Skip to contentSkip to conversation

swiotlb: Allow allocating buffer anywhere in memory

PATCH In review Thiago Jung Bauermann 2 participants 3 messages Aug 15, 2020 – Aug 17, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 swiotlb: Allow allocating buffer anywhere in memory

POWER secure guests (i.e., guests which use the Protection Execution Facility) need to use SWIOTLB to be able to do I/O with the hypervisor, but they don't need the SWIOTLB memory to be in low addresses since the hypervisor doesn't have any addressing limitation.

This solves a SWIOTLB initialization problem we are seeing in secure guests with 128 GB of RAM: they are configured with 4 GB of crashkernel reserved memory, which leaves no space for SWIOTLB in low addresses.

---
 arch/powerpc/mm/mem.c   |  7 ++++++-
 include/linux/swiotlb.h |  8 +++++++-
 kernel/dma/swiotlb.c    | 10 +++++++---
 3 files changed, 20 insertions(+), 5 deletions(-)

Normally I would split changes like this into one patch touching generic code and another for the arch-specific part, but in this case I thought it would be unneeded complexity. I can split though if people prefer it that way.

arch/powerpc/mm/mem.c +6 -1
index c2c11eb8dcfc..13f2e3aff8b5 100644
@@ -50,6 +50,7 @@
 #include <asm/swiotlb.h>
 #include <asm/rtas.h>
 #include <asm/kasan.h>
+#include <asm/svm.h>
 
 #include <mm/mmu_decl.h>
 
@@ -290,7 +291,11 @@ void __init mem_init(void)
 	 * back to to-down.
 	 */
 	memblock_set_bottom_up(true);
-	swiotlb_init(0);
+	/*
+	 * SVM guests can use the SWIOTLB wherever it is in memory,
+	 * even if not DMA-able.
+	 */
+	swiotlb_init_anywhere(0, is_secure_guest());
 #endif
 
 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
include/linux/swiotlb.h +7 -1
index 046bb94bd4d6..433f3dbb35b5 100644
@@ -29,7 +29,13 @@ enum swiotlb_force {
  */
 #define IO_TLB_SHIFT 11
 
-extern void swiotlb_init(int verbose);
+void __init swiotlb_init_anywhere(int verbose, bool allocate_anywhere);
+
+static inline void swiotlb_init(int verbose)
+{
+	swiotlb_init_anywhere(verbose, false);
+}
+
 int swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose);
 extern unsigned long swiotlb_nr_tbl(void);
 unsigned long swiotlb_size_or_default(void);
kernel/dma/swiotlb.c +7 -3
index c19379fabd20..27070aa59e34 100644
@@ -244,7 +244,7 @@ int __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose)
  * structures for the software IO TLB used to implement the DMA API.
  */
 void  __init
-swiotlb_init(int verbose)
+swiotlb_init_anywhere(int verbose, bool allocate_anywhere)
 {
 	size_t default_size = IO_TLB_DEFAULT_SIZE;
 	unsigned char *vstart;
@@ -257,8 +257,12 @@ swiotlb_init(int verbose)
 
 	bytes = io_tlb_nslabs << IO_TLB_SHIFT;
 
-	/* Get IO TLB memory from the low pages */
-	vstart = memblock_alloc_low(PAGE_ALIGN(bytes), PAGE_SIZE);
+	if (allocate_anywhere)
+		vstart = memblock_alloc(PAGE_ALIGN(bytes), PAGE_SIZE);
+	else
+		/* Get IO TLB memory from the low pages */
+		vstart = memblock_alloc_low(PAGE_ALIGN(bytes), PAGE_SIZE);
+
 	if (vstart && !swiotlb_init_with_tbl(vstart, io_tlb_nslabs, verbose))
 		return;
 

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu

#1 lore raw
Quoted 8 lines · Thiago Jung Bauermann

POWER secure guests (i.e., guests which use the Protection Execution Facility) need to use SWIOTLB to be able to do I/O with the hypervisor, but they don't need the SWIOTLB memory to be in low addresses since the hypervisor doesn't have any addressing limitation.

This solves a SWIOTLB initialization problem we are seeing in secure guests with 128 GB of RAM: they are configured with 4 GB of crashkernel reserved memory, which leaves no space for SWIOTLB in low addresses.

What about just open coding the allocation and using swiotlb_init_with_tbl?

Quoted 75 lines
---
 arch/powerpc/mm/mem.c   |  7 ++++++-
 include/linux/swiotlb.h |  8 +++++++-
 kernel/dma/swiotlb.c    | 10 +++++++---
 3 files changed, 20 insertions(+), 5 deletions(-)

Normally I would split changes like this into one patch touching generic code and another for the arch-specific part, but in this case I thought it would be unneeded complexity. I can split though if people prefer it that way.

arch/powerpc/mm/mem.c +6 -1
index c2c11eb8dcfc..13f2e3aff8b5 100644
@@ -50,6 +50,7 @@
 #include <asm/swiotlb.h>
 #include <asm/rtas.h>
 #include <asm/kasan.h>
+#include <asm/svm.h>
 
 #include <mm/mmu_decl.h>
 
@@ -290,7 +291,11 @@ void __init mem_init(void)
 	 * back to to-down.
 	 */
 	memblock_set_bottom_up(true);
-	swiotlb_init(0);
+	/*
+	 * SVM guests can use the SWIOTLB wherever it is in memory,
+	 * even if not DMA-able.
+	 */
+	swiotlb_init_anywhere(0, is_secure_guest());
 #endif
 
 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
include/linux/swiotlb.h +7 -1
index 046bb94bd4d6..433f3dbb35b5 100644
@@ -29,7 +29,13 @@ enum swiotlb_force {
  */
 #define IO_TLB_SHIFT 11
 
-extern void swiotlb_init(int verbose);
+void __init swiotlb_init_anywhere(int verbose, bool allocate_anywhere);
+
+static inline void swiotlb_init(int verbose)
+{
+	swiotlb_init_anywhere(verbose, false);
+}
+
 int swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose);
 extern unsigned long swiotlb_nr_tbl(void);
 unsigned long swiotlb_size_or_default(void);
kernel/dma/swiotlb.c +1 -2
index c19379fabd20..27070aa59e34 100644
@@ -244,7 +244,7 @@ int __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose)
  * structures for the software IO TLB used to implement the DMA API.
  */
 void  __init
-swiotlb_init(int verbose)
+swiotlb_init_anywhere(int verbose, bool allocate_anywhere)
 {
 	size_t default_size = IO_TLB_DEFAULT_SIZE;
 	unsigned char *vstart;
@@ -257,8 +257,12 @@ swiotlb_init(int verbose)
 
 	bytes = io_tlb_nslabs << IO_TLB_SHIFT;
 
-	/* Get IO TLB memory from the low pages */

kernel/dma/swiotlb.c swiotlb_init(int verbose) · line 268

Quoted 10 lines
kernel/dma/swiotlb.c +6 -1
-	vstart = memblock_alloc_low(PAGE_ALIGN(bytes), PAGE_SIZE);
+	if (allocate_anywhere)
+		vstart = memblock_alloc(PAGE_ALIGN(bytes), PAGE_SIZE);
+	else
+		/* Get IO TLB memory from the low pages */
+		vstart = memblock_alloc_low(PAGE_ALIGN(bytes), PAGE_SIZE);
+
 	if (vstart && !swiotlb_init_with_tbl(vstart, io_tlb_nslabs, verbose))
 		return;
 

---end quoted text---

#2 lore raw

Hello Christoph,

Quoted 6 lines · Christoph Hellwig
Quoted 5 lines · Thiago Jung Bauermann

POWER secure guests (i.e., guests which use the Protection Execution Facility) need to use SWIOTLB to be able to do I/O with the hypervisor, but they don't need the SWIOTLB memory to be in low addresses since the hypervisor doesn't have any addressing limitation.

Quoted 6 lines
Quoted 3 lines

This solves a SWIOTLB initialization problem we are seeing in secure guests with 128 GB of RAM: they are configured with 4 GB of crashkernel reserved memory, which leaves no space for SWIOTLB in low addresses.

What about just open coding the allocation and using swiotlb_init_with_tbl?

Yes, that works too. I just sent a v2 implementing that change. I just had to add a small accessor function so that I could set no_iotlb_memory from outside swiotlb.c.

Thank you for the quick review.

Signature · 6 lines
Thiago Jung Bauermann
IBM Linux Technology Center
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu