s390: Add support for large pages
Subsystems (3)
v4 cover s390: Add support for large pages
Introduce support for large (1M) and huge (2G) pages.
Add a simple testcase for EDAT1 and EDAT2.
v3->v4
* replace macros in patch 5 with a union representing TEID fields
* clear the teid in expect_pgm_int and clear_pgm_int
* update testcase to use expect_pgm_int, remove expect_dat_fault
* update testcase to use teid union
v2->v3
* Add proper macros for control register bits
* Improved patch titles and descriptions
* Moved definition of TEID bits to library
* Rebased on the lastest upstream branch
v1->v2
* split patch 2 -> new patch 2 and new patch 3
* patch 2: fixes pgtable.h, also fixes wrong usage of REGION_TABLE_LENGTH instead of SEGMENT_TABLE_LENGTH
* patch 3: introduces new macros and functions for large pages
* patch 4: remove erroneous double call to pte_alloc in get_pte
* patch 4: added comment in mmu.c to bridge the s390x architecural names with the Linux ones used in the kvm-unit-tests
* patch 5: added and fixed lots of comments to explain what's going on
* patch 5: set FC for region 3 after writing the canary, like for segments
* patch 5: use uintptr_t instead of intptr_t for set_prefix
* patch 5: introduce new macro PGD_PAGE_SHIFT instead of using magic value 41
* patch 5: use VIRT(0) instead of mem to make it more clear what we are doing, even though VIRT(0) expands to mem
Claudio Imbrenda (7):
s390x: lib: add and use macros for control register bits
libcflat: add SZ_1M and SZ_2G
s390x: lib: fix pgtable.h
s390x: lib: Add idte and other huge pages functions/macros
s390x: lib: add teid union and clear teid from lowcore
s390x: mmu: add support for large pages
s390x: edat test
s390x/Makefile | 1 + lib/s390x/asm/arch_def.h | 12 ++ lib/s390x/asm/float.h | 4 +- lib/s390x/asm/interrupt.h | 26 +++- lib/s390x/asm/pgtable.h | 44 +++++- lib/libcflat.h | 2 + lib/s390x/mmu.h | 73 +++++++++- lib/s390x/interrupt.c | 2 + lib/s390x/mmu.c | 260 ++++++++++++++++++++++++++++++++---- lib/s390x/sclp.c | 4 +- s390x/diag288.c | 2 +- s390x/edat.c | 274 ++++++++++++++++++++++++++++++++++++++ s390x/gs.c | 2 +- s390x/iep.c | 4 +- s390x/skrf.c | 2 +- s390x/smp.c | 8 +- s390x/vector.c | 2 +- s390x/unittests.cfg | 3 + 18 files changed, 676 insertions(+), 49 deletions(-) create mode 100644 s390x/edat.c
Signature · 1 line
2.31.1
PATCH v4 1/7 s390x: lib: add and use macros for control register bits
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.
--- 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(-)
lib/s390x/asm/arch_def.h +12 -0
@@ -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;
lib/s390x/asm/float.h +2 -2
@@ -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
lib/s390x/asm/interrupt.h +2 -2
@@ -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
lib/s390x/sclp.c +2 -2
@@ -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);
s390x/diag288.c +1 -1
@@ -86,7 +86,7 @@ static void test_bite(void) asmvolatile ("stck %0" :"=Q" (time) : :"cc" ); time += (uint64_t )(16000 *1000 ) <<12 ; asmvolatile ("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);
s390x/gs.c +1 -1
@@ -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 ;
s390x/iep.c +2 -2
@@ -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); }
s390x/skrf.c +1 -1
@@ -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);
s390x/smp.c +4 -4
@@ -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);
s390x/vector.c +1 -1
@@ -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 );
Signature · 1 line
2.31.1
PATCH v4 2/7 libcflat: add SZ_1M and SZ_2G
In reply to Claudio ImbrendaAdd SZ_1M and SZ_2G to libcflat.h
s390x needs those for large/huge pages
--- lib/libcflat.h | 2 ++ 1 file changed, 2 insertions(+)
lib/libcflat.h +2 -0
@@ -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
Signature · 1 line
2.31.1
PATCH v4 3/7 s390x: lib: fix pgtable.h
In reply to Claudio ImbrendaFix 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
--- lib/s390x/asm/pgtable.h | 13 ++++++------- 1 file changed, 6 insertions(+), 7 deletions(-)
lib/s390x/asm/pgtable.h +6 -7
@@ -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; asmvolatile ( " ipte %0,%1\n"
Signature · 1 line
2.31.1
Re: [kvm-unit-tests PATCH v4 3/7] s390x: lib: fix pgtable.h
Quoted 11 lines · Claudio Imbrenda
Fix
pgtable.h:*
SEGMENT_ENTRY_SFAAhad 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 useREGION_ENTRY_TLinstead of*_TABLE_LENGTH
-*_TABLE_LENGTHneed to stay, because they should be used for ASCEs
Quoted 1 line
Signed-off-by Claudio Imbrenda <[email protected]>
Quoted 63 lines
--- lib/s390x/asm/pgtable.h | 13 ++++++------- 1 file changed, 6 insertions(+), 7 deletions(-)lib/s390x/asm/pgtable.h +6
-7@@ -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; asmvolatile ( " ipte %0,%1\n"
PATCH v4 4/7 s390x: lib: Add idte and other huge pages functions/macros
In reply to Claudio ImbrendaImprove pgtable.h:
* add macros to check whether a pmd or a pud are large / huge
* add idte functions for pmd, pud, p4d and pgd
--- lib/s390x/asm/pgtable.h | 31 +++++++++++++++++++++++++++++++ 1 file changed, 31 insertions(+)
lib/s390x/asm/pgtable.h +31 -0
@@ -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 ; + asmvolatile ( +" 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_ */
Signature · 1 line
2.31.1
PATCH v4 5/7 s390x: lib: add teid union and clear teid from lowcore
In reply to Claudio ImbrendaAdd a union to represent Translation-Exception Identification (TEID).
Clear the TEID in expect_pgm_int clear_pgm_int.
--- lib/s390x/asm/interrupt.h | 22 ++++++++++++++++++++++ lib/s390x/interrupt.c | 2 ++ 2 files changed, 24 insertions(+)
lib/s390x/asm/interrupt.h +22 -0
@@ -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);
lib/s390x/interrupt.c +2 -0
@@ -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; }
Signature · 1 line
2.31.1
Re: [kvm-unit-tests PATCH v4 5/7] s390x: lib: add teid union and clear teid from lowcore
Quoted 5 lines · Claudio Imbrenda
Add a union to represent Translation-Exception Identification (TEID).
Clear the TEID in
expect_pgm_intclear_pgm_int.Signed-off-by Claudio Imbrenda <[email protected]>
Quoted 18 lines
--- lib/s390x/asm/interrupt.h | 22 ++++++++++++++++++++++ lib/s390x/interrupt.c | 2 ++ 2 files changed, 24 insertions(+)lib/s390x/asm/interrupt.h +5
-0@@ -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 +
lib/s390x/asm/interrupt.h line 30
Quoted 10 lines
lib/s390x/asm/interrupt.h +10
-0+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
lib/s390x/asm/interrupt.h line 32
Quoted 2 lines
lib/s390x/asm/interrupt.h +2
-0+* 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
*/
Quoted 29 lines
lib/s390x/asm/interrupt.h +5
-0+ 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);lib/s390x/interrupt.c +2
-0@@ -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; }
Re: [kvm-unit-tests PATCH v4 5/7] s390x: lib: add teid union and clear teid from lowcore
Quoted 34 lines · On Thu, 27 May 2021 16:52:48 +0200 Janosch Frank <[email protected]> wrote:
Quoted 6 lines · Claudio Imbrenda
Add a union to represent Translation-Exception Identification (TEID).
Clear the TEID in
expect_pgm_intclear_pgm_int.Signed-off-by Claudio Imbrenda <[email protected]>Reviewed-by Janosch Frank <[email protected]>Quoted 24 lines
--- lib/s390x/asm/interrupt.h | 22 ++++++++++++++++++++++ lib/s390x/interrupt.c | 2 ++ 2 files changed, 24 insertions(+)lib/s390x/asm/interrupt.h +11
-0@@ -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;
Quoted 8 lines
lib/s390x/asm/interrupt.h line 32
Quoted 6 lines
lib/s390x/asm/interrupt.h +4
-0+ 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
Quoted 10 lines
Quoted 4 lines
lib/s390x/asm/interrupt.h +1
-0+* including whether the exception was triggered by a MVPG
lib/s390x/asm/interrupt.h +1
-0+* instruction, or whether the addr field is meaningful */
Could you please convert the comment style to this?
/*
* Text
*/
and fixed
Quoted 30 lines
Quoted 29 lines
lib/s390x/asm/interrupt.h +5
-0+ 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);lib/s390x/interrupt.c +2
-0@@ -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; }
PATCH v4 6/7 s390x: mmu: add support for large pages
In reply to Claudio ImbrendaAdd support for 1M and 2G pages.
--- lib/s390x/mmu.h | 73 +++++++++++++- lib/s390x/mmu.c | 260 +++++++++++++++++++++++++++++++++++++++++++----- 2 files changed, 307 insertions(+), 26 deletions(-)
lib/s390x/mmu.h +71 -2
@@ -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_ */
lib/s390x/mmu.c +236 -24
310 lines not on this page · raw message
Signature · 1 line
2.31.1
Re: [kvm-unit-tests PATCH v4 6/7] s390x: mmu: add support for large pages
Quoted 80 lines · Claudio Imbrenda
Add support for 1M and 2G pages.
Signed-off-by Claudio Imbrenda <[email protected]>--- lib/s390x/mmu.h | 73 +++++++++++++- lib/s390x/mmu.c | 260 +++++++++++++++++++++++++++++++++++++++++++----- 2 files changed, 307 insertions(+), 26 deletions(-)lib/s390x/mmu.h +59
-2@@ -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);
lib/s390x/mmu.h line 84
Quoted 10 lines
lib/s390x/mmu.h +10
-0+ +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
Quoted 8 lines
lib/s390x/mmu.h +2
-0+void *get_dat_entry (pgd_t *pgtable,void *vaddr,unsigned int level); +#endif /* _ASMS390X_MMU_H_ */ lib/s390x/mmu.c +0
-0
lib/s390x/mmu.c line 20
Quoted 6 lines
lib/s390x/mmu.c +3
-0@@ -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
Quoted 34 lines
lib/s390x/mmu.c +18
-4+* 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; +} + +/*
lib/s390x/mmu.c static void mmu_enable(pgd_t *pgtable) · line 84
Quoted 10 lines
lib/s390x/mmu.c +10
-0+* 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.
Quoted 25 lines
lib/s390x/mmu.c +25
-0+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) +{
lib/s390x/mmu.c static void mmu_enable(pgd_t *pgtable) · line 119
Quoted 10 lines
lib/s390x/mmu.c +10
-0+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?
Quoted 22 lines
lib/s390x/mmu.c +22
-0+ +} + +/* +* 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) ®ION3_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 ; +}
Re: [kvm-unit-tests PATCH v4 6/7] s390x: mmu: add support for large pages
Quoted 109 lines · On Fri, 28 May 2021 10:44:32 +0200 Janosch Frank <[email protected]> wrote:
Quoted 85 lines · Claudio Imbrenda
Add support for 1M and 2G pages.
Signed-off-by Claudio Imbrenda <[email protected]>--- lib/s390x/mmu.h | 73 +++++++++++++- lib/s390x/mmu.c | 260+++++++++++++++++++++++++++++++++++++++++++----- 2 files changed, 307 insertions(+), 26 deletions(-)
lib/s390x/mmu.h +18
-1@@ -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, unsignedint 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)
+ */
+voidunprotect_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); -voidunprotect_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 toinstall_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_tphys, void *vaddr); +
+/* Similar toinstall_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_tphys, void *vaddr); +lib/s390x/mmu.h line 84
Quoted 9 lines
lib/s390x/mmu.h +9
-0+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
Quoted 11 lines
+void *get_dat_entry(pgd_t *pgtable, void *vaddr, unsigned int level); +
#endif /* _ASMS390X_MMU_H_ */lib/s390x/mmu.c +0
-0@@ -15,6 +15,18 @@#include <vmalloc.h> #include "mmu.h"
Quoted 7 lines
Quoted 5 lines
lib/s390x/mmu.c +3
-0+/* +* 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
Quoted 42 lines
Quoted 42 lines
+ * 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
+ */lib/s390x/mmu.c +16
-4+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) +{
Quoted 8 lines
lib/s390x/mmu.c static void mmu_enable(pgd_t *pgtable) · line 84
Quoted 4 lines
lib/s390x/mmu.c +4
-0+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.
Quoted 31 lines
Quoted 31 lines
lib/s390x/mmu.c +19
-0+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.
+ */lib/s390x/mmu.c +6
-0+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; +
Quoted 8 lines
lib/s390x/mmu.c static void mmu_enable(pgd_t *pgtable) · line 119
Quoted 6 lines
lib/s390x/mmu.c +6
-0+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
Quoted 24 lines
Quoted 24 lines
lib/s390x/mmu.c +5
-0+ +} + +/* +* 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.
+ */lib/s390x/mmu.c +10
-0+static void split_pud (pud_t *pud,uintptr_t va) +{ +phys_addr_t pa =pud_val (*pud) ®ION3_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;lib/s390x/mmu.c +2
-0+idte_pudp (va, &pud_val (*pud)); + pud_val(*pud) = __pa(pmd) | REGION_ENTRY_TT_REGION3 |
REGION_TABLE_LENGTH; +}
PATCH v4 7/7 s390x: edat test
In reply to Claudio ImbrendaSimple EDAT test.
--- s390x/Makefile | 1 + s390x/edat.c | 274 ++++++++++++++++++++++++++++++++++++++++++++ s390x/unittests.cfg | 3 + 3 files changed, 278 insertions(+) create mode 100644 s390x/edat.c
s390x/Makefile +1 -0
@@ -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),)
s390x/edat.c new file +274 -0
274 lines not on this page · raw message
s390x/unittests.cfg +3 -0
@@ -103,3 +103,6 @@ file = sie.elf [mvpg] file = mvpg.elf timeout = 10 + +[edat] +file = edat.elf
Signature · 1 line
2.31.1
Re: [kvm-unit-tests PATCH v4 7/7] s390x: edat test
Quoted 3 lines · Claudio Imbrenda
Simple EDAT test.
Signed-off-by Claudio Imbrenda <[email protected]>
Some nits below.
Acked-by Janosch Frank <[email protected]>
Quoted 95 lines
--- s390x/Makefile | 1 + s390x/edat.c | 274 ++++++++++++++++++++++++++++++++++++++++++++ s390x/unittests.cfg | 3 + 3 files changed, 278 insertions(+) create mode 100644 s390x/edat.cs390x/Makefile +1
-0@@ -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),)s390x/edat.c new file +70
-0@@ -0,0 +1,274 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* +* EDAT test. +* +* Copyright (c) 2021 IBM Corp +* +* Authors: +* Claudio Imbrenda <[email protected]> +*/ +#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" ); + +/*
s390x/edat.c line 80
Quoted 10 lines
s390x/edat.c new file +10
-0+* 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?
s390x/edat.c line 86
Quoted 6 lines
s390x/edat.c new file +6
-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 );
These page table level magic constants give me a headache. PGTABLE_LEVEL_PMD ?
Quoted 200 lines
s390x/edat.c new file +188
-0+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 (); +}s390x/unittests.cfg +3
-0@@ -103,3 +103,6 @@ file = sie.elf [mvpg] file = mvpg.elf timeout = 10 + +[edat] +file = edat.elf