +28,7 @@ static int __init init_svm(void)
 	 * need to use the SWIOTLB buffer for DMA even if dma_capable() says
 	 * otherwise.
 	 */
-	swiotlb_force = SWIOTLB_FORCE;
+	ppc_swiotlb_flags |= SWIOTLB_ANY | SWIOTLB_FORCE;
 
 	/* Share the SWIOTLB buffer with the host. */
 	swiotlb_update_mem_attributes();
@@ -37,30 +37,6 @@ static int __init init_svm(void)  }  machine_early_initcall(pseries,
 init_svm);
 
-/*
- * Initialize SWIOTLB. Essentially the same as swiotlb_init(), except that it
- * can allocate the buffer anywhere in memory. Since the hypervisor doesn't have
- * any addressing limitation, we don't need to allocate it in low addresses.
- */
-void __init svm_swiotlb_init(void)
-{
-	unsigned char *vstart;
-	unsigned long bytes, io_tlb_nslabs;
-
-	io_tlb_nslabs = (swiotlb_size_or_default() >> IO_TLB_SHIFT);
-	io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
-
-	bytes = io_tlb_nslabs << IO_TLB_SHIFT;
-
-	vstart = memblock_alloc(PAGE_ALIGN(bytes), PAGE_SIZE);
-	if (vstart && !swiotlb_init_with_tbl(vstart, io_tlb_nslabs, false))
-		return;
-
-
-	memblock_free(vstart, PAGE_ALIGN(io_tlb_nslabs << IO_TLB_SHIFT));
-	panic("SVM: Cannot allocate SWIOTLB buffer");
-}
-
 int set_memory_encrypted(unsigned long addr, int numpages)  {
 	if (!cc_platform_has(CC_ATTR_MEM_ENCRYPT))
include/linux/swiotlb.h index renamed from include/linux/swiotlb.h +0 -0

dcecf953f7997..ee655f2e4d28b 100644

include/linux/swiotlb.h +1 -0
@@ -15,6 +15,7 @@ struct scatterlist;
 
 #define SWIOTLB_VERBOSE	(1 << 0) /* verbose initialization */
 #define SWIOTLB_FORCE	(1 << 1) /* force bounce buffering */
+#define SWIOTLB_ANY	(1 << 2) /* allow any memory for the buffer */
 
 /*
  * Maximum allowable number of contiguous slabs to map, diff --git

a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c index 1a40c71c4d51a..77cf73dc20a78 100644

kernel/dma/swiotlb.c +7 -2
@@ -275,8 +275,13 @@ void __init swiotlb_init(bool addressing_limit, unsigned int
 flags)
 	if (swiotlb_force_disable)
 		return;
 
-	/* Get IO TLB memory from the low pages */
-	tlb = memblock_alloc_low(bytes, PAGE_SIZE);
+	/*
+	 * By default allocate the bonuce buffer memory from low memory.
+	 */
+	if (flags & SWIOTLB_ANY)
+		tlb = memblock_alloc(bytes, PAGE_SIZE);
+	else
+		tlb = memblock_alloc_low(bytes, PAGE_SIZE);
 	if (!tlb)
 		goto fail;

if (swiotlb_init_with_tbl(tlb, default_nslabs, flags))
--
2.30.2

#16 lore raw

RE: [PATCH 10/12] swiotlb: add a SWIOTLB_ANY flag to lift the low memory restriction

From: Dongli Zhang <[email protected]> Sent: Friday, March 4, 2022 10:28 AM

Quoted 20 lines

Hi Michael,

Quoted 12 lines · Michael Kelley (LINUX)

From: Christoph Hellwig <[email protected]> Sent: Tuesday, March 1, 2022 2:53 AM

Quoted 4 lines

Power SVM wants to allocate a swiotlb buffer that is not restricted to low memory for the trusted hypervisor scheme. Consolidate the support for this into the swiotlb_init interface by adding a new flag.

Hyper-V Isolated VMs want to do the same thing of not restricting the swiotlb buffer to low memory. That's what Tianyu Lan's patch set[1] is proposing. Hyper-V synthetic devices have no DMA addressing limitations, and the likelihood of using a PCI pass-thru device with addressing limitations in an Isolated VM seems vanishingly small.

Quoted 3 lines

So could use of the SWIOTLB_ANY flag be generalized? Let Hyper-V init code set the flag before swiotlb_init() is called. Or provide a CONFIG variable that Hyper-V Isolated VMs could set.

Quoted 7 lines

I used to send 64-bit swiotlb, while at that time people thought it was the same as Restricted DMA patchset.

https://lore.kernel.org/all/[email protected]/

However, I do not think Restricted DMA patchset is going to supports 64-bit (or high memory) DMA. Is this what you are looking for?

Yes, it looks like your patchset would do what we want for Hyper-V Isolated VMs, but it is a more complex solution than is needed. My assertion is that in some environments, such as Hyper-V Isolated VMs, we're willing to assume all devices are 64-bit DMA capable, and to stop carrying the legacy baggage. Bounce buffering is used for a different scenario (memory encryption), and
the bounce buffers can be allocated in high memory. There's no need for a
2nd swiotlb buffer.

Michael

#17 lore raw

PATCH 11/12 swiotlb: merge swiotlb-xen initialization into swiotlb

In reply to Christoph Hellwig

Allow to pass a remap argument to the swiotlb initialization functions to handle the Xen/x86 remap case. ARM/ARM64 never did any remapping from xen_swiotlb_fixup, so we don't even need that quirk.

---
 arch/arm/xen/mm.c               |  23 +++---
 arch/x86/include/asm/xen/page.h |   5 --
 arch/x86/kernel/pci-dma.c       |  19 +++--
 arch/x86/pci/sta2x11-fixup.c    |   2 +-
 drivers/xen/swiotlb-xen.c       | 128 +-------------------------------
 include/linux/swiotlb.h         |   7 +-
 include/xen/arm/page.h          |   1 -
 include/xen/swiotlb-xen.h       |   8 +-
 kernel/dma/swiotlb.c            | 120 +++++++++++++++---------------
 9 files changed, 96 insertions(+), 217 deletions(-)
arch/arm/xen/mm.c +13 -10
index a7e54a087b802..58b40f87617d3 100644
@@ -23,22 +23,20 @@
 #include <asm/xen/hypercall.h>
 #include <asm/xen/interface.h>
 
-unsigned long xen_get_swiotlb_free_pages(unsigned int order)
+static gfp_t xen_swiotlb_gfp(void)
 {
 	phys_addr_t base;
-	gfp_t flags = __GFP_NOWARN|__GFP_KSWAPD_RECLAIM;
 	u64 i;
 
 	for_each_mem_range(i, &base, NULL) {
 		if (base < (phys_addr_t)0xffffffff) {
 			if (IS_ENABLED(CONFIG_ZONE_DMA32))
-				flags |= __GFP_DMA32;
-			else
-				flags |= __GFP_DMA;
-			break;
+				return __GFP_DMA32;
+			return __GFP_DMA;
 		}
 	}
-	return __get_free_pages(flags, order);
+
+	return GFP_KERNEL;
 }
 
 static bool hypercall_cflush = false;
@@ -143,10 +141,15 @@ static int __init xen_mm_init(void)
 	if (!xen_swiotlb_detect())
 		return 0;
 
-	rc = xen_swiotlb_init();
 	/* we can work with the default swiotlb */
-	if (rc < 0 && rc != -EEXIST)
-		return rc;
+	if (!io_tlb_default_mem.nslabs) {
+		if (!xen_initial_domain())
+			return -EINVAL;
+		rc = swiotlb_init_late(swiotlb_size_or_default(),
+				       xen_swiotlb_gfp(), NULL);
+		if (rc < 0)
+			return rc;
+	}
 
 	cflush.op = 0;
 	cflush.a.dev_bus_addr = 0;
arch/x86/include/asm/xen/page.h +0 -5
index e989bc2269f54..1fc67df500145 100644
@@ -357,9 +357,4 @@ static inline bool xen_arch_need_swiotlb(struct device *dev,
 	return false;
 }
 
-static inline unsigned long xen_get_swiotlb_free_pages(unsigned int order)
-{
-	return __get_free_pages(__GFP_NOWARN, order);
-}
-
 #endif /* _ASM_X86_XEN_PAGE_H */
arch/x86/kernel/pci-dma.c +9 -10
index e0def4b1c3181..2f2c468acb955 100644
@@ -71,15 +71,12 @@ static inline void __init pci_swiotlb_detect(void)
 #endif /* CONFIG_SWIOTLB */
 
 #ifdef CONFIG_SWIOTLB_XEN
-static bool xen_swiotlb;
-
 static void __init pci_xen_swiotlb_init(void)
 {
 	if (!xen_initial_domain() && !x86_swiotlb_enable)
 		return;
 	x86_swiotlb_enable = false;
-	xen_swiotlb = true;
-	xen_swiotlb_init_early();
+	swiotlb_init_remap(true, x86_swiotlb_flags, xen_swiotlb_fixup);
 	dma_ops = &xen_swiotlb_dma_ops;
 	if (IS_ENABLED(CONFIG_PCI))
 		pci_request_acs();
@@ -87,14 +84,16 @@ static void __init pci_xen_swiotlb_init(void)
 
 int pci_xen_swiotlb_init_late(void)
 {
-	int rc;
-
-	if (xen_swiotlb)
+	if (dma_ops = &xen_swiotlb_dma_ops)
 		return 0;
 
-	rc = xen_swiotlb_init();
-	if (rc)
-		return rc;
+	/* we can work with the default swiotlb */
+	if (!io_tlb_default_mem.nslabs) {
+		int rc = swiotlb_init_late(swiotlb_size_or_default(),
+					   GFP_KERNEL, xen_swiotlb_fixup);
+		if (rc < 0)
+			return rc;
+	}
 
 	/* XXX: this switches the dma ops under live devices! */
 	dma_ops = &xen_swiotlb_dma_ops;
arch/x86/pci/sta2x11-fixup.c +1 -1
index c7e6faf59a861..7368afc039987 100644
@@ -57,7 +57,7 @@ static void sta2x11_new_instance(struct pci_dev *pdev)
 		int size = STA2X11_SWIOTLB_SIZE;
 		/* First instance: register your own swiotlb area */
 		dev_info(&pdev->dev, "Using SWIOTLB (size %i)\n", size);
-		if (swiotlb_init_late(size, GFP_DMA))
+		if (swiotlb_init_late(size, GFP_DMA, NULL))
 			dev_emerg(&pdev->dev, "init swiotlb failed\n");
 	}
 	list_add(&instance->list, &sta2x11_instance_list);
drivers/xen/swiotlb-xen.c +1 -127

140 lines not on this page · raw message

include/linux/swiotlb.h +4 -3
index ee655f2e4d28b..919cf82ed978e 100644
@@ -34,10 +34,11 @@ struct scatterlist;
 /* default to 64MB */
 #define IO_TLB_DEFAULT_SIZE (64UL<<20)
 
-int swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, unsigned int flags);
 unsigned long swiotlb_size_or_default(void);
-extern int swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs);
-int swiotlb_init_late(size_t size, gfp_t gfp_mask);
+int swiotlb_init_late(size_t size, gfp_t gfp_mask,
+		int (*remap)(void *tlb, unsigned long nslabs));
+void __init swiotlb_init_remap(bool addressing_limit, unsigned int flags,
+		int (*remap)(void *tlb, unsigned long nslabs));
 extern void __init swiotlb_update_mem_attributes(void);
 
 phys_addr_t swiotlb_tbl_map_single(struct device *hwdev, phys_addr_t phys,
include/xen/arm/page.h +0 -1
index ac1b654705631..7e199c6656b90 100644
@@ -115,6 +115,5 @@ static inline bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
 bool xen_arch_need_swiotlb(struct device *dev,
 			   phys_addr_t phys,
 			   dma_addr_t dev_addr);
-unsigned long xen_get_swiotlb_free_pages(unsigned int order);
 
 #endif /* _ASM_ARM_XEN_PAGE_H */
include/xen/swiotlb-xen.h +6 -2
index b3e647f86e3e2..590ceb923f0c8 100644
@@ -10,8 +10,12 @@ void xen_dma_sync_for_cpu(struct device *dev, dma_addr_t handle,
 void xen_dma_sync_for_device(struct device *dev, dma_addr_t handle,
 			     size_t size, enum dma_data_direction dir);
 
-int xen_swiotlb_init(void);
-void __init xen_swiotlb_init_early(void);
+#ifdef CONFIG_SWIOTLB_XEN
+int xen_swiotlb_fixup(void *buf, unsigned long nslabs);
+#else
+#define xen_swiotlb_fixup NULL
+#endif
+
 extern const struct dma_map_ops xen_swiotlb_dma_ops;
 
 #endif /* __LINUX_SWIOTLB_XEN_H */
kernel/dma/swiotlb.c +62 -58

175 lines not on this page · raw message

Signature · 1 line
2.30.2
maintainer Tested-by on v1 11/12
#18 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 37 lines · Christoph Hellwig

Allow to pass a remap argument to the swiotlb initialization functions to handle the Xen/x86 remap case. ARM/ARM64 never did any remapping from xen_swiotlb_fixup, so we don't even need that quirk.

---
 arch/arm/xen/mm.c               |  23 +++---
 arch/x86/include/asm/xen/page.h |   5 --
 arch/x86/kernel/pci-dma.c       |  19 +++--
 arch/x86/pci/sta2x11-fixup.c    |   2 +-
 drivers/xen/swiotlb-xen.c       | 128 +-------------------------------
 include/linux/swiotlb.h         |   7 +-
 include/xen/arm/page.h          |   1 -
 include/xen/swiotlb-xen.h       |   8 +-
 kernel/dma/swiotlb.c            | 120 +++++++++++++++---------------
 9 files changed, 96 insertions(+), 217 deletions(-)
arch/arm/xen/mm.c +1 -4
index a7e54a087b802..58b40f87617d3 100644
@@ -23,22 +23,20 @@
 #include <asm/xen/hypercall.h>
 #include <asm/xen/interface.h>
 
-unsigned long xen_get_swiotlb_free_pages(unsigned int order)
+static gfp_t xen_swiotlb_gfp(void)
 {
 	phys_addr_t base;
-	gfp_t flags = __GFP_NOWARN|__GFP_KSWAPD_RECLAIM;
 	u64 i;
 
 	for_each_mem_range(i, &base, NULL) {
 		if (base < (phys_addr_t)0xffffffff) {
 			if (IS_ENABLED(CONFIG_ZONE_DMA32))
-				flags |= __GFP_DMA32;
-			else

arch/arm/xen/mm.c line 40

Quoted 10 lines
arch/arm/xen/mm.c +4 -3
-				flags |= __GFP_DMA;
-			break;
+				return __GFP_DMA32;
+			return __GFP_DMA;
 		}
 	}
-	return __get_free_pages(flags, order);
+
+	return GFP_KERNEL;
 }

Unrelated to this specific patch series: now that I think about it, if io_tlb_default_mem.nslabs is already allocated by the time xen_mm_init is called, wouldn't we potentially have an issue with the GFP flags used for the earlier allocation (e.g. GFP_DMA32 not used)? Maybe something for another day.

arch/arm/xen/mm.c static int __init xen_mm_init(void) · line 147

Quoted 12 lines
arch/arm/xen/mm.c +3 -3
 static bool hypercall_cflush = false;
@@ -143,10 +141,15 @@ static int __init xen_mm_init(void)
 	if (!xen_swiotlb_detect())
 		return 0;
 
-	rc = xen_swiotlb_init();
 	/* we can work with the default swiotlb */
-	if (rc < 0 && rc != -EEXIST)
-		return rc;
+	if (!io_tlb_default_mem.nslabs) {
+		if (!xen_initial_domain())
+			return -EINVAL;

I don't think we need this xen_initial_domain() check. It is all already sorted out by the xen_swiotlb_detect() check above.

Aside from that the rest looks OK. Also, you can add my:

Quoted 453 lines
arch/arm/xen/mm.c +5 -0
+		rc = swiotlb_init_late(swiotlb_size_or_default(),
+				       xen_swiotlb_gfp(), NULL);
+		if (rc < 0)
+			return rc;
+	}
 
 	cflush.op = 0;
 	cflush.a.dev_bus_addr = 0;
arch/x86/include/asm/xen/page.h +0 -5
index e989bc2269f54..1fc67df500145 100644
@@ -357,9 +357,4 @@ static inline bool xen_arch_need_swiotlb(struct device *dev,
 	return false;
 }
 
-static inline unsigned long xen_get_swiotlb_free_pages(unsigned int order)
-{
-	return __get_free_pages(__GFP_NOWARN, order);
-}
-
 #endif /* _ASM_X86_XEN_PAGE_H */
arch/x86/kernel/pci-dma.c +9 -10
index e0def4b1c3181..2f2c468acb955 100644
@@ -71,15 +71,12 @@ static inline void __init pci_swiotlb_detect(void)
 #endif /* CONFIG_SWIOTLB */
 
 #ifdef CONFIG_SWIOTLB_XEN
-static bool xen_swiotlb;
-
 static void __init pci_xen_swiotlb_init(void)
 {
 	if (!xen_initial_domain() && !x86_swiotlb_enable)
 		return;
 	x86_swiotlb_enable = false;
-	xen_swiotlb = true;
-	xen_swiotlb_init_early();
+	swiotlb_init_remap(true, x86_swiotlb_flags, xen_swiotlb_fixup);
 	dma_ops = &xen_swiotlb_dma_ops;
 	if (IS_ENABLED(CONFIG_PCI))
 		pci_request_acs();
@@ -87,14 +84,16 @@ static void __init pci_xen_swiotlb_init(void)
 
 int pci_xen_swiotlb_init_late(void)
 {
-	int rc;
-
-	if (xen_swiotlb)
+	if (dma_ops = &xen_swiotlb_dma_ops)
 		return 0;
 
-	rc = xen_swiotlb_init();
-	if (rc)
-		return rc;
+	/* we can work with the default swiotlb */
+	if (!io_tlb_default_mem.nslabs) {
+		int rc = swiotlb_init_late(swiotlb_size_or_default(),
+					   GFP_KERNEL, xen_swiotlb_fixup);
+		if (rc < 0)
+			return rc;
+	}
 
 	/* XXX: this switches the dma ops under live devices! */
 	dma_ops = &xen_swiotlb_dma_ops;
drivers/xen/swiotlb-xen.c +1 -127
index c2da3eb4826e8..df8085b50df10 100644
@@ -104,7 +104,7 @@ static int is_xen_swiotlb_buffer(struct device *dev, dma_addr_t dma_addr)
 	return 0;
 }
 
-static int xen_swiotlb_fixup(void *buf, unsigned long nslabs)
+int xen_swiotlb_fixup(void *buf, unsigned long nslabs)
 {
 	int rc;
 	unsigned int order = get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT);
@@ -130,132 +130,6 @@ static int xen_swiotlb_fixup(void *buf, unsigned long nslabs)
 	return 0;
 }
 
-enum xen_swiotlb_err {
-	XEN_SWIOTLB_UNKNOWN = 0,
-	XEN_SWIOTLB_ENOMEM,
-	XEN_SWIOTLB_EFIXUP
-};
-
-static const char *xen_swiotlb_error(enum xen_swiotlb_err err)
-{
-	switch (err) {
-	case XEN_SWIOTLB_ENOMEM:
-		return "Cannot allocate Xen-SWIOTLB buffer\n";
-	case XEN_SWIOTLB_EFIXUP:
-		return "Failed to get contiguous memory for DMA from Xen!\n"\
-		    "You either: don't have the permissions, do not have"\
-		    " enough free memory under 4GB, or the hypervisor memory"\
-		    " is too fragmented!";
-	default:
-		break;
-	}
-	return "";
-}
-
-int xen_swiotlb_init(void)
-{
-	enum xen_swiotlb_err m_ret = XEN_SWIOTLB_UNKNOWN;
-	unsigned long bytes = swiotlb_size_or_default();
-	unsigned long nslabs = bytes >> IO_TLB_SHIFT;
-	unsigned int order, repeat = 3;
-	int rc = -ENOMEM;
-	char *start;
-
-	if (io_tlb_default_mem.nslabs) {
-		pr_warn("swiotlb buffer already initialized\n");
-		return -EEXIST;
-	}
-
-retry:
-	m_ret = XEN_SWIOTLB_ENOMEM;
-	order = get_order(bytes);
-
-	/*
-	 * Get IO TLB memory from any location.
-	 */
-#define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT))
-#define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT)
-	while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) {
-		start = (void *)xen_get_swiotlb_free_pages(order);
-		if (start)
-			break;
-		order--;
-	}
-	if (!start)
-		goto exit;
-	if (order != get_order(bytes)) {
-		pr_warn("Warning: only able to allocate %ld MB for software IO TLB\n",
-			(PAGE_SIZE << order) >> 20);
-		nslabs = SLABS_PER_PAGE << order;
-		bytes = nslabs << IO_TLB_SHIFT;
-	}
-
-	/*
-	 * And replace that memory with pages under 4GB.
-	 */
-	rc = xen_swiotlb_fixup(start, nslabs);
-	if (rc) {
-		free_pages((unsigned long)start, order);
-		m_ret = XEN_SWIOTLB_EFIXUP;
-		goto error;
-	}
-	rc = swiotlb_late_init_with_tbl(start, nslabs);
-	if (rc)
-		return rc;
-	return 0;
-error:
-	if (nslabs > 1024 && repeat--) {
-		/* Min is 2MB */
-		nslabs = max(1024UL, ALIGN(nslabs >> 1, IO_TLB_SEGSIZE));
-		bytes = nslabs << IO_TLB_SHIFT;
-		pr_info("Lowering to %luMB\n", bytes >> 20);
-		goto retry;
-	}
-exit:
-	pr_err("%s (rc:%d)\n", xen_swiotlb_error(m_ret), rc);
-	return rc;
-}
-
-#ifdef CONFIG_X86
-void __init xen_swiotlb_init_early(void)
-{
-	unsigned long bytes = swiotlb_size_or_default();
-	unsigned long nslabs = bytes >> IO_TLB_SHIFT;
-	unsigned int repeat = 3;
-	char *start;
-	int rc;
-
-retry:
-	/*
-	 * Get IO TLB memory from any location.
-	 */
-	start = memblock_alloc(PAGE_ALIGN(bytes),
-			       IO_TLB_SEGSIZE << IO_TLB_SHIFT);
-	if (!start)
-		panic("%s: Failed to allocate %lu bytes\n",
-		      __func__, PAGE_ALIGN(bytes));
-
-	/*
-	 * And replace that memory with pages under 4GB.
-	 */
-	rc = xen_swiotlb_fixup(start, nslabs);
-	if (rc) {
-		memblock_free(start, PAGE_ALIGN(bytes));
-		if (nslabs > 1024 && repeat--) {
-			/* Min is 2MB */
-			nslabs = max(1024UL, ALIGN(nslabs >> 1, IO_TLB_SEGSIZE));
-			bytes = nslabs << IO_TLB_SHIFT;
-			pr_info("Lowering to %luMB\n", bytes >> 20);
-			goto retry;
-		}
-		panic("%s (rc:%d)", xen_swiotlb_error(XEN_SWIOTLB_EFIXUP), rc);
-	}
-
-	if (swiotlb_init_with_tbl(start, nslabs, SWIOTLB_VERBOSE))
-		panic("Cannot allocate SWIOTLB buffer");
-}
-#endif /* CONFIG_X86 */
-
 static void *
 xen_swiotlb_alloc_coherent(struct device *hwdev, size_t size,
 			   dma_addr_t *dma_handle, gfp_t flags,
include/linux/swiotlb.h +4 -3
index ee655f2e4d28b..919cf82ed978e 100644
@@ -34,10 +34,11 @@ struct scatterlist;
 /* default to 64MB */
 #define IO_TLB_DEFAULT_SIZE (64UL<<20)
 
-int swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, unsigned int flags);
 unsigned long swiotlb_size_or_default(void);
-extern int swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs);
-int swiotlb_init_late(size_t size, gfp_t gfp_mask);
+int swiotlb_init_late(size_t size, gfp_t gfp_mask,
+		int (*remap)(void *tlb, unsigned long nslabs));
+void __init swiotlb_init_remap(bool addressing_limit, unsigned int flags,
+		int (*remap)(void *tlb, unsigned long nslabs));
 extern void __init swiotlb_update_mem_attributes(void);
 
 phys_addr_t swiotlb_tbl_map_single(struct device *hwdev, phys_addr_t phys,
include/xen/arm/page.h +0 -1
index ac1b654705631..7e199c6656b90 100644
@@ -115,6 +115,5 @@ static inline bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
 bool xen_arch_need_swiotlb(struct device *dev,
 			   phys_addr_t phys,
 			   dma_addr_t dev_addr);
-unsigned long xen_get_swiotlb_free_pages(unsigned int order);
 #endif /* _ASM_ARM_XEN_PAGE_H */
include/xen/swiotlb-xen.h +6 -2
index b3e647f86e3e2..590ceb923f0c8 100644
@@ -10,8 +10,12 @@ void xen_dma_sync_for_cpu(struct device *dev, dma_addr_t handle,
 void xen_dma_sync_for_device(struct device *dev, dma_addr_t handle,
 			     size_t size, enum dma_data_direction dir);
 
-int xen_swiotlb_init(void);
-void __init xen_swiotlb_init_early(void);
+#ifdef CONFIG_SWIOTLB_XEN
+int xen_swiotlb_fixup(void *buf, unsigned long nslabs);
+#else
+#define xen_swiotlb_fixup NULL
+#endif
+
 extern const struct dma_map_ops xen_swiotlb_dma_ops;
 
 #endif /* __LINUX_SWIOTLB_XEN_H */
kernel/dma/swiotlb.c +62 -58
index 77cf73dc20a78..128363dc9b5bb 100644
@@ -234,40 +234,17 @@ static void swiotlb_init_io_tlb_mem(struct io_tlb_mem *mem, phys_addr_t start,
 	return;
 }
 
-int __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs,
-		unsigned int flags)
-{
-	struct io_tlb_mem *mem = &io_tlb_default_mem;
-	size_t alloc_size;
-
-	if (swiotlb_force_disable)
-		return 0;
-
-	/* protect against double initialization */
-	if (WARN_ON_ONCE(mem->nslabs))
-		return -ENOMEM;
-
-	alloc_size = PAGE_ALIGN(array_size(sizeof(*mem->slots), nslabs));
-	mem->slots = memblock_alloc(alloc_size, PAGE_SIZE);
-	if (!mem->slots)
-		panic("%s: Failed to allocate %zu bytes align=0x%lx\n",
-		      __func__, alloc_size, PAGE_SIZE);
-
-	swiotlb_init_io_tlb_mem(mem, __pa(tlb), nslabs, false);
-	mem->force_bounce = flags & SWIOTLB_FORCE;
-
-	if (flags & SWIOTLB_VERBOSE)
-		swiotlb_print_info();
-	return 0;
-}
-
 /*
  * Statically reserve bounce buffer space and initialize bounce buffer data
  * structures for the software IO TLB used to implement the DMA API.
  */
-void __init swiotlb_init(bool addressing_limit, unsigned int flags)
+void __init swiotlb_init_remap(bool addressing_limit, unsigned int flags,
+		int (*remap)(void *tlb, unsigned long nslabs))
 {
-	size_t bytes = PAGE_ALIGN(default_nslabs << IO_TLB_SHIFT);
+	struct io_tlb_mem *mem = &io_tlb_default_mem;
+	unsigned long nslabs = default_nslabs;
+	size_t alloc_size = PAGE_ALIGN(array_size(sizeof(*mem->slots), nslabs));
+	size_t bytes;
 	void *tlb;
 
 	if (!addressing_limit && !swiotlb_force_bounce)
@@ -275,23 +252,48 @@ void __init swiotlb_init(bool addressing_limit, unsigned int flags)
 	if (swiotlb_force_disable)
 		return;
 
+	/* protect against double initialization */
+	if (WARN_ON_ONCE(mem->nslabs))
+		return;
+
 	/*
 	 * By default allocate the bonuce buffer memory from low memory.
 	 */
+retry:
+	bytes = PAGE_ALIGN(default_nslabs << IO_TLB_SHIFT);
 	if (flags & SWIOTLB_ANY)
 		tlb = memblock_alloc(bytes, PAGE_SIZE);
 	else
 		tlb = memblock_alloc_low(bytes, PAGE_SIZE);
 	if (!tlb)
-		goto fail;
-	if (swiotlb_init_with_tbl(tlb, default_nslabs, flags))
-		goto fail_free_mem;
-	return;
+		panic("%s: failed to allocate tlb structure\n", __func__);
+
+	if (remap && remap(tlb, nslabs) < 0) {
+		memblock_free(tlb, PAGE_ALIGN(bytes));
+
+		/* Min is 2MB */
+		if (nslabs <= 1024)
+			panic("%s: Failed to remap %zu bytes\n",
+			      __func__, bytes);
+		nslabs = max(1024UL, ALIGN(nslabs >> 1, IO_TLB_SEGSIZE));
+		goto retry;
+	}
+
+	mem->slots = memblock_alloc(alloc_size, PAGE_SIZE);
+	if (!mem->slots)
+		panic("%s: Failed to allocate %zu bytes align=0x%lx\n",
+		      __func__, alloc_size, PAGE_SIZE);
 
-fail_free_mem:
-	memblock_free(tlb, bytes);
-fail:
-	pr_warn("Cannot allocate buffer");
+	swiotlb_init_io_tlb_mem(mem, __pa(tlb), default_nslabs, false);
+	mem->force_bounce = flags & SWIOTLB_FORCE;
+
+	if (flags & SWIOTLB_VERBOSE)
+		swiotlb_print_info();
+}
+
+void __init swiotlb_init(bool addressing_limit, unsigned int flags)
+{
+	return swiotlb_init_remap(addressing_limit, flags, NULL);
 }
 
 /*
@@ -299,8 +301,10 @@ void __init swiotlb_init(bool addressing_limit, unsigned int flags)
  * initialize the swiotlb later using the slab allocator if needed.
  * This should be just like above, but with some error catching.
  */
-int swiotlb_init_late(size_t size, gfp_t gfp_mask)
+int swiotlb_init_late(size_t size, gfp_t gfp_mask,
+		int (*remap)(void *tlb, unsigned long nslabs))
 {
+	struct io_tlb_mem *mem = &io_tlb_default_mem;
 	unsigned long nslabs = ALIGN(size >> IO_TLB_SHIFT, IO_TLB_SEGSIZE);
 	unsigned long bytes;
 	unsigned char *vstart = NULL;
@@ -310,9 +314,14 @@ int swiotlb_init_late(size_t size, gfp_t gfp_mask)
 	if (swiotlb_force_disable)
 		return 0;
 
+	/* protect against double initialization */
+	if (WARN_ON_ONCE(mem->nslabs))
+		return -ENOMEM;
+
 	/*
 	 * Get IO TLB memory from the low pages
 	 */
+retry:
 	order = get_order(nslabs << IO_TLB_SHIFT);
 	nslabs = SLABS_PER_PAGE << order;
 	bytes = nslabs << IO_TLB_SHIFT;
@@ -333,33 +342,28 @@ int swiotlb_init_late(size_t size, gfp_t gfp_mask)
 			(PAGE_SIZE << order) >> 20);
 		nslabs = SLABS_PER_PAGE << order;
 	}
-	rc = swiotlb_late_init_with_tbl(vstart, nslabs);
-	if (rc)
-		free_pages((unsigned long)vstart, order);
-
-	return rc;
-}
-
-int
-swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs)
-{
-	struct io_tlb_mem *mem = &io_tlb_default_mem;
-	unsigned long bytes = nslabs << IO_TLB_SHIFT;
 
-	if (swiotlb_force_disable)
-		return 0;
+	if (remap)
+		rc = remap(vstart, nslabs);
+	if (rc) {
+		free_pages((unsigned long)vstart, order);
 
-	/* protect against double initialization */
-	if (WARN_ON_ONCE(mem->nslabs))
-		return -ENOMEM;
+		/* Min is 2MB */
+		if (nslabs <= 1024)
+			return rc;
+		nslabs = max(1024UL, ALIGN(nslabs >> 1, IO_TLB_SEGSIZE));
+		goto retry;
+	}
 
 	mem->slots = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 		get_order(array_size(sizeof(*mem->slots), nslabs)));
-	if (!mem->slots)
+	if (!mem->slots) {
+		free_pages((unsigned long)vstart, order);
 		return -ENOMEM;
+	}
 
-	set_memory_decrypted((unsigned long)tlb, bytes >> PAGE_SHIFT);
-	swiotlb_init_io_tlb_mem(mem, virt_to_phys(tlb), nslabs, true);
+	set_memory_decrypted((unsigned long)vstart, bytes >> PAGE_SHIFT);
+	swiotlb_init_io_tlb_mem(mem, virt_to_phys(vstart), nslabs, true);
 
 	swiotlb_print_info();
 	return 0;
Signature · 7 lines
2.30.2


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#19 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 5 lines · Stefano Stabellini

Unrelated to this specific patch series: now that I think about it, if io_tlb_default_mem.nslabs is already allocated by the time xen_mm_init is called, wouldn't we potentially have an issue with the GFP flags used for the earlier allocation (e.g. GFP_DMA32 not used)? Maybe something for another day.

swiotlb_init allocates low memory from meblock, which is roughly equivalent to GFP_DMA allocations, so we'll be fine.

Quoted 4 lines
Quoted 4 lines
arch/arm/xen/mm.c +0 -0
@@ -143,10 +141,15 @@ static int __init xen_mm_init(void)
 	if (!xen_swiotlb_detect())
 		return 0;
 
Quoted 10 lines

arch/arm/xen/mm.c static int __init xen_mm_init(void) · line 147

Quoted 7 lines
arch/arm/xen/mm.c +3 -3
-	rc = xen_swiotlb_init();
 	/* we can work with the default swiotlb */
-	if (rc < 0 && rc != -EEXIST)
-		return rc;
+	if (!io_tlb_default_mem.nslabs) {
+		if (!xen_initial_domain())
+			return -EINVAL;

I don't think we need this xen_initial_domain() check. It is all already sorted out by the xen_swiotlb_detect() check above.

Is it?

static inline int xen_swiotlb_detect(void)

{
	if (!xen_domain())
		return 0;
	if (xen_feature(XENFEAT_direct_mapped))
		return 1;
	/* legacy case */
	if (!xen_feature(XENFEAT_not_direct_mapped) && xen_initial_domain())
		return 1;
	return 0;
}

I think I'd keep it as-is for now, as my planned next step would be to fold xen-swiotlb into swiotlb entirely.

#20 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 28 lines · Christoph Hellwig
Quoted 5 lines · Stefano Stabellini

Unrelated to this specific patch series: now that I think about it, if io_tlb_default_mem.nslabs is already allocated by the time xen_mm_init is called, wouldn't we potentially have an issue with the GFP flags used for the earlier allocation (e.g. GFP_DMA32 not used)? Maybe something for another day.

swiotlb_init allocates low memory from meblock, which is roughly equivalent to GFP_DMA allocations, so we'll be fine.

Quoted 4 lines
Quoted 4 lines
arch/arm/xen/mm.c +0 -0
@@ -143,10 +141,15 @@ static int __init xen_mm_init(void)
 	if (!xen_swiotlb_detect())
 		return 0;
 
Quoted 10 lines

arch/arm/xen/mm.c static int __init xen_mm_init(void) · line 147

Quoted 7 lines
arch/arm/xen/mm.c +3 -3
-	rc = xen_swiotlb_init();
 	/* we can work with the default swiotlb */
-	if (rc < 0 && rc != -EEXIST)
-		return rc;
+	if (!io_tlb_default_mem.nslabs) {
+		if (!xen_initial_domain())
+			return -EINVAL;

I don't think we need this xen_initial_domain() check. It is all already sorted out by the xen_swiotlb_detect() check above.

Is it?

static inline int xen_swiotlb_detect(void)

Quoted 10 lines
{
	if (!xen_domain())
		return 0;
	if (xen_feature(XENFEAT_direct_mapped))
		return 1;
	/* legacy case */
	if (!xen_feature(XENFEAT_not_direct_mapped) && xen_initial_domain())
		return 1;
	return 0;
}

It used to be that we had a

if (!xen_initial_domain())

      return -EINVAL;

check in the initialization of swiotlb-xen on ARM. Then we replaced it with the more sophisticated xen_swiotlb_detect().

The reason is that swiotlb-xen on ARM relies on Dom0 being 1:1 mapped (guest physical addresses = physical addresses). Recent changes in Xen allowed also DomUs to be 1:1 mapped. Changes still under discussion will allow Dom0 not to be 1:1 mapped.

So, before all the Xen-side changes, knowing what was going to happen, I introduced a clearer interface: XENFEAT_direct_mapped and XENFEAT_not_direct_mapped tell us whether the guest (Linux) is 1:1 mapped or not. If it is 1:1 mapped then Linux can take advantage of swiotlb-xen. Now xen_swiotlb_detect() returns true if Linux is 1:1 mapped.

Then of course there is the legacy case. That's taken care of by:

if (!xen_feature(XENFEAT_not_direct_mapped) && xen_initial_domain())
return 1;

The intention is to say that if XENFEAT_direct_mapped/XENFEAT_not_direct_mapped are not present, then use xen_initial_domain() like we did before.

So if xen_swiotlb_detect() returns true we know that Linux is either dom0 (xen_initial_domain() = true) or we have very good reasons to think we should initialize swiotlb-xen anyway (xen_feature(XENFEAT_direct_mapped) = true).

Quoted 2 lines

I think I'd keep it as-is for now, as my planned next step would be to fold xen-swiotlb into swiotlb entirely.

Thinking more about it we actually need to drop the xen_initial_domain() check otherwise some cases won't be functional (Dom0 not 1:1 mapped, or DomU 1:1 mapped).

#21 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 3 lines · Stefano Stabellini

Thinking more about it we actually need to drop the xen_initial_domain() check otherwise some cases won't be functional (Dom0 not 1:1 mapped, or DomU 1:1 mapped).

Hmm, but that would be the case even before this series, right?

#22 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 6 lines · Christoph Hellwig
Quoted 3 lines · Stefano Stabellini

Thinking more about it we actually need to drop the xen_initial_domain() check otherwise some cases won't be functional (Dom0 not 1:1 mapped, or DomU 1:1 mapped).

Hmm, but that would be the case even before this series, right?

Before this series we only have the xen_swiotlb_detect() check in xen_mm_init, we don't have a second xen_initial_domain() check.

The issue is that this series is adding one more xen_initial_domain() check in xen_mm_init.

#23 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 6 lines · Stefano Stabellini
Quoted 5 lines · Christoph Hellwig
Quoted 3 lines · Stefano Stabellini

Thinking more about it we actually need to drop the xen_initial_domain() check otherwise some cases won't be functional (Dom0 not 1:1 mapped, or DomU 1:1 mapped).

Quoted 7 lines
Quoted 1 line

Hmm, but that would be the case even before this series, right?

Before this series we only have the xen_swiotlb_detect() check in xen_mm_init, we don't have a second xen_initial_domain() check.

The issue is that this series is adding one more xen_initial_domain() check in xen_mm_init.

In current mainline xen_mm_init calls xen_swiotlb_init unconditionally. But xen_swiotlb_init then calls xen_swiotlb_fixup after allocating the memory, which in turn calls xen_create_contiguous_region. xen_create_contiguous_region fails with -EINVAL for the !xen_initial_domain() and thus caues xen_swiotlb_fixup and xen_swiotlb_init to unwind and return -EINVAL.

So as far as I can tell there is no change in behavior, but maybe I'm missing something subtle?

#24 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 22 lines · Christoph Hellwig
Quoted 6 lines · Stefano Stabellini
Quoted 5 lines · Christoph Hellwig
Quoted 3 lines · Stefano Stabellini

Thinking more about it we actually need to drop the xen_initial_domain() check otherwise some cases won't be functional (Dom0 not 1:1 mapped, or DomU 1:1 mapped).

Quoted 7 lines
Quoted 1 line

Hmm, but that would be the case even before this series, right?

Before this series we only have the xen_swiotlb_detect() check in xen_mm_init, we don't have a second xen_initial_domain() check.

The issue is that this series is adding one more xen_initial_domain() check in xen_mm_init.

In current mainline xen_mm_init calls xen_swiotlb_init unconditionally. But xen_swiotlb_init then calls xen_swiotlb_fixup after allocating the memory, which in turn calls xen_create_contiguous_region. xen_create_contiguous_region fails with -EINVAL for the !xen_initial_domain() and thus caues xen_swiotlb_fixup and xen_swiotlb_init to unwind and return -EINVAL.

Quoted 2 lines

So as far as I can tell there is no change in behavior, but maybe I'm missing something subtle?

You are right.

The xen_initial_domain() check in xen_create_contiguous_region() is wrong and we should get rid of it. It is a leftover from before the xen_swiotlb_detect rework.

We could either remove it or change it into another xen_swiotlb_detect() check.

Feel free to add the patch to your series or fold it with another patch or rework it as you prefer. Thanks for spotting this!

---

arm/xen: don't check for xen_initial_domain() in xen_create_contiguous_region

It used to be that Linux enabled swiotlb-xen when running a dom0 on ARM. Since f5079a9a2a31 "xen/arm: introduce XENFEAT_direct_mapped and XENFEAT_not_direct_mapped", Linux detects whether to enable or disable swiotlb-xen based on the new feature flags: XENFEAT_direct_mapped and XENFEAT_not_direct_mapped.

However, there is still a leftover xen_initial_domain() check in xen_create_contiguous_region. Remove the check as xen_create_contiguous_region is only called by swiotlb-xen during initialization. If xen_create_contiguous_region is called, we know Linux is running 1:1 mapped so there is no need for additional checks.

Also update the in-code comment.

arch/arm/xen/mm.c +1 -4
index a7e54a087b80..28c207060253 100644
@@ -122,10 +122,7 @@ int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
 				 unsigned int address_bits,
 				 dma_addr_t *dma_handle)
 {
-	if (!xen_initial_domain())
-		return -EINVAL;
-
-	/* we assume that dom0 is mapped 1:1 for now */
+	/* the domain is 1:1 mapped to use swiotlb-xen */
 	*dma_handle = pstart;
 	return 0;
 }
#25 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

In reply to Stefano Stabellini
Quoted 9 lines · Stefano Stabellini
Quoted 4 lines · Christoph Hellwig

Allow to pass a remap argument to the swiotlb initialization functions to handle the Xen/x86 remap case. ARM/ARM64 never did any remapping from xen_swiotlb_fixup, so we don't even need that quirk.

Aside from that the rest looks OK. Also, you can add my:

Not for me, I fail to boot with

[   52.202000] bnxt_en 0000:31:00.0: swiotlb buffer is full (sz: 256 bytes), total 0 (slots), used 0 (slots)

(this is iscsi root so I need the NIC).

I bisected it to "x86: remove the IOMMU table infrastructure" but haven't actually looked at the code yet.

-boris

#26 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 14 lines · Boris Ostrovsky
Quoted 9 lines · Stefano Stabellini
Quoted 4 lines · Christoph Hellwig

Allow to pass a remap argument to the swiotlb initialization functions to handle the Xen/x86 remap case.  ARM/ARM64 never did any remapping from xen_swiotlb_fixup, so we don't even need that quirk.

Aside from that the rest looks OK. Also, you can add my:

Quoted 9 lines

Not for me, I fail to boot with

[   52.202000] bnxt_en 0000:31:00.0: swiotlb buffer is full (sz: 256 bytes), total 0 (slots), used 0 (slots)

(this is iscsi root so I need the NIC).

I bisected it to "x86: remove the IOMMU table infrastructure" but haven't actually looked at the code yet.

Again, this is as dom0. Baremetal is fine.

-boris

#27 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

In reply to Boris Ostrovsky
Quoted 8 lines · Boris Ostrovsky

Not for me, I fail to boot with

[   52.202000] bnxt_en 0000:31:00.0: swiotlb buffer is full (sz: 256 bytes), total 0 (slots), used 0 (slots)

(this is iscsi root so I need the NIC).

I bisected it to "x86: remove the IOMMU table infrastructure" but haven't actually looked at the code yet.

Thanks. Looks like the sizing is going wrong. Just to confirm, this is dom0 on x86 and no special command line options?

#28 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 11 lines · Christoph Hellwig
Quoted 8 lines · Boris Ostrovsky

Not for me, I fail to boot with

[   52.202000] bnxt_en 0000:31:00.0: swiotlb buffer is full (sz: 256 bytes), total 0 (slots), used 0 (slots)

(this is iscsi root so I need the NIC).

I bisected it to "x86: remove the IOMMU table infrastructure" but haven't actually looked at the code yet.

Thanks. Looks like the sizing is going wrong. Just to confirm, this is dom0 on x86 and no special command line options?

Right.

module2 /boot/vmlinuz-5.17.0-rc6swiotlb placeholder root=UUIDÛef1262-8c8a-43db-8055-7d9bec7bece0 ro crashkernel=auto LANG=en_US.UTF-8 rd.luks=0 rd.lvm=0 rd.md=0 rd.dm=0 netroot=iscsi:169.254.0.2:::1:iqn.2015-02.oracle.boot:uefi iscsi_param=node.session.timeo.replacement_timeout`00 net.ifnames=1 nvme_core.shutdown_timeout ipmi_si.tryacpi=0 ipmi_si.trydmi=0 ipmi_si.trydefaults=0 libiscsi.debug_libiscsi_eh=1  panic nokaslr earlyprintk=xen console=hvc0 loglevel=8 4

#29 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

In reply to Boris Ostrovsky
Quoted 8 lines · Boris Ostrovsky

Not for me, I fail to boot with

[   52.202000] bnxt_en 0000:31:00.0: swiotlb buffer is full (sz: 256 bytes), total 0 (slots), used 0 (slots)

(this is iscsi root so I need the NIC).

I bisected it to "x86: remove the IOMMU table infrastructure" but haven't actually looked at the code yet.

That looks like the swiotlb buffer did not get initialized at all, but I can't really explain why.

Can you stick in a printk and see if xen_swiotlb_init_early gets called at all?

#30 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 8 lines · Christoph Hellwig
Quoted 7 lines · Boris Ostrovsky

Not for me, I fail to boot with

[   52.202000] bnxt_en 0000:31:00.0: swiotlb buffer is full (sz: 256 bytes), total 0 (slots), used 0 (slots)

(this is iscsi root so I need the NIC).

Quoted 6 lines
Quoted 1 line

I bisected it to "x86: remove the IOMMU table infrastructure" but haven't actually looked at the code yet.

That looks like the swiotlb buffer did not get initialized at all, but I can't really explain why.

Can you stick in a printk and see if xen_swiotlb_init_early gets called at all?

Actually, that's the only thing I did do so far and yes, it does get called.

-boris

#31 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 10 lines · Boris Ostrovsky
Quoted 6 lines
Quoted 1 line

I bisected it to "x86: remove the IOMMU table infrastructure" but haven't actually looked at the code yet.

That looks like the swiotlb buffer did not get initialized at all, but I can't really explain why.

Can you stick in a printk and see if xen_swiotlb_init_early gets called at all?

Actually, that's the only thing I did do so far and yes, it does get called.

So, specifically for "x86: remove the IOMMU table infrastructure" I think we need the one-liner below so that swiotlb_exit doesn't get called for the Xen case. But that should have been fixed up by the next patch already.

arch/x86/kernel/pci-dma.c +1 -1
index 2ac0ef9c2fb76..1173aa282ab27 100644
@@ -70,7 +70,7 @@ static void __init pci_xen_swiotlb_init(void)
 	if (!xen_initial_domain() && !x86_swiotlb_enable &&
 	    swiotlb_force != SWIOTLB_FORCE)
 		return;
-	x86_swiotlb_enable = false;
+	x86_swiotlb_enable = true;
 	xen_swiotlb = true;
 	xen_swiotlb_init_early();
 	dma_ops = &xen_swiotlb_dma_ops;
#32 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 9 lines · Christoph Hellwig
Quoted 8 lines · Boris Ostrovsky
Quoted 6 lines
Quoted 1 line

I bisected it to "x86: remove the IOMMU table infrastructure" but haven't actually looked at the code yet.

That looks like the swiotlb buffer did not get initialized at all, but I can't really explain why.

Can you stick in a printk and see if xen_swiotlb_init_early gets called at all?

Quoted 5 lines
Quoted 1 line

Actually, that's the only thing I did do so far and yes, it does get called.

So, specifically for "x86: remove the IOMMU table infrastructure" I think we need the one-liner below so that swiotlb_exit doesn't get called for the Xen case. But that should have been fixed up by the next patch already.

This indeed allows dom0 to boot. Not sure I see where in the next patch this would have been fixed?

(BTW, just noticed in iommu_setup() you set this variable to 1. Should be 'true')

-boris

#33 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 1 line · Boris Ostrovsky

This indeed allows dom0 to boot. Not sure I see where in the next patch this would have been fixed?

I thought it did, but it doesn't. In the meantime I've pushed out an updated branch with this folded in to:

git://git.infradead.org/users/hch/misc.git swiotlb-init-cleanup

Quoted 1 line

(BTW, just noticed in iommu_setup() you set this variable to 1. Should be 'true')

Thank, I'll fix this up.

#34 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

In reply to Christoph Hellwig
Quoted 3 lines · Christoph Hellwig

Allow to pass a remap argument to the swiotlb initialization functions to handle the Xen/x86 remap case. ARM/ARM64 never did any remapping from xen_swiotlb_fixup, so we don't even need that quirk.

Any chance this patch could be split? Lots of things are happening here and it's somewhat hard to review. (Patch 7 too BTW but I think I managed to get through it)

Quoted 5 lines
arch/x86/kernel/pci-dma.c +0 -0
index e0def4b1c3181..2f2c468acb955 100644

arch/x86/kernel/pci-dma.c static inline void __init pci_swiotlb_detect(void) · line 77

Quoted 9 lines
arch/x86/kernel/pci-dma.c +0 -2
@@ -71,15 +71,12 @@ static inline void __init pci_swiotlb_detect(void)
  #endif /* CONFIG_SWIOTLB */
 
  #ifdef CONFIG_SWIOTLB_XEN
-static bool xen_swiotlb;
-
  static void __init pci_xen_swiotlb_init(void)
  {
  	if (!xen_initial_domain() && !x86_swiotlb_enable)
  		return;

Now that there is a single call site for this routine I think this check can be dropped. We are only called here for xen_initial_domain()=true.

-boris

#35 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 8 lines · Boris Ostrovsky
Quoted 3 lines · Christoph Hellwig

Allow to pass a remap argument to the swiotlb initialization functions to handle the Xen/x86 remap case. ARM/ARM64 never did any remapping from xen_swiotlb_fixup, so we don't even need that quirk.

Any chance this patch could be split? Lots of things are happening here and it's somewhat hard to review. (Patch 7 too BTW but I think I managed to get through it)

What would be your preferred split?

Quoted 8 lines
Quoted 8 lines
arch/x86/kernel/pci-dma.c +0 -1
index e0def4b1c3181..2f2c468acb955 100644
@@ -71,15 +71,12 @@ static inline void __init pci_swiotlb_detect(void)
  #endif /* CONFIG_SWIOTLB */
    #ifdef CONFIG_SWIOTLB_XEN
-static bool xen_swiotlb;
Quoted 8 lines

arch/x86/kernel/pci-dma.c static inline void __init pci_swiotlb_detect(void) · line 77

Quoted 5 lines
arch/x86/kernel/pci-dma.c +0 -1
-
  static void __init pci_xen_swiotlb_init(void)
  {
  	if (!xen_initial_domain() && !x86_swiotlb_enable)
  		return;

Now that there is a single call site for this routine I think this check can be dropped. We are only called here for xen_initial_domain()=true.

The callsite just checks xen_pv_domain() and itself is called unconditionally during initialization.

#36 lore raw

Re: [PATCH 11/12] swiotlb: merge swiotlb-xen initialization into swiotlb

Quoted 8 lines · Christoph Hellwig
Quoted 6 lines · Boris Ostrovsky
Quoted 3 lines · Christoph Hellwig

Allow to pass a remap argument to the swiotlb initialization functions to handle the Xen/x86 remap case. ARM/ARM64 never did any remapping from xen_swiotlb_fixup, so we don't even need that quirk.

Any chance this patch could be split? Lots of things are happening here and it's somewhat hard to review. (Patch 7 too BTW but I think I managed to get through it)

What would be your preferred split?

swiotlb_init() rework to be done separately?

Quoted 16 lines
Quoted 16 lines
Quoted 6 lines
arch/x86/kernel/pci-dma.c +0 -0
index e0def4b1c3181..2f2c468acb955 100644

arch/x86/kernel/pci-dma.c static inline void __init pci_swiotlb_detect(void) · line 76

Quoted 8 lines
arch/x86/kernel/pci-dma.c +0 -2
@@ -71,15 +71,12 @@ static inline void __init pci_swiotlb_detect(void)
   #endif /* CONFIG_SWIOTLB */
     #ifdef CONFIG_SWIOTLB_XEN
-static bool xen_swiotlb;
-
   static void __init pci_xen_swiotlb_init(void)
   {
   	if (!xen_initial_domain() && !x86_swiotlb_enable)
   		return;

Now that there is a single call site for this routine I think this check can be dropped. We are only called here for xen_initial_domain()=true.

Quoted 2 lines

The callsite just checks xen_pv_domain() and itself is called unconditionally during initialization.

Oh, right, nevermind. *pv* domain.

-boris

#37 lore raw

PATCH 12/12 x86: remove cruft from <asm/dma-mapping.h>

In reply to Christoph Hellwig

<asm/dma-mapping.h> gets pulled in by all drivers using the DMA API. Remove x86 internal variables and unnecessary includes from it.

---
 arch/x86/include/asm/dma-mapping.h | 11 -----------
 arch/x86/include/asm/iommu.h       |  2 ++
 2 files changed, 2 insertions(+), 11 deletions(-)
arch/x86/include/asm/dma-mapping.h +0 -11
index 256fd8115223d..1c66708e30623 100644
@@ -2,17 +2,6 @@
 #ifndef _ASM_X86_DMA_MAPPING_H
 #define _ASM_X86_DMA_MAPPING_H
 
-/*
- * IOMMU interface. See Documentation/core-api/dma-api-howto.rst and
- * Documentation/core-api/dma-api.rst for documentation.
- */
-
-#include <linux/scatterlist.h>
-#include <asm/io.h>
-
-extern int iommu_merge;
-extern int panic_on_overflow;
-
 extern const struct dma_map_ops *dma_ops;
 
 static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
arch/x86/include/asm/iommu.h +2 -0
index dba89ed40d38d..0bef44d30a278 100644
@@ -8,6 +8,8 @@
 
 extern int force_iommu, no_iommu;
 extern int iommu_detected;
+extern int iommu_merge;
+extern int panic_on_overflow;
 
 #ifdef CONFIG_SWIOTLB
 extern bool x86_swiotlb_enable;
Signature · 1 line
2.30.2