* [kvm-unit-tests PATCH v4 1/7] s390x: lib: add and use macros for control register bits
2021-05-26 13:42 [kvm-unit-tests PATCH v4 0/7] s390: Add support for large pages Claudio Imbrenda
@ 2021-05-26 13:42 ` Claudio Imbrenda
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 2/7] libcflat: add SZ_1M and SZ_2G Claudio Imbrenda
` (5 subsequent siblings)
6 siblings, 0 replies; 14+ messages in thread
From: Claudio Imbrenda @ 2021-05-26 13:42 UTC (permalink / raw)
To: kvm; +Cc: linux-s390, david, thuth, frankja, cohuck
Add CTL0_* and CTL2_* macros for specific control register bits.
Replace all hardcoded values in the library and in the existing testcases so
that they use the new macros.
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Reviewed-by: Janosch Frank <frankja@linux.ibm.com>
---
lib/s390x/asm/arch_def.h | 12 ++++++++++++
lib/s390x/asm/float.h | 4 ++--
lib/s390x/asm/interrupt.h | 4 ++--
lib/s390x/sclp.c | 4 ++--
s390x/diag288.c | 2 +-
s390x/gs.c | 2 +-
s390x/iep.c | 4 ++--
s390x/skrf.c | 2 +-
s390x/smp.c | 8 ++++----
s390x/vector.c | 2 +-
10 files changed, 28 insertions(+), 16 deletions(-)
diff --git a/lib/s390x/asm/arch_def.h b/lib/s390x/asm/arch_def.h
index 7e2c5e62..c3568ab9 100644
--- a/lib/s390x/asm/arch_def.h
+++ b/lib/s390x/asm/arch_def.h
@@ -229,6 +229,18 @@ static inline uint64_t stctg(int cr)
return value;
}
+#define CTL0_LOW_ADDR_PROT (63 - 35)
+#define CTL0_EDAT (63 - 40)
+#define CTL0_IEP (63 - 43)
+#define CTL0_AFP (63 - 45)
+#define CTL0_VECTOR (63 - 46)
+#define CTL0_EMERGENCY_SIGNAL (63 - 49)
+#define CTL0_EXTERNAL_CALL (63 - 50)
+#define CTL0_CLOCK_COMPARATOR (63 - 52)
+#define CTL0_SERVICE_SIGNAL (63 - 54)
+
+#define CTL2_GUARDED_STORAGE (63 - 59)
+
static inline void ctl_set_bit(int cr, unsigned int bit)
{
uint64_t reg;
diff --git a/lib/s390x/asm/float.h b/lib/s390x/asm/float.h
index 13679447..98829918 100644
--- a/lib/s390x/asm/float.h
+++ b/lib/s390x/asm/float.h
@@ -38,12 +38,12 @@ static inline void set_fpc_dxc(uint8_t dxc)
static inline void afp_enable(void)
{
- ctl_set_bit(0, 63 - 45);
+ ctl_set_bit(0, CTL0_AFP);
}
static inline void afp_disable(void)
{
- ctl_clear_bit(0, 63 - 45);
+ ctl_clear_bit(0, CTL0_AFP);
}
#endif
diff --git a/lib/s390x/asm/interrupt.h b/lib/s390x/asm/interrupt.h
index 31e4766d..bf0eb40d 100644
--- a/lib/s390x/asm/interrupt.h
+++ b/lib/s390x/asm/interrupt.h
@@ -27,13 +27,13 @@ void check_pgm_int_code(uint16_t code);
/* Activate low-address protection */
static inline void low_prot_enable(void)
{
- ctl_set_bit(0, 63 - 35);
+ ctl_set_bit(0, CTL0_LOW_ADDR_PROT);
}
/* Disable low-address protection */
static inline void low_prot_disable(void)
{
- ctl_clear_bit(0, 63 - 35);
+ ctl_clear_bit(0, CTL0_LOW_ADDR_PROT);
}
#endif
diff --git a/lib/s390x/sclp.c b/lib/s390x/sclp.c
index 7a9b2c52..0d0b3d6a 100644
--- a/lib/s390x/sclp.c
+++ b/lib/s390x/sclp.c
@@ -50,7 +50,7 @@ void sclp_setup_int(void)
{
uint64_t mask;
- ctl_set_bit(0, 9);
+ ctl_set_bit(0, CTL0_SERVICE_SIGNAL);
mask = extract_psw_mask();
mask |= PSW_MASK_EXT;
@@ -59,7 +59,7 @@ void sclp_setup_int(void)
void sclp_handle_ext(void)
{
- ctl_clear_bit(0, 9);
+ ctl_clear_bit(0, CTL0_SERVICE_SIGNAL);
spin_lock(&sclp_lock);
sclp_busy = false;
spin_unlock(&sclp_lock);
diff --git a/s390x/diag288.c b/s390x/diag288.c
index e132ff04..82b6ec17 100644
--- a/s390x/diag288.c
+++ b/s390x/diag288.c
@@ -86,7 +86,7 @@ static void test_bite(void)
asm volatile("stck %0" : "=Q" (time) : : "cc");
time += (uint64_t)(16000 * 1000) << 12;
asm volatile("sckc %0" : : "Q" (time));
- ctl_set_bit(0, 11);
+ ctl_set_bit(0, CTL0_CLOCK_COMPARATOR);
mask = extract_psw_mask();
mask |= PSW_MASK_EXT;
load_psw_mask(mask);
diff --git a/s390x/gs.c b/s390x/gs.c
index 1376d0e6..a017a97d 100644
--- a/s390x/gs.c
+++ b/s390x/gs.c
@@ -145,7 +145,7 @@ static void test_special(void)
static void init(void)
{
/* Enable control bit for gs */
- ctl_set_bit(2, 4);
+ ctl_set_bit(2, CTL2_GUARDED_STORAGE);
/* Setup gs registers to guard the gs_area */
gs_cb.gsd = gs_area | 25;
diff --git a/s390x/iep.c b/s390x/iep.c
index fe167ef0..906c77b3 100644
--- a/s390x/iep.c
+++ b/s390x/iep.c
@@ -22,7 +22,7 @@ static void test_iep(void)
void (*fn)(void);
/* Enable IEP */
- ctl_set_bit(0, 20);
+ ctl_set_bit(0, CTL0_IEP);
/* Get and protect a page with the IEP bit */
iepbuf = alloc_page();
@@ -40,7 +40,7 @@ static void test_iep(void)
check_pgm_int_code(PGM_INT_CODE_PROTECTION);
report_prefix_pop();
unprotect_page(iepbuf, PAGE_ENTRY_IEP);
- ctl_clear_bit(0, 20);
+ ctl_clear_bit(0, CTL0_IEP);
free_page(iepbuf);
}
diff --git a/s390x/skrf.c b/s390x/skrf.c
index 57524ba8..94e906a6 100644
--- a/s390x/skrf.c
+++ b/s390x/skrf.c
@@ -150,7 +150,7 @@ static void ecall_setup(void)
/* Put a skey into the ext new psw */
lc->ext_new_psw.mask = 0x00F0000180000000UL;
/* Open up ext masks */
- ctl_set_bit(0, 13);
+ ctl_set_bit(0, CTL0_EXTERNAL_CALL);
mask = extract_psw_mask();
mask |= PSW_MASK_EXT;
load_psw_mask(mask);
diff --git a/s390x/smp.c b/s390x/smp.c
index b0ece491..f25ec769 100644
--- a/s390x/smp.c
+++ b/s390x/smp.c
@@ -154,7 +154,7 @@ static void ecall(void)
struct lowcore *lc = (void *)0x0;
expect_ext_int();
- ctl_set_bit(0, 13);
+ ctl_set_bit(0, CTL0_EXTERNAL_CALL);
mask = extract_psw_mask();
mask |= PSW_MASK_EXT;
load_psw_mask(mask);
@@ -188,7 +188,7 @@ static void emcall(void)
struct lowcore *lc = (void *)0x0;
expect_ext_int();
- ctl_set_bit(0, 14);
+ ctl_set_bit(0, CTL0_EMERGENCY_SIGNAL);
mask = extract_psw_mask();
mask |= PSW_MASK_EXT;
load_psw_mask(mask);
@@ -283,8 +283,8 @@ static void test_local_ints(void)
unsigned long mask;
/* Open masks for ecall and emcall */
- ctl_set_bit(0, 13);
- ctl_set_bit(0, 14);
+ ctl_set_bit(0, CTL0_EXTERNAL_CALL);
+ ctl_set_bit(0, CTL0_EMERGENCY_SIGNAL);
mask = extract_psw_mask();
mask |= PSW_MASK_EXT;
load_psw_mask(mask);
diff --git a/s390x/vector.c b/s390x/vector.c
index b052de55..fdb0eee2 100644
--- a/s390x/vector.c
+++ b/s390x/vector.c
@@ -106,7 +106,7 @@ static void test_bcd_add(void)
static void init(void)
{
/* Enable vector instructions */
- ctl_set_bit(0, 17);
+ ctl_set_bit(0, CTL0_VECTOR);
/* Preset vector registers to 0xff */
memset(pagebuf, 0xff, PAGE_SIZE);
--
2.31.1
^ permalink raw reply related [flat|nested] 14+ messages in thread* [kvm-unit-tests PATCH v4 2/7] libcflat: add SZ_1M and SZ_2G
2021-05-26 13:42 [kvm-unit-tests PATCH v4 0/7] s390: Add support for large pages Claudio Imbrenda
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 1/7] s390x: lib: add and use macros for control register bits Claudio Imbrenda
@ 2021-05-26 13:42 ` Claudio Imbrenda
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 3/7] s390x: lib: fix pgtable.h Claudio Imbrenda
` (4 subsequent siblings)
6 siblings, 0 replies; 14+ messages in thread
From: Claudio Imbrenda @ 2021-05-26 13:42 UTC (permalink / raw)
To: kvm; +Cc: linux-s390, david, thuth, frankja, cohuck
Add SZ_1M and SZ_2G to libcflat.h
s390x needs those for large/huge pages
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
---
lib/libcflat.h | 2 ++
1 file changed, 2 insertions(+)
diff --git a/lib/libcflat.h b/lib/libcflat.h
index 460a1234..8dac0621 100644
--- a/lib/libcflat.h
+++ b/lib/libcflat.h
@@ -157,7 +157,9 @@ extern void setup_vm(void);
#define SZ_8K (1 << 13)
#define SZ_16K (1 << 14)
#define SZ_64K (1 << 16)
+#define SZ_1M (1 << 20)
#define SZ_2M (1 << 21)
#define SZ_1G (1 << 30)
+#define SZ_2G (1ul << 31)
#endif
--
2.31.1
^ permalink raw reply related [flat|nested] 14+ messages in thread* [kvm-unit-tests PATCH v4 3/7] s390x: lib: fix pgtable.h
2021-05-26 13:42 [kvm-unit-tests PATCH v4 0/7] s390: Add support for large pages Claudio Imbrenda
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 1/7] s390x: lib: add and use macros for control register bits Claudio Imbrenda
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 2/7] libcflat: add SZ_1M and SZ_2G Claudio Imbrenda
@ 2021-05-26 13:42 ` Claudio Imbrenda
2021-05-27 14:47 ` Janosch Frank
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 4/7] s390x: lib: Add idte and other huge pages functions/macros Claudio Imbrenda
` (3 subsequent siblings)
6 siblings, 1 reply; 14+ messages in thread
From: Claudio Imbrenda @ 2021-05-26 13:42 UTC (permalink / raw)
To: kvm; +Cc: linux-s390, david, thuth, frankja, cohuck
Fix pgtable.h:
* SEGMENT_ENTRY_SFAA had one extra bit set
* pmd entries don't have a length field
* ipte does not need to clear the lower bits
- clearing the 12 lower bits is technically incorrect, as page tables are
architecturally aligned to 11 bit addresses (even though the unit tests
allocate always one full page)
* region table entries should use REGION_ENTRY_TL instead of *_TABLE_LENGTH
- *_TABLE_LENGTH need to stay, because they should be used for ASCEs
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
---
lib/s390x/asm/pgtable.h | 13 ++++++-------
1 file changed, 6 insertions(+), 7 deletions(-)
diff --git a/lib/s390x/asm/pgtable.h b/lib/s390x/asm/pgtable.h
index 277f3480..1a21f175 100644
--- a/lib/s390x/asm/pgtable.h
+++ b/lib/s390x/asm/pgtable.h
@@ -60,7 +60,7 @@
#define SEGMENT_SHIFT 20
#define SEGMENT_ENTRY_ORIGIN 0xfffffffffffff800UL
-#define SEGMENT_ENTRY_SFAA 0xfffffffffff80000UL
+#define SEGMENT_ENTRY_SFAA 0xfffffffffff00000UL
#define SEGMENT_ENTRY_AV 0x0000000000010000UL
#define SEGMENT_ENTRY_ACC 0x000000000000f000UL
#define SEGMENT_ENTRY_F 0x0000000000000800UL
@@ -143,7 +143,7 @@ static inline p4d_t *p4d_alloc(pgd_t *pgd, unsigned long addr)
if (pgd_none(*pgd)) {
p4d_t *p4d = p4d_alloc_one();
pgd_val(*pgd) = __pa(p4d) | REGION_ENTRY_TT_REGION1 |
- REGION_TABLE_LENGTH;
+ REGION_ENTRY_TL;
}
return p4d_offset(pgd, addr);
}
@@ -163,7 +163,7 @@ static inline pud_t *pud_alloc(p4d_t *p4d, unsigned long addr)
if (p4d_none(*p4d)) {
pud_t *pud = pud_alloc_one();
p4d_val(*p4d) = __pa(pud) | REGION_ENTRY_TT_REGION2 |
- REGION_TABLE_LENGTH;
+ REGION_ENTRY_TL;
}
return pud_offset(p4d, addr);
}
@@ -183,7 +183,7 @@ static inline pmd_t *pmd_alloc(pud_t *pud, unsigned long addr)
if (pud_none(*pud)) {
pmd_t *pmd = pmd_alloc_one();
pud_val(*pud) = __pa(pmd) | REGION_ENTRY_TT_REGION3 |
- REGION_TABLE_LENGTH;
+ REGION_ENTRY_TL;
}
return pmd_offset(pud, addr);
}
@@ -202,15 +202,14 @@ static inline pte_t *pte_alloc(pmd_t *pmd, unsigned long addr)
{
if (pmd_none(*pmd)) {
pte_t *pte = pte_alloc_one();
- pmd_val(*pmd) = __pa(pte) | SEGMENT_ENTRY_TT_SEGMENT |
- SEGMENT_TABLE_LENGTH;
+ pmd_val(*pmd) = __pa(pte) | SEGMENT_ENTRY_TT_SEGMENT;
}
return pte_offset(pmd, addr);
}
static inline void ipte(unsigned long vaddr, pteval_t *p_pte)
{
- unsigned long table_origin = (unsigned long)p_pte & PAGE_MASK;
+ unsigned long table_origin = (unsigned long)p_pte;
asm volatile(
" ipte %0,%1\n"
--
2.31.1
^ permalink raw reply related [flat|nested] 14+ messages in thread* Re: [kvm-unit-tests PATCH v4 3/7] s390x: lib: fix pgtable.h
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 3/7] s390x: lib: fix pgtable.h Claudio Imbrenda
@ 2021-05-27 14:47 ` Janosch Frank
0 siblings, 0 replies; 14+ messages in thread
From: Janosch Frank @ 2021-05-27 14:47 UTC (permalink / raw)
To: Claudio Imbrenda, kvm; +Cc: linux-s390, david, thuth, cohuck
On 5/26/21 3:42 PM, Claudio Imbrenda wrote:
> Fix pgtable.h:
>
> * SEGMENT_ENTRY_SFAA had one extra bit set
> * pmd entries don't have a length field
> * ipte does not need to clear the lower bits
> - clearing the 12 lower bits is technically incorrect, as page tables are
> architecturally aligned to 11 bit addresses (even though the unit tests
> allocate always one full page)
> * region table entries should use REGION_ENTRY_TL instead of *_TABLE_LENGTH
> - *_TABLE_LENGTH need to stay, because they should be used for ASCEs
>
> Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Reviewed-by: Janosch Frank <frankja@linux.ibm.com>
> ---
> lib/s390x/asm/pgtable.h | 13 ++++++-------
> 1 file changed, 6 insertions(+), 7 deletions(-)
>
> diff --git a/lib/s390x/asm/pgtable.h b/lib/s390x/asm/pgtable.h
> index 277f3480..1a21f175 100644
> --- a/lib/s390x/asm/pgtable.h
> +++ b/lib/s390x/asm/pgtable.h
> @@ -60,7 +60,7 @@
> #define SEGMENT_SHIFT 20
>
> #define SEGMENT_ENTRY_ORIGIN 0xfffffffffffff800UL
> -#define SEGMENT_ENTRY_SFAA 0xfffffffffff80000UL
> +#define SEGMENT_ENTRY_SFAA 0xfffffffffff00000UL
> #define SEGMENT_ENTRY_AV 0x0000000000010000UL
> #define SEGMENT_ENTRY_ACC 0x000000000000f000UL
> #define SEGMENT_ENTRY_F 0x0000000000000800UL
> @@ -143,7 +143,7 @@ static inline p4d_t *p4d_alloc(pgd_t *pgd, unsigned long addr)
> if (pgd_none(*pgd)) {
> p4d_t *p4d = p4d_alloc_one();
> pgd_val(*pgd) = __pa(p4d) | REGION_ENTRY_TT_REGION1 |
> - REGION_TABLE_LENGTH;
> + REGION_ENTRY_TL;
> }
> return p4d_offset(pgd, addr);
> }
> @@ -163,7 +163,7 @@ static inline pud_t *pud_alloc(p4d_t *p4d, unsigned long addr)
> if (p4d_none(*p4d)) {
> pud_t *pud = pud_alloc_one();
> p4d_val(*p4d) = __pa(pud) | REGION_ENTRY_TT_REGION2 |
> - REGION_TABLE_LENGTH;
> + REGION_ENTRY_TL;
> }
> return pud_offset(p4d, addr);
> }
> @@ -183,7 +183,7 @@ static inline pmd_t *pmd_alloc(pud_t *pud, unsigned long addr)
> if (pud_none(*pud)) {
> pmd_t *pmd = pmd_alloc_one();
> pud_val(*pud) = __pa(pmd) | REGION_ENTRY_TT_REGION3 |
> - REGION_TABLE_LENGTH;
> + REGION_ENTRY_TL;
> }
> return pmd_offset(pud, addr);
> }
> @@ -202,15 +202,14 @@ static inline pte_t *pte_alloc(pmd_t *pmd, unsigned long addr)
> {
> if (pmd_none(*pmd)) {
> pte_t *pte = pte_alloc_one();
> - pmd_val(*pmd) = __pa(pte) | SEGMENT_ENTRY_TT_SEGMENT |
> - SEGMENT_TABLE_LENGTH;
> + pmd_val(*pmd) = __pa(pte) | SEGMENT_ENTRY_TT_SEGMENT;
> }
> return pte_offset(pmd, addr);
> }
>
> static inline void ipte(unsigned long vaddr, pteval_t *p_pte)
> {
> - unsigned long table_origin = (unsigned long)p_pte & PAGE_MASK;
> + unsigned long table_origin = (unsigned long)p_pte;
>
> asm volatile(
> " ipte %0,%1\n"
>
^ permalink raw reply [flat|nested] 14+ messages in thread
* [kvm-unit-tests PATCH v4 4/7] s390x: lib: Add idte and other huge pages functions/macros
2021-05-26 13:42 [kvm-unit-tests PATCH v4 0/7] s390: Add support for large pages Claudio Imbrenda
` (2 preceding siblings ...)
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 3/7] s390x: lib: fix pgtable.h Claudio Imbrenda
@ 2021-05-26 13:42 ` Claudio Imbrenda
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 5/7] s390x: lib: add teid union and clear teid from lowcore Claudio Imbrenda
` (2 subsequent siblings)
6 siblings, 0 replies; 14+ messages in thread
From: Claudio Imbrenda @ 2021-05-26 13:42 UTC (permalink / raw)
To: kvm; +Cc: linux-s390, david, thuth, frankja, cohuck
Improve pgtable.h:
* add macros to check whether a pmd or a pud are large / huge
* add idte functions for pmd, pud, p4d and pgd
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Acked-by: Janosch Frank <frankja@linux.ibm.com>
---
lib/s390x/asm/pgtable.h | 31 +++++++++++++++++++++++++++++++
1 file changed, 31 insertions(+)
diff --git a/lib/s390x/asm/pgtable.h b/lib/s390x/asm/pgtable.h
index 1a21f175..f166dcc6 100644
--- a/lib/s390x/asm/pgtable.h
+++ b/lib/s390x/asm/pgtable.h
@@ -100,6 +100,9 @@
#define pmd_none(entry) (pmd_val(entry) & SEGMENT_ENTRY_I)
#define pte_none(entry) (pte_val(entry) & PAGE_ENTRY_I)
+#define pud_huge(entry) (pud_val(entry) & REGION3_ENTRY_FC)
+#define pmd_large(entry) (pmd_val(entry) & SEGMENT_ENTRY_FC)
+
#define pgd_addr(entry) __va(pgd_val(entry) & REGION_ENTRY_ORIGIN)
#define p4d_addr(entry) __va(p4d_val(entry) & REGION_ENTRY_ORIGIN)
#define pud_addr(entry) __va(pud_val(entry) & REGION_ENTRY_ORIGIN)
@@ -216,6 +219,34 @@ static inline void ipte(unsigned long vaddr, pteval_t *p_pte)
: : "a" (table_origin), "a" (vaddr) : "memory");
}
+static inline void idte(unsigned long table_origin, unsigned long vaddr)
+{
+ vaddr &= SEGMENT_ENTRY_SFAA;
+ asm volatile(
+ " idte %0,0,%1\n"
+ : : "a" (table_origin), "a" (vaddr) : "memory");
+}
+
+static inline void idte_pmdp(unsigned long vaddr, pmdval_t *pmdp)
+{
+ idte((unsigned long)(pmdp - pmd_index(vaddr)) | ASCE_DT_SEGMENT, vaddr);
+}
+
+static inline void idte_pudp(unsigned long vaddr, pudval_t *pudp)
+{
+ idte((unsigned long)(pudp - pud_index(vaddr)) | ASCE_DT_REGION3, vaddr);
+}
+
+static inline void idte_p4dp(unsigned long vaddr, p4dval_t *p4dp)
+{
+ idte((unsigned long)(p4dp - p4d_index(vaddr)) | ASCE_DT_REGION2, vaddr);
+}
+
+static inline void idte_pgdp(unsigned long vaddr, pgdval_t *pgdp)
+{
+ idte((unsigned long)(pgdp - pgd_index(vaddr)) | ASCE_DT_REGION1, vaddr);
+}
+
void configure_dat(int enable);
#endif /* _ASMS390X_PGTABLE_H_ */
--
2.31.1
^ permalink raw reply related [flat|nested] 14+ messages in thread* [kvm-unit-tests PATCH v4 5/7] s390x: lib: add teid union and clear teid from lowcore
2021-05-26 13:42 [kvm-unit-tests PATCH v4 0/7] s390: Add support for large pages Claudio Imbrenda
` (3 preceding siblings ...)
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 4/7] s390x: lib: Add idte and other huge pages functions/macros Claudio Imbrenda
@ 2021-05-26 13:42 ` Claudio Imbrenda
2021-05-27 14:52 ` Janosch Frank
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 6/7] s390x: mmu: add support for large pages Claudio Imbrenda
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 7/7] s390x: edat test Claudio Imbrenda
6 siblings, 1 reply; 14+ messages in thread
From: Claudio Imbrenda @ 2021-05-26 13:42 UTC (permalink / raw)
To: kvm; +Cc: linux-s390, david, thuth, frankja, cohuck
Add a union to represent Translation-Exception Identification (TEID).
Clear the TEID in expect_pgm_int clear_pgm_int.
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
---
lib/s390x/asm/interrupt.h | 22 ++++++++++++++++++++++
lib/s390x/interrupt.c | 2 ++
2 files changed, 24 insertions(+)
diff --git a/lib/s390x/asm/interrupt.h b/lib/s390x/asm/interrupt.h
index bf0eb40d..b40def65 100644
--- a/lib/s390x/asm/interrupt.h
+++ b/lib/s390x/asm/interrupt.h
@@ -13,6 +13,28 @@
#define EXT_IRQ_EXTERNAL_CALL 0x1202
#define EXT_IRQ_SERVICE_SIG 0x2401
+#define TEID_ASCE_PRIMARY 0
+#define TEID_ASCE_AR 1
+#define TEID_ASCE_SECONDARY 2
+#define TEID_ASCE_HOME 3
+
+union teid {
+ unsigned long val;
+ struct {
+ unsigned long addr:52;
+ unsigned long fetch:1;
+ unsigned long store:1;
+ unsigned long reserved:6;
+ unsigned long acc_list_prot:1;
+ /* depending on the exception and the installed facilities,
+ * the m field can indicate severel different things,
+ * including whether the exception was triggered by a MVPG
+ * instruction, or whether the addr field is meaningful */
+ unsigned long m:1;
+ unsigned long asce_id:2;
+ };
+};
+
void register_pgm_cleanup_func(void (*f)(void));
void handle_pgm_int(struct stack_frame_int *stack);
void handle_ext_int(struct stack_frame_int *stack);
diff --git a/lib/s390x/interrupt.c b/lib/s390x/interrupt.c
index ce0003de..b627942f 100644
--- a/lib/s390x/interrupt.c
+++ b/lib/s390x/interrupt.c
@@ -22,6 +22,7 @@ void expect_pgm_int(void)
{
pgm_int_expected = true;
lc->pgm_int_code = 0;
+ lc->trans_exc_id = 0;
mb();
}
@@ -39,6 +40,7 @@ uint16_t clear_pgm_int(void)
mb();
code = lc->pgm_int_code;
lc->pgm_int_code = 0;
+ lc->trans_exc_id = 0;
pgm_int_expected = false;
return code;
}
--
2.31.1
^ permalink raw reply related [flat|nested] 14+ messages in thread* Re: [kvm-unit-tests PATCH v4 5/7] s390x: lib: add teid union and clear teid from lowcore
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 5/7] s390x: lib: add teid union and clear teid from lowcore Claudio Imbrenda
@ 2021-05-27 14:52 ` Janosch Frank
2021-05-27 15:05 ` Claudio Imbrenda
0 siblings, 1 reply; 14+ messages in thread
From: Janosch Frank @ 2021-05-27 14:52 UTC (permalink / raw)
To: Claudio Imbrenda, kvm; +Cc: linux-s390, david, thuth, cohuck
On 5/26/21 3:42 PM, Claudio Imbrenda wrote:
> Add a union to represent Translation-Exception Identification (TEID).
>
> Clear the TEID in expect_pgm_int clear_pgm_int.
>
> Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Reviewed-by: Janosch Frank <frankja@linux.ibm.com>
> ---
> lib/s390x/asm/interrupt.h | 22 ++++++++++++++++++++++
> lib/s390x/interrupt.c | 2 ++
> 2 files changed, 24 insertions(+)
>
> diff --git a/lib/s390x/asm/interrupt.h b/lib/s390x/asm/interrupt.h
> index bf0eb40d..b40def65 100644
> --- a/lib/s390x/asm/interrupt.h
> +++ b/lib/s390x/asm/interrupt.h
> @@ -13,6 +13,28 @@
> #define EXT_IRQ_EXTERNAL_CALL 0x1202
> #define EXT_IRQ_SERVICE_SIG 0x2401
>
> +#define TEID_ASCE_PRIMARY 0
> +#define TEID_ASCE_AR 1
> +#define TEID_ASCE_SECONDARY 2
> +#define TEID_ASCE_HOME 3
> +
> +union teid {
> + unsigned long val;
> + struct {
> + unsigned long addr:52;
> + unsigned long fetch:1;
> + unsigned long store:1;
> + unsigned long reserved:6;
> + unsigned long acc_list_prot:1;
> + /* depending on the exception and the installed facilities,
> + * the m field can indicate severel different things,
several
> + * including whether the exception was triggered by a MVPG
> + * instruction, or whether the addr field is meaningful */
Could you please convert the comment style to this?
/*
* Text
*/
> + unsigned long m:1;
> + unsigned long asce_id:2;
> + };
> +};
> +
> void register_pgm_cleanup_func(void (*f)(void));
> void handle_pgm_int(struct stack_frame_int *stack);
> void handle_ext_int(struct stack_frame_int *stack);
> diff --git a/lib/s390x/interrupt.c b/lib/s390x/interrupt.c
> index ce0003de..b627942f 100644
> --- a/lib/s390x/interrupt.c
> +++ b/lib/s390x/interrupt.c
> @@ -22,6 +22,7 @@ void expect_pgm_int(void)
> {
> pgm_int_expected = true;
> lc->pgm_int_code = 0;
> + lc->trans_exc_id = 0;
> mb();
> }
>
> @@ -39,6 +40,7 @@ uint16_t clear_pgm_int(void)
> mb();
> code = lc->pgm_int_code;
> lc->pgm_int_code = 0;
> + lc->trans_exc_id = 0;
> pgm_int_expected = false;
> return code;
> }
>
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [kvm-unit-tests PATCH v4 5/7] s390x: lib: add teid union and clear teid from lowcore
2021-05-27 14:52 ` Janosch Frank
@ 2021-05-27 15:05 ` Claudio Imbrenda
0 siblings, 0 replies; 14+ messages in thread
From: Claudio Imbrenda @ 2021-05-27 15:05 UTC (permalink / raw)
To: Janosch Frank; +Cc: kvm, linux-s390, david, thuth, cohuck
On Thu, 27 May 2021 16:52:48 +0200
Janosch Frank <frankja@linux.ibm.com> wrote:
> On 5/26/21 3:42 PM, Claudio Imbrenda wrote:
> > Add a union to represent Translation-Exception Identification
> > (TEID).
> >
> > Clear the TEID in expect_pgm_int clear_pgm_int.
> >
> > Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
>
> Reviewed-by: Janosch Frank <frankja@linux.ibm.com>
>
> > ---
> > lib/s390x/asm/interrupt.h | 22 ++++++++++++++++++++++
> > lib/s390x/interrupt.c | 2 ++
> > 2 files changed, 24 insertions(+)
> >
> > diff --git a/lib/s390x/asm/interrupt.h b/lib/s390x/asm/interrupt.h
> > index bf0eb40d..b40def65 100644
> > --- a/lib/s390x/asm/interrupt.h
> > +++ b/lib/s390x/asm/interrupt.h
> > @@ -13,6 +13,28 @@
> > #define EXT_IRQ_EXTERNAL_CALL 0x1202
> > #define EXT_IRQ_SERVICE_SIG 0x2401
> >
> > +#define TEID_ASCE_PRIMARY 0
> > +#define TEID_ASCE_AR 1
> > +#define TEID_ASCE_SECONDARY 2
> > +#define TEID_ASCE_HOME 3
> > +
> > +union teid {
> > + unsigned long val;
> > + struct {
> > + unsigned long addr:52;
> > + unsigned long fetch:1;
> > + unsigned long store:1;
> > + unsigned long reserved:6;
> > + unsigned long acc_list_prot:1;
> > + /* depending on the exception and the installed
> > facilities,
> > + * the m field can indicate severel different
> > things,
>
> several
fixed
> > + * including whether the exception was triggered
> > by a MVPG
> > + * instruction, or whether the addr field is
> > meaningful */
>
> Could you please convert the comment style to this?
>
> /*
> * Text
> */
and fixed
> > + unsigned long m:1;
> > + unsigned long asce_id:2;
> > + };
> > +};
> > +
> > void register_pgm_cleanup_func(void (*f)(void));
> > void handle_pgm_int(struct stack_frame_int *stack);
> > void handle_ext_int(struct stack_frame_int *stack);
> > diff --git a/lib/s390x/interrupt.c b/lib/s390x/interrupt.c
> > index ce0003de..b627942f 100644
> > --- a/lib/s390x/interrupt.c
> > +++ b/lib/s390x/interrupt.c
> > @@ -22,6 +22,7 @@ void expect_pgm_int(void)
> > {
> > pgm_int_expected = true;
> > lc->pgm_int_code = 0;
> > + lc->trans_exc_id = 0;
> > mb();
> > }
> >
> > @@ -39,6 +40,7 @@ uint16_t clear_pgm_int(void)
> > mb();
> > code = lc->pgm_int_code;
> > lc->pgm_int_code = 0;
> > + lc->trans_exc_id = 0;
> > pgm_int_expected = false;
> > return code;
> > }
> >
>
^ permalink raw reply [flat|nested] 14+ messages in thread
* [kvm-unit-tests PATCH v4 6/7] s390x: mmu: add support for large pages
2021-05-26 13:42 [kvm-unit-tests PATCH v4 0/7] s390: Add support for large pages Claudio Imbrenda
` (4 preceding siblings ...)
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 5/7] s390x: lib: add teid union and clear teid from lowcore Claudio Imbrenda
@ 2021-05-26 13:42 ` Claudio Imbrenda
2021-05-28 8:44 ` Janosch Frank
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 7/7] s390x: edat test Claudio Imbrenda
6 siblings, 1 reply; 14+ messages in thread
From: Claudio Imbrenda @ 2021-05-26 13:42 UTC (permalink / raw)
To: kvm; +Cc: linux-s390, david, thuth, frankja, cohuck
Add support for 1M and 2G pages.
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
---
lib/s390x/mmu.h | 73 +++++++++++++-
lib/s390x/mmu.c | 260 +++++++++++++++++++++++++++++++++++++++++++-----
2 files changed, 307 insertions(+), 26 deletions(-)
diff --git a/lib/s390x/mmu.h b/lib/s390x/mmu.h
index 603f289e..93208467 100644
--- a/lib/s390x/mmu.h
+++ b/lib/s390x/mmu.h
@@ -10,9 +10,78 @@
#ifndef _ASMS390X_MMU_H_
#define _ASMS390X_MMU_H_
-void protect_page(void *vaddr, unsigned long prot);
+/*
+ * Splits the pagetables down to the given DAT tables level.
+ * Returns a pointer to the DAT table entry of the given level.
+ * @pgtable root of the page table tree
+ * @vaddr address whose page tables are to split
+ * @level 3 (for 2GB pud), 4 (for 1 MB pmd) or 5 (for 4KB pages)
+ */
+void *split_page(pgd_t *pgtable, void *vaddr, unsigned int level);
+
+/*
+ * Applies the given protection bits to the given DAT tables level,
+ * splitting if necessary.
+ * @pgtable root of the page table tree
+ * @vaddr address whose protection bits are to be changed
+ * @prot the protection bits to set
+ * @level 3 (for 2GB pud), 4 (for 1MB pmd) or 5 (for 4KB pages)
+ */
+void protect_dat_entry(void *vaddr, unsigned long prot, unsigned int level);
+/*
+ * Clears the given protection bits from the given DAT tables level,
+ * splitting if necessary.
+ * @pgtable root of the page table tree
+ * @vaddr address whose protection bits are to be changed
+ * @prot the protection bits to clear
+ * @level 3 (for 2GB pud), 4 (for 1MB pmd) or 5 (for 4kB pages)
+ */
+void unprotect_dat_entry(void *vaddr, unsigned long prot, unsigned int level);
+
+/*
+ * Applies the given protection bits to the given 4kB pages range,
+ * splitting if necessary.
+ * @start starting address whose protection bits are to be changed
+ * @len size in bytes
+ * @prot the protection bits to set
+ */
void protect_range(void *start, unsigned long len, unsigned long prot);
-void unprotect_page(void *vaddr, unsigned long prot);
+/*
+ * Clears the given protection bits from the given 4kB pages range,
+ * splitting if necessary.
+ * @start starting address whose protection bits are to be changed
+ * @len size in bytes
+ * @prot the protection bits to set
+ */
void unprotect_range(void *start, unsigned long len, unsigned long prot);
+/* Similar to install_page, maps the virtual address to the physical address
+ * for the given page tables, using 1MB large pages.
+ * Returns a pointer to the DAT table entry.
+ * @pgtable root of the page table tree
+ * @phys physical address to map, must be 1MB aligned!
+ * @vaddr virtual address to map, must be 1MB aligned!
+ */
+pmdval_t *install_large_page(pgd_t *pgtable, phys_addr_t phys, void *vaddr);
+
+/* Similar to install_page, maps the virtual address to the physical address
+ * for the given page tables, using 2GB huge pages.
+ * Returns a pointer to the DAT table entry.
+ * @pgtable root of the page table tree
+ * @phys physical address to map, must be 2GB aligned!
+ * @vaddr virtual address to map, must be 2GB aligned!
+ */
+pudval_t *install_huge_page(pgd_t *pgtable, phys_addr_t phys, void *vaddr);
+
+static inline void protect_page(void *vaddr, unsigned long prot)
+{
+ protect_dat_entry(vaddr, prot, 5);
+}
+
+static inline void unprotect_page(void *vaddr, unsigned long prot)
+{
+ unprotect_dat_entry(vaddr, prot, 5);
+}
+void *get_dat_entry(pgd_t *pgtable, void *vaddr, unsigned int level);
+
#endif /* _ASMS390X_MMU_H_ */
diff --git a/lib/s390x/mmu.c b/lib/s390x/mmu.c
index 5c517366..def91334 100644
--- a/lib/s390x/mmu.c
+++ b/lib/s390x/mmu.c
@@ -15,6 +15,18 @@
#include <vmalloc.h>
#include "mmu.h"
+/*
+ * The naming convention used here is the same as used in the Linux kernel,
+ * and this is the corrispondence between the s390x architectural names and
+ * the Linux ones:
+ *
+ * pgd - region 1 table entry
+ * p4d - region 2 table entry
+ * pud - region 3 table entry
+ * pmd - segment table entry
+ * pte - page table entry
+ */
+
static pgd_t *table_root;
void configure_dat(int enable)
@@ -46,54 +58,254 @@ static void mmu_enable(pgd_t *pgtable)
lc->pgm_new_psw.mask |= PSW_MASK_DAT;
}
-static pteval_t *get_pte(pgd_t *pgtable, uintptr_t vaddr)
+/*
+ * Get the pud (region 3) DAT table entry for the given address and root,
+ * allocating it if necessary
+ */
+static inline pud_t *get_pud(pgd_t *pgtable, uintptr_t vaddr)
{
pgd_t *pgd = pgd_offset(pgtable, vaddr);
p4d_t *p4d = p4d_alloc(pgd, vaddr);
pud_t *pud = pud_alloc(p4d, vaddr);
- pmd_t *pmd = pmd_alloc(pud, vaddr);
- pte_t *pte = pte_alloc(pmd, vaddr);
- return &pte_val(*pte);
+ return pud;
+}
+
+/*
+ * Get the pmd (segment) DAT table entry for the given address and pud,
+ * allocating it if necessary.
+ * The pud must not be huge.
+ */
+static inline pmd_t *get_pmd(pud_t *pud, uintptr_t vaddr)
+{
+ pmd_t *pmd;
+
+ assert(!pud_huge(*pud));
+ pmd = pmd_alloc(pud, vaddr);
+ return pmd;
+}
+
+/*
+ * Get the pte (page) DAT table entry for the given address and pmd,
+ * allocating it if necessary.
+ * The pmd must not be large.
+ */
+static inline pte_t *get_pte(pmd_t *pmd, uintptr_t vaddr)
+{
+ pte_t *pte;
+
+ assert(!pmd_large(*pmd));
+ pte = pte_alloc(pmd, vaddr);
+ return pte;
+}
+
+/*
+ * Splits a large pmd (segment) DAT table entry into equivalent 4kB small
+ * pages.
+ * @pmd The pmd to split, it must be large.
+ * @va the virtual address corresponding to this pmd.
+ */
+static void split_pmd(pmd_t *pmd, uintptr_t va)
+{
+ phys_addr_t pa = pmd_val(*pmd) & SEGMENT_ENTRY_SFAA;
+ unsigned long i;
+ pte_t *pte;
+
+ assert(pmd_large(*pmd));
+ pte = alloc_pages(PAGE_TABLE_ORDER);
+ for (i = 0; i < PAGE_TABLE_ENTRIES; i++)
+ pte_val(pte[i]) = pa | PAGE_SIZE * i;
+ idte_pmdp(va, &pmd_val(*pmd));
+ pmd_val(*pmd) = __pa(pte) | SEGMENT_ENTRY_TT_SEGMENT;
+
+}
+
+/*
+ * Splits a huge pud (region 3) DAT table entry into equivalent 1MB large
+ * pages.
+ * @pud The pud to split, it must be huge.
+ * @va the virtual address corresponding to this pud.
+ */
+static void split_pud(pud_t *pud, uintptr_t va)
+{
+ phys_addr_t pa = pud_val(*pud) & REGION3_ENTRY_RFAA;
+ unsigned long i;
+ pmd_t *pmd;
+
+ assert(pud_huge(*pud));
+ pmd = alloc_pages(SEGMENT_TABLE_ORDER);
+ for (i = 0; i < SEGMENT_TABLE_ENTRIES; i++)
+ pmd_val(pmd[i]) = pa | SZ_1M * i | SEGMENT_ENTRY_FC | SEGMENT_ENTRY_TT_SEGMENT;
+ idte_pudp(va, &pud_val(*pud));
+ pud_val(*pud) = __pa(pmd) | REGION_ENTRY_TT_REGION3 | REGION_TABLE_LENGTH;
+}
+
+void *get_dat_entry(pgd_t *pgtable, void *vaddr, unsigned int level)
+{
+ uintptr_t va = (uintptr_t)vaddr;
+ pgd_t *pgd;
+ p4d_t *p4d;
+ pud_t *pud;
+ pmd_t *pmd;
+
+ assert(level && (level <= 5));
+ pgd = pgd_offset(pgtable, va);
+ if (level == 1)
+ return pgd;
+ p4d = p4d_alloc(pgd, va);
+ if (level == 2)
+ return p4d;
+ pud = pud_alloc(p4d, va);
+
+ if (level == 3)
+ return pud;
+ if (!pud_none(*pud) && pud_huge(*pud))
+ split_pud(pud, va);
+ pmd = get_pmd(pud, va);
+ if (level == 4)
+ return pmd;
+ if (!pmd_none(*pmd) && pmd_large(*pmd))
+ split_pmd(pmd, va);
+ return get_pte(pmd, va);
+}
+
+void *split_page(pgd_t *pgtable, void *vaddr, unsigned int level)
+{
+ assert((level >= 3) && (level <= 5));
+ return get_dat_entry(pgtable ? pgtable : table_root, vaddr, level);
}
phys_addr_t virt_to_pte_phys(pgd_t *pgtable, void *vaddr)
{
- return (*get_pte(pgtable, (uintptr_t)vaddr) & PAGE_MASK) +
- ((unsigned long)vaddr & ~PAGE_MASK);
+ uintptr_t va = (uintptr_t)vaddr;
+ pud_t *pud;
+ pmd_t *pmd;
+ pte_t *pte;
+
+ pud = get_pud(pgtable, va);
+ if (pud_huge(*pud))
+ return (pud_val(*pud) & REGION3_ENTRY_RFAA) | (va & ~REGION3_ENTRY_RFAA);
+ pmd = get_pmd(pud, va);
+ if (pmd_large(*pmd))
+ return (pmd_val(*pmd) & SEGMENT_ENTRY_SFAA) | (va & ~SEGMENT_ENTRY_SFAA);
+ pte = get_pte(pmd, va);
+ return (pte_val(*pte) & PAGE_MASK) | (va & ~PAGE_MASK);
+}
+
+/*
+ * Get the DAT table entry of the given level for the given address,
+ * splitting if necessary. If the entry was not invalid, invalidate it, and
+ * return the pointer to the entry and, if requested, its old value.
+ * @pgtable root of the page tables
+ * @vaddr virtual address
+ * @level 3 (for 2GB pud), 4 (for 1MB pmd) or 5 (for 4kB pages)
+ * @old if not NULL, will be written with the old value of the DAT table
+ * entry before invalidation
+ */
+static void *dat_get_and_invalidate(pgd_t *pgtable, void *vaddr, unsigned int level, unsigned long *old)
+{
+ unsigned long va = (unsigned long)vaddr;
+ void *ptr;
+
+ ptr = get_dat_entry(pgtable, vaddr, level);
+ if (old)
+ *old = *(unsigned long *)ptr;
+ if ((level == 1) && !pgd_none(*(pgd_t *)ptr))
+ idte_pgdp(va, ptr);
+ else if ((level == 2) && !p4d_none(*(p4d_t *)ptr))
+ idte_p4dp(va, ptr);
+ else if ((level == 3) && !pud_none(*(pud_t *)ptr))
+ idte_pudp(va, ptr);
+ else if ((level == 4) && !pmd_none(*(pmd_t *)ptr))
+ idte_pmdp(va, ptr);
+ else if (!pte_none(*(pte_t *)ptr))
+ ipte(va, ptr);
+ return ptr;
}
-static pteval_t *set_pte(pgd_t *pgtable, pteval_t val, void *vaddr)
+static void cleanup_pmd(pmd_t *pmd)
{
- pteval_t *p_pte = get_pte(pgtable, (uintptr_t)vaddr);
+ /* was invalid or large, nothing to do */
+ if (pmd_none(*pmd) || pmd_large(*pmd))
+ return;
+ /* was not large, free the corresponding page table */
+ free_pages((void *)(pmd_val(*pmd) & PAGE_MASK));
+}
- /* first flush the old entry (if we're replacing anything) */
- if (!(*p_pte & PAGE_ENTRY_I))
- ipte((uintptr_t)vaddr, p_pte);
+static void cleanup_pud(pud_t *pud)
+{
+ unsigned long i;
+ pmd_t *pmd;
- *p_pte = val;
- return p_pte;
+ /* was invalid or large, nothing to do */
+ if (pud_none(*pud) || pud_huge(*pud))
+ return;
+ /* recursively clean up all pmds if needed */
+ pmd = (pmd_t *)(pud_val(*pud) & PAGE_MASK);
+ for (i = 0; i < SEGMENT_TABLE_ENTRIES; i++)
+ cleanup_pmd(pmd + i);
+ /* free the corresponding segment table */
+ free_pages(pmd);
+}
+
+/*
+ * Set the DAT entry for the given level of the given virtual address. If a
+ * mapping already existed, it is overwritten. If an existing mapping with
+ * smaller pages existed, all the lower tables are freed.
+ * Returns the pointer to the DAT table entry.
+ * @pgtable root of the page tables
+ * @val the new value for the DAT table entry
+ * @vaddr the virtual address
+ * @level 3 for pud (region 3), 4 for pmd (segment) and 5 for pte (pages)
+ */
+static void *set_dat_entry(pgd_t *pgtable, unsigned long val, void *vaddr, unsigned int level)
+{
+ unsigned long old, *res;
+
+ res = dat_get_and_invalidate(pgtable, vaddr, level, &old);
+ if (level == 4)
+ cleanup_pmd((pmd_t *)&old);
+ if (level == 3)
+ cleanup_pud((pud_t *)&old);
+ *res = val;
+ return res;
}
pteval_t *install_page(pgd_t *pgtable, phys_addr_t phys, void *vaddr)
{
- return set_pte(pgtable, __pa(phys), vaddr);
+ assert(IS_ALIGNED(phys, PAGE_SIZE));
+ assert(IS_ALIGNED((uintptr_t)vaddr, PAGE_SIZE));
+ return set_dat_entry(pgtable, phys, vaddr, 5);
+}
+
+pmdval_t *install_large_page(pgd_t *pgtable, phys_addr_t phys, void *vaddr)
+{
+ assert(IS_ALIGNED(phys, SZ_1M));
+ assert(IS_ALIGNED((uintptr_t)vaddr, SZ_1M));
+ return set_dat_entry(pgtable, phys | SEGMENT_ENTRY_FC, vaddr, 4);
+}
+
+pudval_t *install_huge_page(pgd_t *pgtable, phys_addr_t phys, void *vaddr)
+{
+ assert(IS_ALIGNED(phys, SZ_2G));
+ assert(IS_ALIGNED((uintptr_t)vaddr, SZ_2G));
+ return set_dat_entry(pgtable, phys | REGION3_ENTRY_FC | REGION_ENTRY_TT_REGION3, vaddr, 3);
}
-void protect_page(void *vaddr, unsigned long prot)
+void protect_dat_entry(void *vaddr, unsigned long prot, unsigned int level)
{
- pteval_t *p_pte = get_pte(table_root, (uintptr_t)vaddr);
- pteval_t n_pte = *p_pte | prot;
+ unsigned long old, *ptr;
- set_pte(table_root, n_pte, vaddr);
+ ptr = dat_get_and_invalidate(table_root, vaddr, level, &old);
+ *ptr = old | prot;
}
-void unprotect_page(void *vaddr, unsigned long prot)
+void unprotect_dat_entry(void *vaddr, unsigned long prot, unsigned int level)
{
- pteval_t *p_pte = get_pte(table_root, (uintptr_t)vaddr);
- pteval_t n_pte = *p_pte & ~prot;
+ unsigned long old, *ptr;
- set_pte(table_root, n_pte, vaddr);
+ ptr = dat_get_and_invalidate(table_root, vaddr, level, &old);
+ *ptr = old & ~prot;
}
void protect_range(void *start, unsigned long len, unsigned long prot)
@@ -102,7 +314,7 @@ void protect_range(void *start, unsigned long len, unsigned long prot)
len &= PAGE_MASK;
for (; len; len -= PAGE_SIZE, curr += PAGE_SIZE)
- protect_page((void *)curr, prot);
+ protect_dat_entry((void *)curr, prot, 5);
}
void unprotect_range(void *start, unsigned long len, unsigned long prot)
@@ -111,7 +323,7 @@ void unprotect_range(void *start, unsigned long len, unsigned long prot)
len &= PAGE_MASK;
for (; len; len -= PAGE_SIZE, curr += PAGE_SIZE)
- unprotect_page((void *)curr, prot);
+ unprotect_dat_entry((void *)curr, prot, 5);
}
static void setup_identity(pgd_t *pgtable, phys_addr_t start_addr,
--
2.31.1
^ permalink raw reply related [flat|nested] 14+ messages in thread* Re: [kvm-unit-tests PATCH v4 6/7] s390x: mmu: add support for large pages
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 6/7] s390x: mmu: add support for large pages Claudio Imbrenda
@ 2021-05-28 8:44 ` Janosch Frank
2021-05-28 10:03 ` Claudio Imbrenda
0 siblings, 1 reply; 14+ messages in thread
From: Janosch Frank @ 2021-05-28 8:44 UTC (permalink / raw)
To: Claudio Imbrenda, kvm; +Cc: linux-s390, david, thuth, cohuck
On 5/26/21 3:42 PM, Claudio Imbrenda wrote:
> Add support for 1M and 2G pages.
>
> Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
> ---
> lib/s390x/mmu.h | 73 +++++++++++++-
> lib/s390x/mmu.c | 260 +++++++++++++++++++++++++++++++++++++++++++-----
> 2 files changed, 307 insertions(+), 26 deletions(-)
>
> diff --git a/lib/s390x/mmu.h b/lib/s390x/mmu.h
> index 603f289e..93208467 100644
> --- a/lib/s390x/mmu.h
> +++ b/lib/s390x/mmu.h
> @@ -10,9 +10,78 @@
> #ifndef _ASMS390X_MMU_H_
> #define _ASMS390X_MMU_H_
>
> -void protect_page(void *vaddr, unsigned long prot);
> +/*
> + * Splits the pagetables down to the given DAT tables level.
> + * Returns a pointer to the DAT table entry of the given level.
> + * @pgtable root of the page table tree
> + * @vaddr address whose page tables are to split
> + * @level 3 (for 2GB pud), 4 (for 1 MB pmd) or 5 (for 4KB pages)
> + */
> +void *split_page(pgd_t *pgtable, void *vaddr, unsigned int level);
> +
> +/*
> + * Applies the given protection bits to the given DAT tables level,
> + * splitting if necessary.
> + * @pgtable root of the page table tree
> + * @vaddr address whose protection bits are to be changed
> + * @prot the protection bits to set
> + * @level 3 (for 2GB pud), 4 (for 1MB pmd) or 5 (for 4KB pages)
> + */
> +void protect_dat_entry(void *vaddr, unsigned long prot, unsigned int level);
> +/*
> + * Clears the given protection bits from the given DAT tables level,
> + * splitting if necessary.
> + * @pgtable root of the page table tree
> + * @vaddr address whose protection bits are to be changed
> + * @prot the protection bits to clear
> + * @level 3 (for 2GB pud), 4 (for 1MB pmd) or 5 (for 4kB pages)
> + */
> +void unprotect_dat_entry(void *vaddr, unsigned long prot, unsigned int level);
> +
> +/*
> + * Applies the given protection bits to the given 4kB pages range,
> + * splitting if necessary.
> + * @start starting address whose protection bits are to be changed
> + * @len size in bytes
> + * @prot the protection bits to set
> + */
> void protect_range(void *start, unsigned long len, unsigned long prot);
> -void unprotect_page(void *vaddr, unsigned long prot);
> +/*
> + * Clears the given protection bits from the given 4kB pages range,
> + * splitting if necessary.
> + * @start starting address whose protection bits are to be changed
> + * @len size in bytes
> + * @prot the protection bits to set
> + */
> void unprotect_range(void *start, unsigned long len, unsigned long prot);
>
> +/* Similar to install_page, maps the virtual address to the physical address
> + * for the given page tables, using 1MB large pages.
> + * Returns a pointer to the DAT table entry.
> + * @pgtable root of the page table tree
> + * @phys physical address to map, must be 1MB aligned!
> + * @vaddr virtual address to map, must be 1MB aligned!
> + */
> +pmdval_t *install_large_page(pgd_t *pgtable, phys_addr_t phys, void *vaddr);
> +
> +/* Similar to install_page, maps the virtual address to the physical address
> + * for the given page tables, using 2GB huge pages.
> + * Returns a pointer to the DAT table entry.
> + * @pgtable root of the page table tree
> + * @phys physical address to map, must be 2GB aligned!
> + * @vaddr virtual address to map, must be 2GB aligned!
> + */
> +pudval_t *install_huge_page(pgd_t *pgtable, phys_addr_t phys, void *vaddr);
> +
> +static inline void protect_page(void *vaddr, unsigned long prot)
> +{
> + protect_dat_entry(vaddr, prot, 5);
> +}
> +
> +static inline void unprotect_page(void *vaddr, unsigned long prot)
> +{
> + unprotect_dat_entry(vaddr, prot, 5);
> +}
\n
> +void *get_dat_entry(pgd_t *pgtable, void *vaddr, unsigned int level);
> +
> #endif /* _ASMS390X_MMU_H_ */
> diff --git a/lib/s390x/mmu.c b/lib/s390x/mmu.c
> index 5c517366..def91334 100644
> --- a/lib/s390x/mmu.c
> +++ b/lib/s390x/mmu.c
> @@ -15,6 +15,18 @@
> #include <vmalloc.h>
> #include "mmu.h"
>
> +/*
> + * The naming convention used here is the same as used in the Linux kernel,
> + * and this is the corrispondence between the s390x architectural names and
corresponds
> + * the Linux ones:
> + *
> + * pgd - region 1 table entry
> + * p4d - region 2 table entry
> + * pud - region 3 table entry
> + * pmd - segment table entry
> + * pte - page table entry
> + */
> +
> static pgd_t *table_root;
>
> void configure_dat(int enable)
> @@ -46,54 +58,254 @@ static void mmu_enable(pgd_t *pgtable)
> lc->pgm_new_psw.mask |= PSW_MASK_DAT;
> }
>
> -static pteval_t *get_pte(pgd_t *pgtable, uintptr_t vaddr)
> +/*
> + * Get the pud (region 3) DAT table entry for the given address and root,
> + * allocating it if necessary
> + */
> +static inline pud_t *get_pud(pgd_t *pgtable, uintptr_t vaddr)
> {
> pgd_t *pgd = pgd_offset(pgtable, vaddr);
> p4d_t *p4d = p4d_alloc(pgd, vaddr);
> pud_t *pud = pud_alloc(p4d, vaddr);
> - pmd_t *pmd = pmd_alloc(pud, vaddr);
> - pte_t *pte = pte_alloc(pmd, vaddr);
>
> - return &pte_val(*pte);
> + return pud;
> +}
> +
> +/*
> + * Get the pmd (segment) DAT table entry for the given address and pud,
> + * allocating it if necessary.
> + * The pud must not be huge.
> + */
> +static inline pmd_t *get_pmd(pud_t *pud, uintptr_t vaddr)
> +{
> + pmd_t *pmd;
> +
> + assert(!pud_huge(*pud));
> + pmd = pmd_alloc(pud, vaddr);
Don't we have the *_alloc_map() functions in the kernel whic either map
or allocate? I'd prefer that naming over *_alloc() if you also map if
already allocated.
> + return pmd;
> +}
> +
> +/*
> + * Get the pte (page) DAT table entry for the given address and pmd,
> + * allocating it if necessary.
> + * The pmd must not be large.
> + */
> +static inline pte_t *get_pte(pmd_t *pmd, uintptr_t vaddr)
> +{
> + pte_t *pte;
> +
> + assert(!pmd_large(*pmd));
> + pte = pte_alloc(pmd, vaddr);
> + return pte;
> +}
> +
> +/*
> + * Splits a large pmd (segment) DAT table entry into equivalent 4kB small
> + * pages.
> + * @pmd The pmd to split, it must be large.
> + * @va the virtual address corresponding to this pmd.
> + */
> +static void split_pmd(pmd_t *pmd, uintptr_t va)
> +{
> + phys_addr_t pa = pmd_val(*pmd) & SEGMENT_ENTRY_SFAA;
> + unsigned long i;
> + pte_t *pte;
> +
> + assert(pmd_large(*pmd));
> + pte = alloc_pages(PAGE_TABLE_ORDER);
> + for (i = 0; i < PAGE_TABLE_ENTRIES; i++)
> + pte_val(pte[i]) = pa | PAGE_SIZE * i;
> + idte_pmdp(va, &pmd_val(*pmd));
> + pmd_val(*pmd) = __pa(pte) | SEGMENT_ENTRY_TT_SEGMENT;
Equivalent would mean we carry over protection, no?
> +
> +}
> +
> +/*
> + * Splits a huge pud (region 3) DAT table entry into equivalent 1MB large
> + * pages.
> + * @pud The pud to split, it must be huge.
> + * @va the virtual address corresponding to this pud.
> + */
> +static void split_pud(pud_t *pud, uintptr_t va)
> +{
> + phys_addr_t pa = pud_val(*pud) & REGION3_ENTRY_RFAA;
> + unsigned long i;
> + pmd_t *pmd;
> +
> + assert(pud_huge(*pud));
> + pmd = alloc_pages(SEGMENT_TABLE_ORDER);
> + for (i = 0; i < SEGMENT_TABLE_ENTRIES; i++)
> + pmd_val(pmd[i]) = pa | SZ_1M * i | SEGMENT_ENTRY_FC | SEGMENT_ENTRY_TT_SEGMENT;
> + idte_pudp(va, &pud_val(*pud));
> + pud_val(*pud) = __pa(pmd) | REGION_ENTRY_TT_REGION3 | REGION_TABLE_LENGTH;
> +}
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [kvm-unit-tests PATCH v4 6/7] s390x: mmu: add support for large pages
2021-05-28 8:44 ` Janosch Frank
@ 2021-05-28 10:03 ` Claudio Imbrenda
0 siblings, 0 replies; 14+ messages in thread
From: Claudio Imbrenda @ 2021-05-28 10:03 UTC (permalink / raw)
To: Janosch Frank; +Cc: kvm, linux-s390, david, thuth, cohuck
On Fri, 28 May 2021 10:44:32 +0200
Janosch Frank <frankja@linux.ibm.com> wrote:
> On 5/26/21 3:42 PM, Claudio Imbrenda wrote:
> > Add support for 1M and 2G pages.
> >
> > Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
> > ---
> > lib/s390x/mmu.h | 73 +++++++++++++-
> > lib/s390x/mmu.c | 260
> > +++++++++++++++++++++++++++++++++++++++++++----- 2 files changed,
> > 307 insertions(+), 26 deletions(-)
> >
> > diff --git a/lib/s390x/mmu.h b/lib/s390x/mmu.h
> > index 603f289e..93208467 100644
> > --- a/lib/s390x/mmu.h
> > +++ b/lib/s390x/mmu.h
> > @@ -10,9 +10,78 @@
> > #ifndef _ASMS390X_MMU_H_
> > #define _ASMS390X_MMU_H_
> >
> > -void protect_page(void *vaddr, unsigned long prot);
> > +/*
> > + * Splits the pagetables down to the given DAT tables level.
> > + * Returns a pointer to the DAT table entry of the given level.
> > + * @pgtable root of the page table tree
> > + * @vaddr address whose page tables are to split
> > + * @level 3 (for 2GB pud), 4 (for 1 MB pmd) or 5 (for 4KB pages)
> > + */
> > +void *split_page(pgd_t *pgtable, void *vaddr, unsigned int level);
> > +
> > +/*
> > + * Applies the given protection bits to the given DAT tables level,
> > + * splitting if necessary.
> > + * @pgtable root of the page table tree
> > + * @vaddr address whose protection bits are to be changed
> > + * @prot the protection bits to set
> > + * @level 3 (for 2GB pud), 4 (for 1MB pmd) or 5 (for 4KB pages)
> > + */
> > +void protect_dat_entry(void *vaddr, unsigned long prot, unsigned
> > int level); +/*
> > + * Clears the given protection bits from the given DAT tables
> > level,
> > + * splitting if necessary.
> > + * @pgtable root of the page table tree
> > + * @vaddr address whose protection bits are to be changed
> > + * @prot the protection bits to clear
> > + * @level 3 (for 2GB pud), 4 (for 1MB pmd) or 5 (for 4kB pages)
> > + */
> > +void unprotect_dat_entry(void *vaddr, unsigned long prot, unsigned
> > int level); +
> > +/*
> > + * Applies the given protection bits to the given 4kB pages range,
> > + * splitting if necessary.
> > + * @start starting address whose protection bits are to be changed
> > + * @len size in bytes
> > + * @prot the protection bits to set
> > + */
> > void protect_range(void *start, unsigned long len, unsigned long
> > prot); -void unprotect_page(void *vaddr, unsigned long prot);
> > +/*
> > + * Clears the given protection bits from the given 4kB pages range,
> > + * splitting if necessary.
> > + * @start starting address whose protection bits are to be changed
> > + * @len size in bytes
> > + * @prot the protection bits to set
> > + */
> > void unprotect_range(void *start, unsigned long len, unsigned long
> > prot);
> > +/* Similar to install_page, maps the virtual address to the
> > physical address
> > + * for the given page tables, using 1MB large pages.
> > + * Returns a pointer to the DAT table entry.
> > + * @pgtable root of the page table tree
> > + * @phys physical address to map, must be 1MB aligned!
> > + * @vaddr virtual address to map, must be 1MB aligned!
> > + */
> > +pmdval_t *install_large_page(pgd_t *pgtable, phys_addr_t phys,
> > void *vaddr); +
> > +/* Similar to install_page, maps the virtual address to the
> > physical address
> > + * for the given page tables, using 2GB huge pages.
> > + * Returns a pointer to the DAT table entry.
> > + * @pgtable root of the page table tree
> > + * @phys physical address to map, must be 2GB aligned!
> > + * @vaddr virtual address to map, must be 2GB aligned!
> > + */
> > +pudval_t *install_huge_page(pgd_t *pgtable, phys_addr_t phys, void
> > *vaddr); +
> > +static inline void protect_page(void *vaddr, unsigned long prot)
> > +{
> > + protect_dat_entry(vaddr, prot, 5);
> > +}
> > +
> > +static inline void unprotect_page(void *vaddr, unsigned long prot)
> > +{
> > + unprotect_dat_entry(vaddr, prot, 5);
> > +}
>
> \n
>
> > +void *get_dat_entry(pgd_t *pgtable, void *vaddr, unsigned int
> > level); +
> > #endif /* _ASMS390X_MMU_H_ */
> > diff --git a/lib/s390x/mmu.c b/lib/s390x/mmu.c
> > index 5c517366..def91334 100644
> > --- a/lib/s390x/mmu.c
> > +++ b/lib/s390x/mmu.c
> > @@ -15,6 +15,18 @@
> > #include <vmalloc.h>
> > #include "mmu.h"
> >
> > +/*
> > + * The naming convention used here is the same as used in the
> > Linux kernel,
> > + * and this is the corrispondence between the s390x architectural
> > names and
>
> corresponds
oops
> > + * the Linux ones:
> > + *
> > + * pgd - region 1 table entry
> > + * p4d - region 2 table entry
> > + * pud - region 3 table entry
> > + * pmd - segment table entry
> > + * pte - page table entry
> > + */
> > +
> > static pgd_t *table_root;
> >
> > void configure_dat(int enable)
> > @@ -46,54 +58,254 @@ static void mmu_enable(pgd_t *pgtable)
> > lc->pgm_new_psw.mask |= PSW_MASK_DAT;
> > }
> >
> > -static pteval_t *get_pte(pgd_t *pgtable, uintptr_t vaddr)
> > +/*
> > + * Get the pud (region 3) DAT table entry for the given address
> > and root,
> > + * allocating it if necessary
> > + */
> > +static inline pud_t *get_pud(pgd_t *pgtable, uintptr_t vaddr)
> > {
> > pgd_t *pgd = pgd_offset(pgtable, vaddr);
> > p4d_t *p4d = p4d_alloc(pgd, vaddr);
> > pud_t *pud = pud_alloc(p4d, vaddr);
> > - pmd_t *pmd = pmd_alloc(pud, vaddr);
> > - pte_t *pte = pte_alloc(pmd, vaddr);
> >
> > - return &pte_val(*pte);
> > + return pud;
> > +}
> > +
> > +/*
> > + * Get the pmd (segment) DAT table entry for the given address and
> > pud,
> > + * allocating it if necessary.
> > + * The pud must not be huge.
> > + */
> > +static inline pmd_t *get_pmd(pud_t *pud, uintptr_t vaddr)
> > +{
> > + pmd_t *pmd;
> > +
> > + assert(!pud_huge(*pud));
> > + pmd = pmd_alloc(pud, vaddr);
>
> Don't we have the *_alloc_map() functions in the kernel whic either
> map or allocate? I'd prefer that naming over *_alloc() if you also
> map if already allocated.
the functions existed already, I'm only reusing them.
> > + return pmd;
> > +}
> > +
> > +/*
> > + * Get the pte (page) DAT table entry for the given address and
> > pmd,
> > + * allocating it if necessary.
> > + * The pmd must not be large.
> > + */
> > +static inline pte_t *get_pte(pmd_t *pmd, uintptr_t vaddr)
> > +{
> > + pte_t *pte;
> > +
> > + assert(!pmd_large(*pmd));
> > + pte = pte_alloc(pmd, vaddr);
> > + return pte;
> > +}
> > +
> > +/*
> > + * Splits a large pmd (segment) DAT table entry into equivalent
> > 4kB small
> > + * pages.
> > + * @pmd The pmd to split, it must be large.
> > + * @va the virtual address corresponding to this pmd.
> > + */
> > +static void split_pmd(pmd_t *pmd, uintptr_t va)
> > +{
> > + phys_addr_t pa = pmd_val(*pmd) & SEGMENT_ENTRY_SFAA;
> > + unsigned long i;
> > + pte_t *pte;
> > +
> > + assert(pmd_large(*pmd));
> > + pte = alloc_pages(PAGE_TABLE_ORDER);
> > + for (i = 0; i < PAGE_TABLE_ENTRIES; i++)
> > + pte_val(pte[i]) = pa | PAGE_SIZE * i;
> > + idte_pmdp(va, &pmd_val(*pmd));
> > + pmd_val(*pmd) = __pa(pte) | SEGMENT_ENTRY_TT_SEGMENT;
>
> Equivalent would mean we carry over protection, no?
that is a good point... I will need to fix it
> > +
> > +}
> > +
> > +/*
> > + * Splits a huge pud (region 3) DAT table entry into equivalent
> > 1MB large
> > + * pages.
> > + * @pud The pud to split, it must be huge.
> > + * @va the virtual address corresponding to this pud.
> > + */
> > +static void split_pud(pud_t *pud, uintptr_t va)
> > +{
> > + phys_addr_t pa = pud_val(*pud) & REGION3_ENTRY_RFAA;
> > + unsigned long i;
> > + pmd_t *pmd;
> > +
> > + assert(pud_huge(*pud));
> > + pmd = alloc_pages(SEGMENT_TABLE_ORDER);
> > + for (i = 0; i < SEGMENT_TABLE_ENTRIES; i++)
> > + pmd_val(pmd[i]) = pa | SZ_1M * i |
> > SEGMENT_ENTRY_FC | SEGMENT_ENTRY_TT_SEGMENT;
> > + idte_pudp(va, &pud_val(*pud));
> > + pud_val(*pud) = __pa(pmd) | REGION_ENTRY_TT_REGION3 |
> > REGION_TABLE_LENGTH; +}
^ permalink raw reply [flat|nested] 14+ messages in thread
* [kvm-unit-tests PATCH v4 7/7] s390x: edat test
2021-05-26 13:42 [kvm-unit-tests PATCH v4 0/7] s390: Add support for large pages Claudio Imbrenda
` (5 preceding siblings ...)
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 6/7] s390x: mmu: add support for large pages Claudio Imbrenda
@ 2021-05-26 13:42 ` Claudio Imbrenda
2021-05-31 14:29 ` Janosch Frank
6 siblings, 1 reply; 14+ messages in thread
From: Claudio Imbrenda @ 2021-05-26 13:42 UTC (permalink / raw)
To: kvm; +Cc: linux-s390, david, thuth, frankja, cohuck
Simple EDAT test.
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
---
s390x/Makefile | 1 +
s390x/edat.c | 274 ++++++++++++++++++++++++++++++++++++++++++++
s390x/unittests.cfg | 3 +
3 files changed, 278 insertions(+)
create mode 100644 s390x/edat.c
diff --git a/s390x/Makefile b/s390x/Makefile
index b92de9c5..26bb1684 100644
--- a/s390x/Makefile
+++ b/s390x/Makefile
@@ -21,6 +21,7 @@ tests += $(TEST_DIR)/css.elf
tests += $(TEST_DIR)/uv-guest.elf
tests += $(TEST_DIR)/sie.elf
tests += $(TEST_DIR)/mvpg.elf
+tests += $(TEST_DIR)/edat.elf
tests_binary = $(patsubst %.elf,%.bin,$(tests))
ifneq ($(HOST_KEY_DOCUMENT),)
diff --git a/s390x/edat.c b/s390x/edat.c
new file mode 100644
index 00000000..0d56ed78
--- /dev/null
+++ b/s390x/edat.c
@@ -0,0 +1,274 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * EDAT test.
+ *
+ * Copyright (c) 2021 IBM Corp
+ *
+ * Authors:
+ * Claudio Imbrenda <imbrenda@linux.ibm.com>
+ */
+#include <libcflat.h>
+#include <vmalloc.h>
+#include <asm/facility.h>
+#include <asm/interrupt.h>
+#include <mmu.h>
+#include <asm/pgtable.h>
+#include <asm-generic/barrier.h>
+
+#define PGD_PAGE_SHIFT (REGION1_SHIFT - PAGE_SHIFT)
+
+#define LC_SIZE (2 * PAGE_SIZE)
+#define VIRT(x) ((void *)((unsigned long)(x) + (unsigned long)mem))
+
+static uint8_t prefix_buf[LC_SIZE] __attribute__((aligned(LC_SIZE)));
+static unsigned int tmp[1024] __attribute__((aligned(PAGE_SIZE)));
+static void *root, *mem, *m;
+static struct lowcore *lc;
+volatile unsigned int *p;
+
+/*
+ * Check if a non-access-list protection exception happened for the given
+ * address, in the primary address space.
+ */
+static bool check_pgm_prot(void *ptr)
+{
+ union teid teid;
+
+ if (lc->pgm_int_code != PGM_INT_CODE_PROTECTION)
+ return false;
+
+ teid.val = lc->trans_exc_id;
+
+ /*
+ * depending on the presence of the ESOP feature, the rest of the
+ * field might or might not be meaningful when the m field is 0.
+ */
+ if (!teid.m)
+ return true;
+ return (!teid.acc_list_prot && !teid.asce_id &&
+ (teid.addr == ((unsigned long)ptr >> PAGE_SHIFT)));
+}
+
+static void test_dat(void)
+{
+ report_prefix_push("edat off");
+ /* disable EDAT */
+ ctl_clear_bit(0, CTL0_EDAT);
+
+ /* Check some basics */
+ p[0] = 42;
+ report(p[0] == 42, "pte, r/w");
+ p[0] = 0;
+
+ /* Write protect the page and try to write, expect a fault */
+ protect_page(m, PAGE_ENTRY_P);
+ expect_pgm_int();
+ p[0] = 42;
+ unprotect_page(m, PAGE_ENTRY_P);
+ report(!p[0] && check_pgm_prot(m), "pte, ro");
+
+ /*
+ * The FC bit (for large pages) should be ignored because EDAT is
+ * off. We set a value and then we try to read it back again after
+ * setting the FC bit. This way we can check if large pages were
+ * erroneously enabled despite EDAT being off.
+ */
+ p[0] = 42;
+ protect_dat_entry(m, SEGMENT_ENTRY_FC, 4);
+ report(p[0] == 42, "pmd, fc=1, r/w");
+ unprotect_dat_entry(m, SEGMENT_ENTRY_FC, 4);
+ p[0] = 0;
+
+ /*
+ * Segment protection should work even with EDAT off, try to write
+ * anyway and expect a fault
+ */
+ protect_dat_entry(m, SEGMENT_ENTRY_P, 4);
+ expect_pgm_int();
+ p[0] = 42;
+ report(!p[0] && check_pgm_prot(m), "pmd, ro");
+ unprotect_dat_entry(m, SEGMENT_ENTRY_P, 4);
+
+ /* The FC bit should be ignored because EDAT is off, like above */
+ p[0] = 42;
+ protect_dat_entry(m, REGION3_ENTRY_FC, 3);
+ report(p[0] == 42, "pud, fc=1, r/w");
+ unprotect_dat_entry(m, REGION3_ENTRY_FC, 3);
+ p[0] = 0;
+
+ /*
+ * Region1/2/3 protection should not work, because EDAT is off.
+ * Protect the various region1/2/3 entries and write, expect the
+ * write to be successful.
+ */
+ protect_dat_entry(m, REGION_ENTRY_P, 3);
+ p[0] = 42;
+ report(p[0] == 42, "pud, ro");
+ unprotect_dat_entry(m, REGION_ENTRY_P, 3);
+ p[0] = 0;
+
+ protect_dat_entry(m, REGION_ENTRY_P, 2);
+ p[0] = 42;
+ report(p[0] == 42, "p4d, ro");
+ unprotect_dat_entry(m, REGION_ENTRY_P, 2);
+ p[0] = 0;
+
+ protect_dat_entry(m, REGION_ENTRY_P, 1);
+ p[0] = 42;
+ report(p[0] == 42, "pgd, ro");
+ unprotect_dat_entry(m, REGION_ENTRY_P, 1);
+ p[0] = 0;
+
+ report_prefix_pop();
+}
+
+static void test_edat1(void)
+{
+ report_prefix_push("edat1");
+ /* Enable EDAT */
+ ctl_set_bit(0, CTL0_EDAT);
+ p[0] = 0;
+
+ /*
+ * Segment protection should work normally, try to write and expect
+ * a fault.
+ */
+ expect_pgm_int();
+ protect_dat_entry(m, SEGMENT_ENTRY_P, 4);
+ p[0] = 42;
+ report(!p[0] && check_pgm_prot(m), "pmd, ro");
+ unprotect_dat_entry(m, SEGMENT_ENTRY_P, 4);
+
+ /*
+ * Region1/2/3 protection should work now, because EDAT is on. Try
+ * to write anyway and expect a fault.
+ */
+ expect_pgm_int();
+ protect_dat_entry(m, REGION_ENTRY_P, 3);
+ p[0] = 42;
+ report(!p[0] && check_pgm_prot(m), "pud, ro");
+ unprotect_dat_entry(m, REGION_ENTRY_P, 3);
+
+ expect_pgm_int();
+ protect_dat_entry(m, REGION_ENTRY_P, 2);
+ p[0] = 42;
+ report(!p[0] && check_pgm_prot(m), "p4d, ro");
+ unprotect_dat_entry(m, REGION_ENTRY_P, 2);
+
+ expect_pgm_int();
+ protect_dat_entry(m, REGION_ENTRY_P, 1);
+ p[0] = 42;
+ report(!p[0] && check_pgm_prot(m), "pgd, ro");
+ unprotect_dat_entry(m, REGION_ENTRY_P, 1);
+
+ /* Large pages should work */
+ p[0] = 42;
+ install_large_page(root, 0, mem);
+ report(p[0] == 42, "pmd, large");
+
+ /*
+ * Prefixing should not work with large pages. Since the lower
+ * addresses are mapped with small pages, which are subject to
+ * prefixing, and the pages mapped with large pages are not subject
+ * to prefixing, this is the resulting scenario:
+ *
+ * virtual 0 = real 0 -> absolute prefix_buf
+ * virtual prefix_buf = real prefix_buf -> absolute 0
+ * VIRT(0) -> absolute 0
+ * VIRT(prefix_buf) -> absolute prefix_buf
+ *
+ * The testcase checks if the memory at virtual 0 has the same
+ * content as the memory at VIRT(prefix_buf) and the memory at
+ * VIRT(0) has the same content as the memory at virtual prefix_buf.
+ * If prefixing is erroneously applied for large pages, the testcase
+ * will therefore fail.
+ */
+ report(!memcmp(0, VIRT(prefix_buf), LC_SIZE) &&
+ !memcmp(prefix_buf, VIRT(0), LC_SIZE),
+ "pmd, large, prefixing");
+
+ report_prefix_pop();
+}
+
+static void test_edat2(void)
+{
+ report_prefix_push("edat2");
+ p[0] = 42;
+
+ /* Huge pages should work */
+ install_huge_page(root, 0, mem);
+ report(p[0] == 42, "pud, huge");
+
+ /* Prefixing should not work with huge pages, just like large pages */
+ report(!memcmp(0, VIRT(prefix_buf), LC_SIZE) &&
+ !memcmp(prefix_buf, VIRT(0), LC_SIZE),
+ "pmd, large, prefixing");
+
+ report_prefix_pop();
+}
+
+static unsigned int setup(void)
+{
+ bool has_edat1 = test_facility(8);
+ bool has_edat2 = test_facility(78);
+ unsigned long pa, va;
+
+ if (has_edat2 && !has_edat1)
+ report_abort("EDAT2 available, but EDAT1 not available");
+
+ /* Setup DAT 1:1 mapping and memory management */
+ setup_vm();
+ root = (void *)(stctg(1) & PAGE_MASK);
+
+ /*
+ * Get a pgd worth of virtual memory, so we can test things later
+ * without interfering with the test code or the interrupt handler
+ */
+ mem = alloc_vpages_aligned(BIT_ULL(PGD_PAGE_SHIFT), PGD_PAGE_SHIFT);
+ assert(mem);
+ va = (unsigned long)mem;
+
+ /* Map the first 1GB of real memory */
+ for (pa = 0; pa < SZ_1G; pa += PAGE_SIZE, va += PAGE_SIZE)
+ install_page(root, pa, (void *)va);
+
+ /*
+ * Move the lowcore to a known non-zero location. This is needed
+ * later to check whether prefixing is working with large pages.
+ */
+ assert((unsigned long)&prefix_buf < SZ_2G);
+ memcpy(prefix_buf, 0, LC_SIZE);
+ set_prefix((uint32_t)(uintptr_t)prefix_buf);
+ /* Clear the old copy */
+ memset(prefix_buf, 0, LC_SIZE);
+
+ /* m will point to tmp through the new virtual mapping */
+ m = VIRT(&tmp);
+ /* p is the same as m but volatile */
+ p = (volatile unsigned int *)m;
+
+ return has_edat1 + has_edat2;
+}
+
+int main(void)
+{
+ unsigned int edat;
+
+ report_prefix_push("edat");
+ edat = setup();
+
+ test_dat();
+
+ if (edat)
+ test_edat1();
+ else
+ report_skip("EDAT not available");
+
+ if (edat >= 2)
+ test_edat2();
+ else
+ report_skip("EDAT2 not available");
+
+ report_prefix_pop();
+ return report_summary();
+}
diff --git a/s390x/unittests.cfg b/s390x/unittests.cfg
index 9f81a608..a0ec8864 100644
--- a/s390x/unittests.cfg
+++ b/s390x/unittests.cfg
@@ -103,3 +103,6 @@ file = sie.elf
[mvpg]
file = mvpg.elf
timeout = 10
+
+[edat]
+file = edat.elf
--
2.31.1
^ permalink raw reply related [flat|nested] 14+ messages in thread* Re: [kvm-unit-tests PATCH v4 7/7] s390x: edat test
2021-05-26 13:42 ` [kvm-unit-tests PATCH v4 7/7] s390x: edat test Claudio Imbrenda
@ 2021-05-31 14:29 ` Janosch Frank
0 siblings, 0 replies; 14+ messages in thread
From: Janosch Frank @ 2021-05-31 14:29 UTC (permalink / raw)
To: Claudio Imbrenda, kvm; +Cc: linux-s390, david, thuth, cohuck
On 5/26/21 3:42 PM, Claudio Imbrenda wrote:
> Simple EDAT test.
>
> Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Some nits below.
Acked-by Janosch Frank <frankja@linux.ibm.com>
> ---
> s390x/Makefile | 1 +
> s390x/edat.c | 274 ++++++++++++++++++++++++++++++++++++++++++++
> s390x/unittests.cfg | 3 +
> 3 files changed, 278 insertions(+)
> create mode 100644 s390x/edat.c
>
> diff --git a/s390x/Makefile b/s390x/Makefile
> index b92de9c5..26bb1684 100644
> --- a/s390x/Makefile
> +++ b/s390x/Makefile
> @@ -21,6 +21,7 @@ tests += $(TEST_DIR)/css.elf
> tests += $(TEST_DIR)/uv-guest.elf
> tests += $(TEST_DIR)/sie.elf
> tests += $(TEST_DIR)/mvpg.elf
> +tests += $(TEST_DIR)/edat.elf
>
> tests_binary = $(patsubst %.elf,%.bin,$(tests))
> ifneq ($(HOST_KEY_DOCUMENT),)
> diff --git a/s390x/edat.c b/s390x/edat.c
> new file mode 100644
> index 00000000..0d56ed78
> --- /dev/null
> +++ b/s390x/edat.c
> @@ -0,0 +1,274 @@
> +/* SPDX-License-Identifier: GPL-2.0-only */
> +/*
> + * EDAT test.
> + *
> + * Copyright (c) 2021 IBM Corp
> + *
> + * Authors:
> + * Claudio Imbrenda <imbrenda@linux.ibm.com>
> + */
> +#include <libcflat.h>
> +#include <vmalloc.h>
> +#include <asm/facility.h>
> +#include <asm/interrupt.h>
> +#include <mmu.h>
> +#include <asm/pgtable.h>
> +#include <asm-generic/barrier.h>
> +
> +#define PGD_PAGE_SHIFT (REGION1_SHIFT - PAGE_SHIFT)
> +
> +#define LC_SIZE (2 * PAGE_SIZE)
> +#define VIRT(x) ((void *)((unsigned long)(x) + (unsigned long)mem))
> +
> +static uint8_t prefix_buf[LC_SIZE] __attribute__((aligned(LC_SIZE)));
> +static unsigned int tmp[1024] __attribute__((aligned(PAGE_SIZE)));
> +static void *root, *mem, *m;
> +static struct lowcore *lc;
> +volatile unsigned int *p;
> +
> +/*
> + * Check if a non-access-list protection exception happened for the given
> + * address, in the primary address space.
> + */
> +static bool check_pgm_prot(void *ptr)
> +{
> + union teid teid;
> +
> + if (lc->pgm_int_code != PGM_INT_CODE_PROTECTION)
> + return false;
> +
> + teid.val = lc->trans_exc_id;
> +
> + /*
> + * depending on the presence of the ESOP feature, the rest of the
> + * field might or might not be meaningful when the m field is 0.
> + */
> + if (!teid.m)
> + return true;
> + return (!teid.acc_list_prot && !teid.asce_id &&
> + (teid.addr == ((unsigned long)ptr >> PAGE_SHIFT)));
> +}
> +
> +static void test_dat(void)
> +{
> + report_prefix_push("edat off");
> + /* disable EDAT */
> + ctl_clear_bit(0, CTL0_EDAT);
> +
> + /* Check some basics */
> + p[0] = 42;
> + report(p[0] == 42, "pte, r/w");
> + p[0] = 0;
> +
> + /* Write protect the page and try to write, expect a fault */
> + protect_page(m, PAGE_ENTRY_P);
> + expect_pgm_int();
> + p[0] = 42;
> + unprotect_page(m, PAGE_ENTRY_P);
> + report(!p[0] && check_pgm_prot(m), "pte, ro");
> +
> + /*
> + * The FC bit (for large pages) should be ignored because EDAT is
> + * off. We set a value and then we try to read it back again after
> + * setting the FC bit. This way we can check if large pages were
> + * erroneously enabled despite EDAT being off.
> + */
> + p[0] = 42;
> + protect_dat_entry(m, SEGMENT_ENTRY_FC, 4);
> + report(p[0] == 42, "pmd, fc=1, r/w");
> + unprotect_dat_entry(m, SEGMENT_ENTRY_FC, 4);
> + p[0] = 0;
Could we pack that into a function and re-use it for the checks below?
> +
> + /*
> + * Segment protection should work even with EDAT off, try to write
> + * anyway and expect a fault
> + */
> + protect_dat_entry(m, SEGMENT_ENTRY_P, 4);
These page table level magic constants give me a headache.
PGTABLE_LEVEL_PMD ?
> + expect_pgm_int();
> + p[0] = 42;
> + report(!p[0] && check_pgm_prot(m), "pmd, ro");
> + unprotect_dat_entry(m, SEGMENT_ENTRY_P, 4);
> +
> + /* The FC bit should be ignored because EDAT is off, like above */
> + p[0] = 42;
> + protect_dat_entry(m, REGION3_ENTRY_FC, 3);
> + report(p[0] == 42, "pud, fc=1, r/w");
> + unprotect_dat_entry(m, REGION3_ENTRY_FC, 3);
> + p[0] = 0;
> +
> + /*
> + * Region1/2/3 protection should not work, because EDAT is off.
> + * Protect the various region1/2/3 entries and write, expect the
> + * write to be successful.
> + */
> + protect_dat_entry(m, REGION_ENTRY_P, 3);
> + p[0] = 42;
> + report(p[0] == 42, "pud, ro");
> + unprotect_dat_entry(m, REGION_ENTRY_P, 3);
> + p[0] = 0;
> +
> + protect_dat_entry(m, REGION_ENTRY_P, 2);
> + p[0] = 42;
> + report(p[0] == 42, "p4d, ro");
> + unprotect_dat_entry(m, REGION_ENTRY_P, 2);
> + p[0] = 0;
> +
> + protect_dat_entry(m, REGION_ENTRY_P, 1);
> + p[0] = 42;
> + report(p[0] == 42, "pgd, ro");
> + unprotect_dat_entry(m, REGION_ENTRY_P, 1);
> + p[0] = 0;
> +
> + report_prefix_pop();
> +}
> +
> +static void test_edat1(void)
> +{
> + report_prefix_push("edat1");
> + /* Enable EDAT */
> + ctl_set_bit(0, CTL0_EDAT);
> + p[0] = 0;
> +
> + /*
> + * Segment protection should work normally, try to write and expect
> + * a fault.
> + */
> + expect_pgm_int();
> + protect_dat_entry(m, SEGMENT_ENTRY_P, 4);
> + p[0] = 42;
> + report(!p[0] && check_pgm_prot(m), "pmd, ro");
> + unprotect_dat_entry(m, SEGMENT_ENTRY_P, 4);
> +
> + /*
> + * Region1/2/3 protection should work now, because EDAT is on. Try
> + * to write anyway and expect a fault.
> + */
> + expect_pgm_int();
> + protect_dat_entry(m, REGION_ENTRY_P, 3);
> + p[0] = 42;
> + report(!p[0] && check_pgm_prot(m), "pud, ro");
> + unprotect_dat_entry(m, REGION_ENTRY_P, 3);
> +
> + expect_pgm_int();
> + protect_dat_entry(m, REGION_ENTRY_P, 2);
> + p[0] = 42;
> + report(!p[0] && check_pgm_prot(m), "p4d, ro");
> + unprotect_dat_entry(m, REGION_ENTRY_P, 2);
> +
> + expect_pgm_int();
> + protect_dat_entry(m, REGION_ENTRY_P, 1);
> + p[0] = 42;
> + report(!p[0] && check_pgm_prot(m), "pgd, ro");
> + unprotect_dat_entry(m, REGION_ENTRY_P, 1);
> +
> + /* Large pages should work */
> + p[0] = 42;
> + install_large_page(root, 0, mem);
> + report(p[0] == 42, "pmd, large");
> +
> + /*
> + * Prefixing should not work with large pages. Since the lower
> + * addresses are mapped with small pages, which are subject to
> + * prefixing, and the pages mapped with large pages are not subject
> + * to prefixing, this is the resulting scenario:
> + *
> + * virtual 0 = real 0 -> absolute prefix_buf
> + * virtual prefix_buf = real prefix_buf -> absolute 0
> + * VIRT(0) -> absolute 0
> + * VIRT(prefix_buf) -> absolute prefix_buf
> + *
> + * The testcase checks if the memory at virtual 0 has the same
> + * content as the memory at VIRT(prefix_buf) and the memory at
> + * VIRT(0) has the same content as the memory at virtual prefix_buf.
> + * If prefixing is erroneously applied for large pages, the testcase
> + * will therefore fail.
> + */
> + report(!memcmp(0, VIRT(prefix_buf), LC_SIZE) &&
> + !memcmp(prefix_buf, VIRT(0), LC_SIZE),
> + "pmd, large, prefixing");
> +
> + report_prefix_pop();
> +}
> +
> +static void test_edat2(void)
> +{
> + report_prefix_push("edat2");
> + p[0] = 42;
> +
> + /* Huge pages should work */
> + install_huge_page(root, 0, mem);
> + report(p[0] == 42, "pud, huge");
> +
> + /* Prefixing should not work with huge pages, just like large pages */
> + report(!memcmp(0, VIRT(prefix_buf), LC_SIZE) &&
> + !memcmp(prefix_buf, VIRT(0), LC_SIZE),
> + "pmd, large, prefixing");
> +
> + report_prefix_pop();
> +}
> +
> +static unsigned int setup(void)
> +{
> + bool has_edat1 = test_facility(8);
> + bool has_edat2 = test_facility(78);
> + unsigned long pa, va;
> +
> + if (has_edat2 && !has_edat1)
> + report_abort("EDAT2 available, but EDAT1 not available");
> +
> + /* Setup DAT 1:1 mapping and memory management */
> + setup_vm();
> + root = (void *)(stctg(1) & PAGE_MASK);
> +
> + /*
> + * Get a pgd worth of virtual memory, so we can test things later
> + * without interfering with the test code or the interrupt handler
> + */
> + mem = alloc_vpages_aligned(BIT_ULL(PGD_PAGE_SHIFT), PGD_PAGE_SHIFT);
> + assert(mem);
> + va = (unsigned long)mem;
> +
> + /* Map the first 1GB of real memory */
> + for (pa = 0; pa < SZ_1G; pa += PAGE_SIZE, va += PAGE_SIZE)
> + install_page(root, pa, (void *)va);
> +
> + /*
> + * Move the lowcore to a known non-zero location. This is needed
> + * later to check whether prefixing is working with large pages.
> + */
> + assert((unsigned long)&prefix_buf < SZ_2G);
> + memcpy(prefix_buf, 0, LC_SIZE);
> + set_prefix((uint32_t)(uintptr_t)prefix_buf);
> + /* Clear the old copy */
> + memset(prefix_buf, 0, LC_SIZE);
> +
> + /* m will point to tmp through the new virtual mapping */
> + m = VIRT(&tmp);
> + /* p is the same as m but volatile */
> + p = (volatile unsigned int *)m;
> +
> + return has_edat1 + has_edat2;
> +}
> +
> +int main(void)
> +{
> + unsigned int edat;
> +
> + report_prefix_push("edat");
> + edat = setup();
> +
> + test_dat();
> +
> + if (edat)
> + test_edat1();
> + else
> + report_skip("EDAT not available");
> +
> + if (edat >= 2)
> + test_edat2();
> + else
> + report_skip("EDAT2 not available");
> +
> + report_prefix_pop();
> + return report_summary();
> +}
> diff --git a/s390x/unittests.cfg b/s390x/unittests.cfg
> index 9f81a608..a0ec8864 100644
> --- a/s390x/unittests.cfg
> +++ b/s390x/unittests.cfg
> @@ -103,3 +103,6 @@ file = sie.elf
> [mvpg]
> file = mvpg.elf
> timeout = 10
> +
> +[edat]
> +file = edat.elf
>
^ permalink raw reply [flat|nested] 14+ messages in thread