| Message ID | 20120823050958.GB2332@ram-ThinkPad-T61 (mailing list archive) |
|---|---|
| State | New, archived |
| Delegated to: | Bjorn Helgaas |
| Headers | show |
Hi, Ram and Bjorn, On Wed, Aug 22, 2012 at 10:09 PM, Ram Pai <linuxram@us.ibm.com> wrote: > +static inline int pci_next_resource_idx(int i, int flag) > +{ > + while (++i < PCI_NUM_RESOURCES) { > + if ((i >= 0 && i < PCI_ROM_RESOURCE && (flag & PCI_STD_RES)) || > + (i == PCI_ROM_RESOURCE && (flag & PCI_ROM_RES)) || > +#ifdef CONFIG_PCI_IOV > + (i <= PCI_IOV_RESOURCE_END && (flag & PCI_IOV_RES)) || > +#endif > + (i <= PCI_BRIDGE_RESOURCE_END && (flag & PCI_BRIDGE_RES))) > + return i; > + } > + return -1; > +} no, you can not merge them. when it start as -1, and user only need the bridge resource, it will loop from 0 to 16. I optimized it more to skip some searching. please check v5 and v6. v5 will store aside the mask, and use the bit map later. v6 will still to do the local checking, but will skip some ++i loop. it seems v5 is more efficient while it will need 16*8 bytes for mask storage. v5: pci_next_resource_idx: 000000000000098c <pci_next_resource_idx>: 98c: 55 push %rbp 98d: 48 89 e5 mov %rsp,%rbp 990: 53 push %rbx 991: 41 50 push %r8 993: e8 00 00 00 00 callq 998 <pci_next_resource_idx+0xc> 998: 89 fb mov %edi,%ebx 99a: 89 f7 mov %esi,%edi 99c: ff c3 inc %ebx 99e: e8 d1 fe ff ff callq 874 <get_res_idx_mask> 9a3: 48 89 c7 mov %rax,%rdi 9a6: 83 c8 ff or $0xffffffffffffffff,%eax 9a9: 83 fb 10 cmp $0x10,%ebx 9ac: 7f 18 jg 9c6 <pci_next_resource_idx+0x3a> 9ae: 48 63 d3 movslq %ebx,%rdx 9b1: be 11 00 00 00 mov $0x11,%esi 9b6: e8 00 00 00 00 callq 9bb <pci_next_resource_idx+0x2f> 9bb: ba ff ff ff ff mov $0xffffffff,%edx 9c0: 83 f8 11 cmp $0x11,%eax 9c3: 0f 4d c2 cmovge %edx,%eax 9c6: 59 pop %rcx 9c7: 5b pop %rbx 9c8: 5d pop %rbp 9c9: c3 retq v6: would need to more code in pci_next_resource_idx: 00000000000008e1 <pci_next_resource_idx>: 8e1: 55 push %rbp 8e2: 48 89 e5 mov %rsp,%rbp 8e5: e8 00 00 00 00 callq 8ea <pci_next_resource_idx+0x9> 8ea: 89 f1 mov %esi,%ecx 8ec: 89 f2 mov %esi,%edx 8ee: 89 f8 mov %edi,%eax 8f0: 89 f7 mov %esi,%edi 8f2: 83 e1 04 and $0x4,%ecx 8f5: 83 e2 02 and $0x2,%edx 8f8: 83 e7 08 and $0x8,%edi 8fb: 83 e6 01 and $0x1,%esi 8fe: eb 35 jmp 935 <pci_next_resource_idx+0x54> 900: 85 c0 test %eax,%eax 902: 78 1c js 920 <pci_next_resource_idx+0x3f> 904: 83 f8 05 cmp $0x5,%eax 907: 7f 07 jg 910 <pci_next_resource_idx+0x2f> 909: 40 84 f6 test %sil,%sil 90c: 75 38 jne 946 <pci_next_resource_idx+0x65> 90e: eb 05 jmp 915 <pci_next_resource_idx+0x34> 910: 83 f8 06 cmp $0x6,%eax 913: 75 0b jne 920 <pci_next_resource_idx+0x3f> 915: 85 d2 test %edx,%edx 917: 75 28 jne 941 <pci_next_resource_idx+0x60> 919: b8 07 00 00 00 mov $0x7,%eax 91e: eb 0a jmp 92a <pci_next_resource_idx+0x49> 920: 83 f8 06 cmp $0x6,%eax 923: 7e 10 jle 935 <pci_next_resource_idx+0x54> 925: 83 f8 0c cmp $0xc,%eax 928: 7f 1e jg 948 <pci_next_resource_idx+0x67> 92a: 85 c9 test %ecx,%ecx 92c: 75 18 jne 946 <pci_next_resource_idx+0x65> 92e: b8 0d 00 00 00 mov $0xd,%eax 933: eb 13 jmp 948 <pci_next_resource_idx+0x67> 935: ff c0 inc %eax 937: 83 f8 10 cmp $0x10,%eax 93a: 7e c4 jle 900 <pci_next_resource_idx+0x1f> 93c: 83 c8 ff or $0xffffffffffffffff,%eax 93f: eb 05 jmp 946 <pci_next_resource_idx+0x65> 941: b8 06 00 00 00 mov $0x6,%eax 946: 5d pop %rbp 947: c3 retq 948: 85 ff test %edi,%edi 94a: 75 fa jne 946 <pci_next_resource_idx+0x65> 94c: b8 11 00 00 00 mov $0x11,%eax 951: eb e2 jmp 935 <pci_next_resource_idx+0x54> So just -v5 from now? Please let me know which one is good, so I could rebase for-pci-for-each-res-addon branch, and then Ram could rebase his patches to top of new for-pci-each-res-addon branch. Thanks Yinghai
diff --git a/include/linux/pci.h b/include/linux/pci.h index e444f5b..7101da2 100644 --- a/include/linux/pci.h +++ b/include/linux/pci.h @@ -1351,6 +1351,48 @@ static inline int pci_domain_nr(struct pci_bus *bus) (pci_resource_end((dev), (bar)) - \ pci_resource_start((dev), (bar)) + 1)) +#define PCI_STD_RES (1<<0) +#define PCI_ROM_RES (1<<1) +#define PCI_BRIDGE_RES (1<<2) +#define PCI_IOV_RES (1<<3) +#define PCI_ALL_RES (PCI_STD_RES|PCI_ROM_RES|PCI_BRIDGE_RES|PCI_IOV_RES) +#define PCI_NOSTD_RES (PCI_ALL_RES & ~PCI_STD_RES) +#define PCI_NOIOV_RES (PCI_ALL_RES & ~PCI_IOV_RES) +#define PCI_NOROM_RES (PCI_ALL_RES & ~PCI_ROM_RES) +#define PCI_NOBRIDGE_RES (PCI_ALL_RES & ~PCI_BRIDGE_RES) +#define PCI_STD_ROM_RES (PCI_STD_RES | PCI_ROM_RES) +#define PCI_STD_IOV_RES (PCI_STD_RES | PCI_IOV_RES) +#define PCI_STD_ROM_IOV_RES (PCI_NOBRIDGE_RES) + +static inline struct resource *pci_dev_resource_n(struct pci_dev *dev, int n) +{ + if (n >= 0 && n < PCI_NUM_RESOURCES) + return &dev->resource[n]; + + return NULL; +} + +static inline int pci_next_resource_idx(int i, int flag) +{ + while (++i < PCI_NUM_RESOURCES) { + if ((i >= 0 && i < PCI_ROM_RESOURCE && (flag & PCI_STD_RES)) || + (i == PCI_ROM_RESOURCE && (flag & PCI_ROM_RES)) || +#ifdef CONFIG_PCI_IOV + (i <= PCI_IOV_RESOURCE_END && (flag & PCI_IOV_RES)) || +#endif + (i <= PCI_BRIDGE_RESOURCE_END && (flag & PCI_BRIDGE_RES))) + return i; + } + return -1; +} + +#define for_each_pci_resource(dev, res, flag) \ + for (i = pci_next_resource_idx(-1, flag), \ + res = pci_dev_resource_n(dev, i); \ + res; \ + i = pci_next_resource_idx(i, flag), \ + res = pci_dev_resource_n(dev, i)) + /* Similar to the helpers above, these manipulate per-pci_dev * driver-specific data. They are really just a wrapper around * the generic device structure functions of these calls.
PCI: pci resource iterator Currently pci_dev structure holds an array of 17 PCI resources; six base BARs, one ROM BAR, four BRIDGE BARs, six sriov BARs. This is wasteful. A bridge device just needs the 4 bridge resources. A non-bridge device just needs the six base resources and one ROM resource. The sriov resources are needed only if the device has SRIOV capability. The pci_dev structure needs to be re-organized to avoid unnecessary bloating. However too much code outside the pci-bus driver, assumes the internal details of the pci_dev structure, thus making it hard to re-organize the datastructure. As a first step this patch provides generic methods to access the resource structure of the pci_dev. Next step, change the code in all the drivers that access the resource structure to use the new iterator. Finally we can re-organize the resource structure in the pci_dev structure and correspondingly update the methods. This patch is compile tested only. Changelog v2: Consolidated iterator interface as per Bjorn's suggestion. Changelog v3: ability to get index of the returned resource. Suggested by Yinghai. Incorporates Yinghai's code snippets. <TODO get Yinghai's SOB> Signed-off-by: Ram Pai <linuxram@us.ibm.com> -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html