aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/parisc
diff options
context:
space:
mode:
authorGrant Grundler <grundler@parisc-linux.org>2005-10-21 22:37:20 -0400
committerKyle McMartin <kyle@parisc-linux.org>2005-10-21 22:37:20 -0400
commit64908ad95c34f25849412d6d4735ac10f8fb6575 (patch)
tree544173d34dadfb80339e3d4ba147875ad1f5c731 /drivers/parisc
parent53f01bba49938f115237fe43a261c31ac13ae5c6 (diff)
[PARISC] Update sba_iommu from parisc tree
revert use of %%sr0 in fdc asm. Thanks to Joel Soete for pointing out this oversight. Signed-off-by: Grant Grundler <grundler@parisc-linux.org> 2.6.14-rc2-pa3 move "sync" outside the main loop that fills IO Pdir. Signed-off-by: Grant Grundler <grundler@parisc-linux.org> remove explicit use of sr0 in fdc ops. Thanks to Joel Soete for reminding me were I added those... Signed-off-by: Grant Grundler <grundler@parisc-linux.org> 2.6.14-rc2-pa2 - make SBA more anal about invalidating pdir entries Previous code cleared the valid flag a pdir entry but it did NOT guarantee this change was visible to the PDIR before writing the PCOM register. Ie the SBA could pick up a stale entry if the write happened to hit the SBA before the cacheline was flushed from the cache. Long term, I think I want to make this a compile time flag. Developement tree should enable anal pdir checking by default and Debian can disable it with either a CONFIG option or one-line patch. fdc/sync options can only negatively affect performance though I haven't measure how much yet. If someone can run netperf TCP_RR across gige and compare -pa1 and -pa2, that would be sufficient. Cleaned up the use of "fdc" to make sure it's using "kernel" space id (specify sr0 but maps to sr4-7). It seems a bit fragile to assume "sr1" gets loaded with KERNEL_SPACE which is how the code works today. Tested on 32 and 64-bit SMP kernels on j6k. Signed-off-by: Grant Grundler <grundler@parisc-linux.org> remove PDC_NARROW from SBA and document history of PDC_NARROW a bit. It will still show up in an older kernel's .config file. Signed-off-by: Grant Grundler <grundler@parisc-linux.org> if/ifdef cleanups from Joel Soete. Signed-off-by: Grant Grundler <grundler@parisc-linux.org> 2.6.12-rc4-pa2 fix 32-bit support for Astro platforms o Since my last SBA code change, SBA could allocate more than 1GB of IOVA space on Astro boxes with more than 1GB of RAM when running 32-bit kernel. This is bad since IOMMU can only talk to the first 1GB at most. Kudos to jejb for quickly spotting that bug. o jejb also noted SBA should *always* reject DMA masks > 32-bits since DMA-mapping.txt indicates caller should try again with 32-bits. o off-by-one error when comparing the mask to IOVA space size. Signed-off-by: Grant Grundler <grundler@parisc-linux.org> Signed-off-by: Kyle McMartin <kyle@parisc-linux.org>
Diffstat (limited to 'drivers/parisc')
-rw-r--r--drivers/parisc/sba_iommu.c140
1 files changed, 91 insertions, 49 deletions
diff --git a/drivers/parisc/sba_iommu.c b/drivers/parisc/sba_iommu.c
index 6256ad365d0b..48591badd4ce 100644
--- a/drivers/parisc/sba_iommu.c
+++ b/drivers/parisc/sba_iommu.c
@@ -91,8 +91,8 @@ extern struct proc_dir_entry * proc_mckinley_root;
91#define DBG_RES(x...) 91#define DBG_RES(x...)
92#endif 92#endif
93 93
94#if defined(__LP64__) && !defined(CONFIG_PDC_NARROW) 94#if defined(CONFIG_64BIT)
95/* "low end" PA8800 machines use ZX1 chipset */ 95/* "low end" PA8800 machines use ZX1 chipset: PAT PDC and only run 64-bit */
96#define ZX1_SUPPORT 96#define ZX1_SUPPORT
97#endif 97#endif
98 98
@@ -231,7 +231,7 @@ struct ioc {
231 spinlock_t res_lock; 231 spinlock_t res_lock;
232 unsigned int res_bitshift; /* from the LEFT! */ 232 unsigned int res_bitshift; /* from the LEFT! */
233 unsigned int res_size; /* size of resource map in bytes */ 233 unsigned int res_size; /* size of resource map in bytes */
234#if SBA_HINT_SUPPORT 234#ifdef SBA_HINT_SUPPORT
235/* FIXME : DMA HINTs not used */ 235/* FIXME : DMA HINTs not used */
236 unsigned long hint_mask_pdir; /* bits used for DMA hints */ 236 unsigned long hint_mask_pdir; /* bits used for DMA hints */
237 unsigned int hint_shift_pdir; 237 unsigned int hint_shift_pdir;
@@ -294,7 +294,7 @@ static unsigned long piranha_bad_128k = 0;
294/* Looks nice and keeps the compiler happy */ 294/* Looks nice and keeps the compiler happy */
295#define SBA_DEV(d) ((struct sba_device *) (d)) 295#define SBA_DEV(d) ((struct sba_device *) (d))
296 296
297#if SBA_AGP_SUPPORT 297#ifdef SBA_AGP_SUPPORT
298static int reserve_sba_gart = 1; 298static int reserve_sba_gart = 1;
299#endif 299#endif
300 300
@@ -314,7 +314,7 @@ static int reserve_sba_gart = 1;
314#define WRITE_REG32(val, addr) __raw_writel(cpu_to_le32(val), addr) 314#define WRITE_REG32(val, addr) __raw_writel(cpu_to_le32(val), addr)
315#define WRITE_REG64(val, addr) __raw_writeq(cpu_to_le64(val), addr) 315#define WRITE_REG64(val, addr) __raw_writeq(cpu_to_le64(val), addr)
316 316
317#ifdef __LP64__ 317#ifdef CONFIG_64BIT
318#define READ_REG(addr) READ_REG64(addr) 318#define READ_REG(addr) READ_REG64(addr)
319#define WRITE_REG(value, addr) WRITE_REG64(value, addr) 319#define WRITE_REG(value, addr) WRITE_REG64(value, addr)
320#else 320#else
@@ -324,7 +324,7 @@ static int reserve_sba_gart = 1;
324 324
325#ifdef DEBUG_SBA_INIT 325#ifdef DEBUG_SBA_INIT
326 326
327/* NOTE: When __LP64__ isn't defined, READ_REG64() is two 32-bit reads */ 327/* NOTE: When CONFIG_64BIT isn't defined, READ_REG64() is two 32-bit reads */
328 328
329/** 329/**
330 * sba_dump_ranges - debugging only - print ranges assigned to this IOA 330 * sba_dump_ranges - debugging only - print ranges assigned to this IOA
@@ -364,7 +364,7 @@ static void sba_dump_tlb(void __iomem *hpa)
364#else 364#else
365#define sba_dump_ranges(x) 365#define sba_dump_ranges(x)
366#define sba_dump_tlb(x) 366#define sba_dump_tlb(x)
367#endif 367#endif /* DEBUG_SBA_INIT */
368 368
369 369
370#ifdef ASSERT_PDIR_SANITY 370#ifdef ASSERT_PDIR_SANITY
@@ -674,7 +674,7 @@ sba_free_range(struct ioc *ioc, dma_addr_t iova, size_t size)
674* 674*
675***************************************************************/ 675***************************************************************/
676 676
677#if SBA_HINT_SUPPORT 677#ifdef SBA_HINT_SUPPORT
678#define SBA_DMA_HINT(ioc, val) ((val) << (ioc)->hint_shift_pdir) 678#define SBA_DMA_HINT(ioc, val) ((val) << (ioc)->hint_shift_pdir)
679#endif 679#endif
680 680
@@ -743,9 +743,8 @@ sba_io_pdir_entry(u64 *pdir_ptr, space_t sid, unsigned long vba,
743 * (bit #61, big endian), we have to flush and sync every time 743 * (bit #61, big endian), we have to flush and sync every time
744 * IO-PDIR is changed in Ike/Astro. 744 * IO-PDIR is changed in Ike/Astro.
745 */ 745 */
746 if (ioc_needs_fdc) { 746 if (ioc_needs_fdc)
747 asm volatile("fdc 0(%%sr1,%0)\n\tsync" : : "r" (pdir_ptr)); 747 asm volatile("fdc %%r0(%0)" : : "r" (pdir_ptr));
748 }
749} 748}
750 749
751 750
@@ -769,42 +768,57 @@ static SBA_INLINE void
769sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt) 768sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt)
770{ 769{
771 u32 iovp = (u32) SBA_IOVP(ioc,iova); 770 u32 iovp = (u32) SBA_IOVP(ioc,iova);
772 771 u64 *pdir_ptr = &ioc->pdir_base[PDIR_INDEX(iovp)];
773 /* Even though this is a big-endian machine, the entries
774 ** in the iopdir are little endian. That's why we clear the byte
775 ** at +7 instead of at +0.
776 */
777 int off = PDIR_INDEX(iovp)*sizeof(u64)+7;
778 772
779#ifdef ASSERT_PDIR_SANITY 773#ifdef ASSERT_PDIR_SANITY
780 /* Assert first pdir entry is set */ 774 /* Assert first pdir entry is set.
781 if (0x80 != (((u8 *) ioc->pdir_base)[off])) { 775 **
776 ** Even though this is a big-endian machine, the entries
777 ** in the iopdir are little endian. That's why we look at
778 ** the byte at +7 instead of at +0.
779 */
780 if (0x80 != (((u8 *) pdir_ptr)[7])) {
782 sba_dump_pdir_entry(ioc,"sba_mark_invalid()", PDIR_INDEX(iovp)); 781 sba_dump_pdir_entry(ioc,"sba_mark_invalid()", PDIR_INDEX(iovp));
783 } 782 }
784#endif 783#endif
785 784
786 if (byte_cnt <= IOVP_SIZE) 785 if (byte_cnt > IOVP_SIZE)
787 { 786 {
788 iovp |= IOVP_SHIFT; /* set "size" field for PCOM */ 787#if 0
788 unsigned long entries_per_cacheline = ioc_needs_fdc ?
789 L1_CACHE_ALIGN(((unsigned long) pdir_ptr))
790 - (unsigned long) pdir_ptr;
791 : 262144;
792#endif
789 793
790 /* 794 /* set "size" field for PCOM */
791 ** clear I/O PDIR entry "valid" bit 795 iovp |= get_order(byte_cnt) + PAGE_SHIFT;
792 ** Do NOT clear the rest - save it for debugging.
793 ** We should only clear bits that have previously
794 ** been enabled.
795 */
796 ((u8 *)(ioc->pdir_base))[off] = 0;
797 } else {
798 u32 t = get_order(byte_cnt) + PAGE_SHIFT;
799 796
800 iovp |= t;
801 do { 797 do {
802 /* clear I/O Pdir entry "valid" bit first */ 798 /* clear I/O Pdir entry "valid" bit first */
803 ((u8 *)(ioc->pdir_base))[off] = 0; 799 ((u8 *) pdir_ptr)[7] = 0;
804 off += sizeof(u64); 800 if (ioc_needs_fdc) {
801 asm volatile("fdc %%r0(%0)" : : "r" (pdir_ptr));
802#if 0
803 entries_per_cacheline = L1_CACHE_SHIFT - 3;
804#endif
805 }
806 pdir_ptr++;
805 byte_cnt -= IOVP_SIZE; 807 byte_cnt -= IOVP_SIZE;
806 } while (byte_cnt > 0); 808 } while (byte_cnt > IOVP_SIZE);
807 } 809 } else
810 iovp |= IOVP_SHIFT; /* set "size" field for PCOM */
811
812 /*
813 ** clear I/O PDIR entry "valid" bit.
814 ** We have to R/M/W the cacheline regardless how much of the
815 ** pdir entry that we clobber.
816 ** The rest of the entry would be useful for debugging if we
817 ** could dump core on HPMC.
818 */
819 ((u8 *) pdir_ptr)[7] = 0;
820 if (ioc_needs_fdc)
821 asm volatile("fdc %%r0(%0)" : : "r" (pdir_ptr));
808 822
809 WRITE_REG( SBA_IOVA(ioc, iovp, 0, 0), ioc->ioc_hpa+IOC_PCOM); 823 WRITE_REG( SBA_IOVA(ioc, iovp, 0, 0), ioc->ioc_hpa+IOC_PCOM);
810} 824}
@@ -819,18 +833,29 @@ sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt)
819static int sba_dma_supported( struct device *dev, u64 mask) 833static int sba_dma_supported( struct device *dev, u64 mask)
820{ 834{
821 struct ioc *ioc; 835 struct ioc *ioc;
836
822 if (dev == NULL) { 837 if (dev == NULL) {
823 printk(KERN_ERR MODULE_NAME ": EISA/ISA/et al not supported\n"); 838 printk(KERN_ERR MODULE_NAME ": EISA/ISA/et al not supported\n");
824 BUG(); 839 BUG();
825 return(0); 840 return(0);
826 } 841 }
827 842
828 ioc = GET_IOC(dev); 843 /* Documentation/DMA-mapping.txt tells drivers to try 64-bit first,
844 * then fall back to 32-bit if that fails.
845 * We are just "encouraging" 32-bit DMA masks here since we can
846 * never allow IOMMU bypass unless we add special support for ZX1.
847 */
848 if (mask > ~0U)
849 return 0;
829 850
830 /* check if mask is > than the largest IO Virt Address */ 851 ioc = GET_IOC(dev);
831 852
832 return((int) (mask >= (ioc->ibase + 853 /*
833 (ioc->pdir_size / sizeof(u64) * IOVP_SIZE) ))); 854 * check if mask is >= than the current max IO Virt Address
855 * The max IO Virt address will *always* < 30 bits.
856 */
857 return((int)(mask >= (ioc->ibase - 1 +
858 (ioc->pdir_size / sizeof(u64) * IOVP_SIZE) )));
834} 859}
835 860
836 861
@@ -898,11 +923,17 @@ sba_map_single(struct device *dev, void *addr, size_t size,
898 size -= IOVP_SIZE; 923 size -= IOVP_SIZE;
899 pdir_start++; 924 pdir_start++;
900 } 925 }
901 /* form complete address */ 926
927 /* force FDC ops in io_pdir_entry() to be visible to IOMMU */
928 if (ioc_needs_fdc)
929 asm volatile("sync" : : );
930
902#ifdef ASSERT_PDIR_SANITY 931#ifdef ASSERT_PDIR_SANITY
903 sba_check_pdir(ioc,"Check after sba_map_single()"); 932 sba_check_pdir(ioc,"Check after sba_map_single()");
904#endif 933#endif
905 spin_unlock_irqrestore(&ioc->res_lock, flags); 934 spin_unlock_irqrestore(&ioc->res_lock, flags);
935
936 /* form complete address */
906 return SBA_IOVA(ioc, iovp, offset, DEFAULT_DMA_HINT_REG); 937 return SBA_IOVA(ioc, iovp, offset, DEFAULT_DMA_HINT_REG);
907} 938}
908 939
@@ -958,12 +989,19 @@ sba_unmap_single(struct device *dev, dma_addr_t iova, size_t size,
958 d--; 989 d--;
959 } 990 }
960 ioc->saved_cnt = 0; 991 ioc->saved_cnt = 0;
992
961 READ_REG(ioc->ioc_hpa+IOC_PCOM); /* flush purges */ 993 READ_REG(ioc->ioc_hpa+IOC_PCOM); /* flush purges */
962 } 994 }
963#else /* DELAYED_RESOURCE_CNT == 0 */ 995#else /* DELAYED_RESOURCE_CNT == 0 */
964 sba_free_range(ioc, iova, size); 996 sba_free_range(ioc, iova, size);
997
998 /* If fdc's were issued, force fdc's to be visible now */
999 if (ioc_needs_fdc)
1000 asm volatile("sync" : : );
1001
965 READ_REG(ioc->ioc_hpa+IOC_PCOM); /* flush purges */ 1002 READ_REG(ioc->ioc_hpa+IOC_PCOM); /* flush purges */
966#endif /* DELAYED_RESOURCE_CNT == 0 */ 1003#endif /* DELAYED_RESOURCE_CNT == 0 */
1004
967 spin_unlock_irqrestore(&ioc->res_lock, flags); 1005 spin_unlock_irqrestore(&ioc->res_lock, flags);
968 1006
969 /* XXX REVISIT for 2.5 Linux - need syncdma for zero-copy support. 1007 /* XXX REVISIT for 2.5 Linux - need syncdma for zero-copy support.
@@ -1106,6 +1144,10 @@ sba_map_sg(struct device *dev, struct scatterlist *sglist, int nents,
1106 */ 1144 */
1107 filled = iommu_fill_pdir(ioc, sglist, nents, 0, sba_io_pdir_entry); 1145 filled = iommu_fill_pdir(ioc, sglist, nents, 0, sba_io_pdir_entry);
1108 1146
1147 /* force FDC ops in io_pdir_entry() to be visible to IOMMU */
1148 if (ioc_needs_fdc)
1149 asm volatile("sync" : : );
1150
1109#ifdef ASSERT_PDIR_SANITY 1151#ifdef ASSERT_PDIR_SANITY
1110 if (sba_check_pdir(ioc,"Check after sba_map_sg()")) 1152 if (sba_check_pdir(ioc,"Check after sba_map_sg()"))
1111 { 1153 {
@@ -1234,8 +1276,10 @@ sba_alloc_pdir(unsigned int pdir_size)
1234 unsigned long pdir_order = get_order(pdir_size); 1276 unsigned long pdir_order = get_order(pdir_size);
1235 1277
1236 pdir_base = __get_free_pages(GFP_KERNEL, pdir_order); 1278 pdir_base = __get_free_pages(GFP_KERNEL, pdir_order);
1237 if (NULL == (void *) pdir_base) 1279 if (NULL == (void *) pdir_base) {
1238 panic("sba_ioc_init() could not allocate I/O Page Table\n"); 1280 panic("%s() could not allocate I/O Page Table\n",
1281 __FUNCTION__);
1282 }
1239 1283
1240 /* If this is not PA8700 (PCX-W2) 1284 /* If this is not PA8700 (PCX-W2)
1241 ** OR newer than ver 2.2 1285 ** OR newer than ver 2.2
@@ -1353,7 +1397,7 @@ sba_ioc_init_pluto(struct parisc_device *sba, struct ioc *ioc, int ioc_num)
1353 u32 iova_space_mask; 1397 u32 iova_space_mask;
1354 u32 iova_space_size; 1398 u32 iova_space_size;
1355 int iov_order, tcnfg; 1399 int iov_order, tcnfg;
1356#if SBA_AGP_SUPPORT 1400#ifdef SBA_AGP_SUPPORT
1357 int agp_found = 0; 1401 int agp_found = 0;
1358#endif 1402#endif
1359 /* 1403 /*
@@ -1390,7 +1434,7 @@ sba_ioc_init_pluto(struct parisc_device *sba, struct ioc *ioc, int ioc_num)
1390 DBG_INIT("%s() pdir %p size %x\n", 1434 DBG_INIT("%s() pdir %p size %x\n",
1391 __FUNCTION__, ioc->pdir_base, ioc->pdir_size); 1435 __FUNCTION__, ioc->pdir_base, ioc->pdir_size);
1392 1436
1393#if SBA_HINT_SUPPORT 1437#ifdef SBA_HINT_SUPPORT
1394 ioc->hint_shift_pdir = iov_order + PAGE_SHIFT; 1438 ioc->hint_shift_pdir = iov_order + PAGE_SHIFT;
1395 ioc->hint_mask_pdir = ~(0x3 << (iov_order + PAGE_SHIFT)); 1439 ioc->hint_mask_pdir = ~(0x3 << (iov_order + PAGE_SHIFT));
1396 1440
@@ -1414,7 +1458,7 @@ sba_ioc_init_pluto(struct parisc_device *sba, struct ioc *ioc, int ioc_num)
1414 1458
1415 WRITE_REG(ioc->imask, ioc->ioc_hpa + IOC_IMASK); 1459 WRITE_REG(ioc->imask, ioc->ioc_hpa + IOC_IMASK);
1416 1460
1417#ifdef __LP64__ 1461#ifdef CONFIG_64BIT
1418 /* 1462 /*
1419 ** Setting the upper bits makes checking for bypass addresses 1463 ** Setting the upper bits makes checking for bypass addresses
1420 ** a little faster later on. 1464 ** a little faster later on.
@@ -1447,7 +1491,7 @@ sba_ioc_init_pluto(struct parisc_device *sba, struct ioc *ioc, int ioc_num)
1447 */ 1491 */
1448 WRITE_REG(ioc->ibase | 31, ioc->ioc_hpa + IOC_PCOM); 1492 WRITE_REG(ioc->ibase | 31, ioc->ioc_hpa + IOC_PCOM);
1449 1493
1450#if SBA_AGP_SUPPORT 1494#ifdef SBA_AGP_SUPPORT
1451 /* 1495 /*
1452 ** If an AGP device is present, only use half of the IOV space 1496 ** If an AGP device is present, only use half of the IOV space
1453 ** for PCI DMA. Unfortunately we can't know ahead of time 1497 ** for PCI DMA. Unfortunately we can't know ahead of time
@@ -1499,11 +1543,9 @@ sba_ioc_init(struct parisc_device *sba, struct ioc *ioc, int ioc_num)
1499 if (iova_space_size < (1 << (20 - PAGE_SHIFT))) { 1543 if (iova_space_size < (1 << (20 - PAGE_SHIFT))) {
1500 iova_space_size = 1 << (20 - PAGE_SHIFT); 1544 iova_space_size = 1 << (20 - PAGE_SHIFT);
1501 } 1545 }
1502#ifdef __LP64__
1503 else if (iova_space_size > (1 << (30 - PAGE_SHIFT))) { 1546 else if (iova_space_size > (1 << (30 - PAGE_SHIFT))) {
1504 iova_space_size = 1 << (30 - PAGE_SHIFT); 1547 iova_space_size = 1 << (30 - PAGE_SHIFT);
1505 } 1548 }
1506#endif
1507 1549
1508 /* 1550 /*
1509 ** iova space must be log2() in size. 1551 ** iova space must be log2() in size.
@@ -1529,7 +1571,7 @@ sba_ioc_init(struct parisc_device *sba, struct ioc *ioc, int ioc_num)
1529 DBG_INIT("%s() pdir %p size %x\n", 1571 DBG_INIT("%s() pdir %p size %x\n",
1530 __FUNCTION__, ioc->pdir_base, pdir_size); 1572 __FUNCTION__, ioc->pdir_base, pdir_size);
1531 1573
1532#if SBA_HINT_SUPPORT 1574#ifdef SBA_HINT_SUPPORT
1533 /* FIXME : DMA HINTs not used */ 1575 /* FIXME : DMA HINTs not used */
1534 ioc->hint_shift_pdir = iov_order + PAGE_SHIFT; 1576 ioc->hint_shift_pdir = iov_order + PAGE_SHIFT;
1535 ioc->hint_mask_pdir = ~(0x3 << (iov_order + PAGE_SHIFT)); 1577 ioc->hint_mask_pdir = ~(0x3 << (iov_order + PAGE_SHIFT));